intelligent communication for future automobile...

INTELLIGENT COMMUNICATION FOR FUTURE AUTOMOBILE NETWORKS

by

RADOVAN MIUCIC

DISSERTATION

Submitted to the Graduate School

of Wayne State University,

Detroit, Michigan

in partial fulfillment of the requirements

for the degree of

DOCTOR OF PHILOSOPHY

2009

MAJOR: COMPUTER ENGINEERING

Approved by:

______________________________

Advisor Date

______________________________

______________________________

______________________________

______________________________

© COPYRIGHT BY

RADOVAN MIUCIC

2009

All Rights Reserved

ii

DEDICATION

I dedicate this work to my family and friends.

iii

ACKNOWLEDGMENTS

My foremost thank goes to my dissertation adviser Dr. Syed Masud Mahmud. Dr.

Mahmud's courses and research in embedded engineering inspired my carrier

orientation and interest for PhD work. His patience, valuable feedback, insights and

suggestions contributed greatly to this dissertation. I thank him for encouragement that

carried me on through harsh times. Because of his skills, patience, and easily being

able to get hold of I will always recommend Dr. Mahmud to anyone who wishes to

pursue an advance degree in electrical or computer engineering. In addition, I would like

to thank Dr. Pepe Siy, Dr. Harpreet Singh, and Dr. Sheran Alles for being members of

my PhD dissertation committee.

My gratitude also goes to my parents for their unlimited support and inspiration.

Without their support and guidance my academic achievements would not be possible.

iv

TABLE OF CONTENTS

LIST OF FIGURES .................................................................................................. viii

LIST OF TABLES .................................................................................................... xi

CHAPTER 1. INTRODUCTION ............................................................................. 1

CHAPTER 2. BACKGROUND MATERIAL ............................................................ 5

2.1. Multicast data dissemination ............................................................................ 5

2.1.1. Embedded Software Update in Automobile Today ..................................... 6

2.1.1.2. Disadvantages of the Wire-Based Reprogramming Process................... 7

2.1.1.3. Advantages in the Wire-Based Reprogramming Process ....................... 8

2.2. Vehicle network Data Reduction ...................................................................... 9

2.2.1. Data Reduction (DR) ................................................................................ 13

2.2.2. Adaptive Data Reduction (ADR) ............................................................... 15

2.2.3. Improved Adaptive Data Reduction (IADR) .............................................. 17

2.3. Distributed Technique for Remote Programming ........................................... 18

2.4. Dedicated Short Range Communication ........................................................ 20

2.4.1. DSRC Overview ....................................................................................... 23

2.4.2. Existing Studies Addressing Channel Characteristics .............................. 26

CHAPTER 3. AN ALGORITHM FOR REMOTE SOFTWARE UPDATE IN INTELLIGENT VEHICLES ............................................................. 29

3.1. Firmware over the Air ..................................................................................... 29

3.2. Wireless Unicasting vs. Multicasting .............................................................. 30

3.3. Vehicle Wireless Technologies....................................................................... 30

3.4. Infrastructure Components ............................................................................. 31

3.4.1. Central Server .......................................................................................... 31

3.4.2. Regional Manager, Towers and Connections........................................... 32

3.5. Vehicle Hardware ........................................................................................... 32

v

3.5.1. Telematics Control Unit ............................................................................ 32

3.5.2. Memory Buffer .......................................................................................... 33

3.6. Remote Software Update ............................................................................... 33

3.6.1. Downloading ............................................................................................. 33

3.6.2. Reprogramming ........................................................................................ 34

3.6.3. Verification................................................................................................ 35

3.6.4. Security .................................................................................................... 36

3.6.5. Error Detection and Correction ................................................................. 36

3.7. Simulation Description, Results and Discussion............................................. 37

3.7.1. Inputs Description ..................................................................................... 37

3.7.2. Outputs Description .................................................................................. 38

3.8. Simulation Scenarios ...................................................................................... 38

3.9. Simulation Results .......................................................................................... 41

3.9.1. Number of towers actively broadcasting ................................................... 41

3.9.2. Time duration for programming vehicles .................................................. 41

3.9.3. Unprogrammed vehicles in coverage area ............................................... 42

3.10. Analysis of the two scenarios ......................................................................... 42

CHAPTER 4. AN ENHANCED DATA REDUCTION (EDR) ALGORITHM FOR EVENT-TRIGGERED NETWORKS ............................................... 47

4.1. Signal Types and Signal Reduction Mechanisms .......................................... 48

4.2. Summary of Signal Representations in the EDR Algorithm ............................ 52

4.3. EDR Message Encoding and Decoding Algorithms ....................................... 53

4.4. Cost OF EDR Algorithm ................................................................................. 56

4.5. Impact of EDR Algorithm on Message Latency .............................................. 58

4.5.1. Synchronization ........................................................................................ 60

4.5.2. Handling of Initial Transients .................................................................... 63

vi

4.5.3. Adaptation of EDR Algorithm to other In-Vehicle Buses ........................... 64

4.6. Advantages of EDR Algorithm compared to other existing Data Reduction Algorithms for Vehicular Applications ............................................................. 65

4.6.1. Advantages of EDR compared to DR ....................................................... 65

4.6.2. Advantages of EDR compared to ADR .................................................... 67

4.6.3. Advantages of EDR compared to IADR ................................................... 68

4.7. PERFORMANCE ANALYSIS ......................................................................... 68

4.7.1. CAN Protocol ............................................................................................ 68

4.7.2. Theoretical Analysis of the EDR Algorithm ............................................... 70

4.8. Simulation, Test Results and Discussion ....................................................... 76

4.8.1. Test Methodology ..................................................................................... 76

4.8.2. Test of Real-Life Message Logs ............................................................... 77

4.8.3. Test of Artificially Created Message Logs ................................................ 80

CHAPTER 5. DISTRIBUTED TECHNIQUE FOR REMOTE PROGRAMMING ... 86

5.1. Bootloader ...................................................................................................... 86

5.2. Description of the current reprogramming process ......................................... 86

5.3. Proposed Implementation .............................................................................. 88

5.4. Feasibility Study ............................................................................................. 92

CHAPTER 6. EXPERIMENTAL CHARACTERIZATION OF THE DEDICATED SHORT RANGE COMMUNICATION ............................................. 96

6.1. Interference Model Based on Two-Ray Ground Reflection ............................ 96

6.2. Diffraction ..................................................................................................... 107

6.3. Scattering and Surface Roughness Consideration ....................................... 107

6.4. Experimental Method ................................................................................... 108

6.4.1. Experimental Vehicles ............................................................................ 108

6.4.2. Experimental Setting .............................................................................. 111

vii

6.4.3. Experimental Procedure ......................................................................... 111

6.4.4. Data Logging and Processing ................................................................ 112

6.5. Experimental Results ................................................................................... 112

6.6. Null-Point Prediction Application to Experimental Data ................................ 113

6.6.1. Analysis Author’s Experimental Data ..................................................... 114

6.6.2. Analysis of VII-C Open Area Test ........................................................... 114

6.6.3. Analysis of VSC Open Area Test ........................................................... 116

6.6.4. Analysis of Breakpoints from Masui, Kobayashi and Akaike .................. 117

6.6.5. Analysis of Critical distance from Cheng, Henty, Stancil, Bai, and Mudalige ............................................................................................................... 118

CHAPTER 7. CONCLUSION ............................................................................. 120

ACRONYMS ................................................................................................. 122

REFERENCES ................................................................................................. 126

ABSTRACT ................................................................................................. 136

AUTOBIOGRAPHICAL STATEMENT ......................................................................... 137

viii

LIST OF FIGURES

Figure 1: The First Message and the Consequent Reduced Message of the DR Algorithm. ..................................................................................................... 15

Figure 2: The First Message and the Consequent Reduced Message of the ADR Algorithm [40]. .............................................................................................. 15

Figure 3: The First Message and the Consequent Reduced Message of the IADR Algorithm [20]. .............................................................................................. 17

Figure 4: a) FCC-allocated DSRC spectrum b) channels 172 through 184 are separated into six Service Channels (SCHs) and one control channel (CCH) c) Extraction of sub-carriers of the channel 172 (Pi denotes pilot and dj denotes data sub-carriers). ........................................................................................ 24

Figure 5: DSRC power level and use plan .................................................................... 25

Figure 6: Data content byte-by-byte of the current SAE J2735 Basic Safety Message . 26

Figure 7: Wireless Infrastructure Overview. Note That Unicasting And Multicasting Can Be Different Channels Of A Single Cellular Tower (i.e., MT And UT Can Be One Tower). ................................................................................................. 31

Figure 8: Software downloading: wireless communication from a tower to a vehicle .... 35

Figure 9: Software reprogramming: wired transfer of the software from TCU to the targeted ECU. .............................................................................................. 36

Figure 10: Greater Michigan map .................................................................................. 39

Figure 11: Representation of the tower arrangement in the Greater Michigan area. ..... 39

Figure 12: Scenario A: Number of towers actively broadcasting vs. time as the size of the retransmit buffer varies. ......................................................................... 43

Figure 13: Scenario B: Number of towers actively broadcasting vs. time as the size of the retransmit buffer varies. ......................................................................... 44

Figure 14: Scenario A: Number of programmed vehicles vs. time as the size of the retransmit buffer varies. ............................................................................... 44

Figure 15: Scenario B: Number of programmed vehicles vs. time as the size of the retransmit buffer varies. ............................................................................... 45

Figure 16: Scenario A: Number of unprogrammed vehicles in coverage area .............. 45

Figure 17: Scenario B: Number of unprogrammed vehicles in coverage area .............. 46

ix

Figure 18: Two examples of Vehicle Speed Signal and Its Delta Representation. ........ 50

Figure 19: An example of the original message and its mask with SDN and SN type groupings. .................................................................................................... 51

Figure 20: An example of the original message and its mask with only SN type groupings. .................................................................................................... 52

Figure 21: An example of the Data Field of a Reduced Message. ................................ 55

Figure 22: Flowchart of EDR message encoding algorithm. ......................................... 56

Figure 23: Flowchart of EDR message decoding algorithm. ......................................... 57

Figure 24: Example of a Message with an 8-Bit Signal. ................................................ 65

Figure 25: Example of Signal Behavior and Signal Representation in EDR and DR. .... 66

Figure 26: Data frame. .................................................................................................. 69

Figure 27: The Data Frame of a CAN Message. ........................................................... 69

Figure 28: Performance Improvement vs. Signal Type Probability for a Message Containing a 64-Bit Long Signal. Note that ),(),(1),( jiPjiPjiP DSNCES −−= . ........ 75

Figure 29: Performance Improvement vs. Signal Type Probability for a Message containing twelve 5-bit Signals and one 4-Bit Signal. Note that

),(),(1),( jiPjiPjiP DSNCES −−= . ............................................................................ 76

Figure 30: Performance Comparison of EDR Considering All Messages (Test1) and the Most Frequent Messages (Test2). ............................................................... 79

Figure 31: Comparison of EDR, ADR, IADR, DR and Uncompressed Bus Utilization for the Real-Life Message Log. ......................................................................... 80

Figure 32: Number of Bytes Saved vs. Signal Type Probability for EDR, DR, IADR, and ADR Algorithms. ........................................................................................... 83

Figure 33: BU Comparison Between Uncompressed and EDR Traffic for Varying Number of Messages. .................................................................................. 84

Figure 34: Typical Execution Flow of Embedded Software in ECU ............................... 86

Figure 35: Current typical reprogramming session. ....................................................... 88

Figure 36: Proposed Reprogramming Session part 1 ................................................... 90


x


Figure 39: Programmer Connected to the Vehicle Serial Bus ....................................... 94

Figure 40: Illustration of: a) two-ray propagation between the two vehicle and b) method of images to find the path difference between the LOS and NLOS component. .................................................................................................. 98

Figure 41: Absolute real value of reflection coefficient for vertical and horizontal polarization for εr = 25. Brewster angle is 78.50. .......................................... 99

Figure 42: Polarization by reflection ............................................................................ 100

Figure 43: Illustration of energy propagation from the antenna mounted on vehicle’s roof. ............................................................................................................ 102

Figure 44: Surface roughness protuberances ............................................................. 108

Figure 45: Hardware added to trunk to convert original vehicle into experimental vehicle. ................................................................................................................... 109

Figure 46: DSRC and GPS antennas mounted on vehicle roof. .................................. 110

Figure 47: Logical blocks of vehicle equipment. .......................................................... 110

Figure 48: Location of the open area test. Open road with 55 mph speed limit ........... 111

Figure 49: Visualization of two vehicles approaching each other in the opposite direction. .................................................................................................... 112

Figure 50: RSSI versus separation distance for two vehicles approaching each other on open road. .................................................................................................. 113

Figure 51: VIIC test setup ............................................................................................ 115

Figure 52: VII Project: Relative average RSSI vs distance between transmitter and receiver. Predicted points d0, d1, and d2 are for the ht = 5 m and hr = 1.47 m. Reproduced from [68] with critical distance marking superimposed for this dissertation. (Note that RSSI in this figure is a relative value not expressed in dBm.) ......................................................................................................... 116

Figure 53: VSC Project: Relative average RSSI vs distance between transmitter and receiver. Predicted points d0 and d1, are for the ht = 3.04 m and hr = 1.45 m. Reproduced from [74] with null points marking added. (Note that RSSI in this figure is a relative value not a dBm.) .......................................................... 117

xi

LIST OF TABLES

Table I: Selected DSRC Physical Layer Parameters..................................................... 24

Table II: Summary of characteristics of the DSRC, Satellite, and Cellular characteristics ..................................................................................................................... 30

Table III Configurable Inputs to the Simulator ............................................................... 40

Table IV Output parameters of the simulator ................................................................. 40

Table V Simulation input values .................................................................................... 41

Table VI 90% completely programmed vehicles threshold ............................................ 46

Table VII Delta Span versus Signal Size ....................................................................... 50

Table VIII Worst-Case Computation Time to Encode a Signal using EDR Algorithm .... 58

Table IX Worst-Case Computation Time to Decode a Signal using EDR Algorithm. .... 59

Table X Maximum number of bits in a Standard ID (11 bit) CAN message ................... 71

Table XI Possible ),,( jitSL Values ........................................................................... 72

Table XII Configuration of the Messages Used in the Analysis. (Messages marked with "*" are the biggest contributors to bus utilization.) ........................................ 78

Table XIII Message Configuration ................................................................................. 81

Table XIV Number of reduced messages per second for different signal type probability. ..................................................................................................................... 82

Table XV -Programmer's time saved ............................................................................. 95

Table XVI - Notation ...................................................................................................... 96

Table XVII Predicted null points, incident angles in degrees and reflection coefficients for vertical and horizontal polarization for hr = ht = 1.45 m ........................ 106

Table XVIII: DSRC and GPS Component Specifics .................................................... 110

Table XIX: Null point predictions for low (hr =1.47 m) and high (hr= 1.78 m) profile vehicles and transmitter antenna heights ht=5 m. ...................................... 115

Table XX Masui Data Breakpoints ............................................................................. 118

1

CHAPTER 1.

INTRODUCTION

With every new model vehicle, functionality continues to increase in computation

complexity. Software size in the vehicle continues to grow. With an increase in software

size, ability to modify and change the original software becomes a necessity for

automotive companies. Traditionally and currently, automotive companies update

software in the vehicle using old-fashioned, hand-held, wired programmer. Advances in

the wireless communication technologies have allowed a new approach in updating in-

vehicle software. This work presents a unique proposal to manage software download

using wireless communication. In addition, this work presents improvements in the in-

vehicle network for efficient distribution of data required by software modules of various

electronic control units (ECUs). Also, this work examines channel characteristics of the

wireless protocol, commonly known as Dedicated Short Range Communication

(DSRC), for automotive usage such as software uploads, safety and other telematics

applications.

There are numerous reasons for reprogramming vehicle ECUs. New software

may remove bugs, improve performance and durability, solve problems, introduce new

features, etc. Reprogramming ECUs wirelessly saves time and if widely accepted it has

potential to reduce cost. There are numerous examples in literature for wireless

reprogramming methods applied to sensor networks [1], [2], [3] and [4]. Most of the

previous work envisions updating sensors using a single or multiple hops unicasting

packet delivery. This is because sensor nodes are low power limited communication-

range wireless nodes. For commercial and passenger vehicles, energy conservation is

2

less demanding than for sensor networks and therefore vehicle can enjoy a bit longer

communication range. This dissertation assumes radio communication of ranges of 300

m to several kilometers. In 2001 Bromley, et al filed a general-purpose patent [5] for

system, method and computer program product for remote vehicle diagnostics,

monitoring, configuring and reprogramming. The patent vaguely describes wireless

module but does not discuss the infrastructure needed for complete reprogramming.

Unlike currently employed wire based reprogramming this dissertation describes a

concept idea of wireless reprogramming based on multicasting software packets to the

specific moving vehicles in need of software update. Envisioned system is composed of

central server that controls regional managers connected to wireless towers. Software

flows from central server to regional managers via wire, optical or high-speed wireless

network. Finally, the software is delivered to the vehicles via wireless protocol.

Existing in-vehicle networks such as Controller Area Network (CAN) are reaching

bandwidth limits because of ever increasing needs for communication. This limitation

may become a bottleneck for reprogramming. One solution is to switch to a higher

speed in-vehicle network. Another may be to add additional CAN networks in vehicle.

Both mentioned solutions result in increase in cost and complexity. Data reduction

technique offers great benefit for better utilizing the existing bandwidth for the portion of

the cost of the two solutions mentioned earlier. There are several data reduction

techniques in literature. Misbahuddin’s et al [36] suggested general data reduction

technique based on repeatability of data bytes in a message. Ramteke et al [40] in his

paper proposes data reduction technique based on signal stability. Both algorithms are

designed to work with the CAN [41] but can be applied to any serial protocol. This

3

dissertation proposes improvements of the two data reduction protocols such as

introducing innovative way to detect data-reduced message by examining its length and

eliminating initial transient bus-load peaks by prioritizing timing of the initial messages.

Bootloader is an important part of every embedded system. An ECU generally

consists of an operating software and a bootloader. A bootloader is the resident

software in the ECU that facilitates reprogramming of the operating software.

Bootloader is generally programmed in during the manufacture of an ECU and never

erased, not even when the operating software is being reprogrammed. A modern

bootloader in an ECU consumes the available resources (memory, processor,

networks) of that particular ECU. Examples of bootloaders are described in [6] and [7].

Unlike typical a bootloader that uses one ECU’s memory, this dissertation proposes

novel approach of using memory resources of all available ECUs on in-vehicle network.

The ECU that is being reprogrammed (targeted ECU) queries other ECUs on the in-

vehicle network for availability of their resources (RAM). The targeted ECU then

instructs available ECUs to collect arriving software packets (in RAM) while the targeted

ECU performs lengthy memory (flash) erase and necessary preparations. By

distributing workload and employing many ECUs among participants, this solution saves

time for download.

Dedicated Short Range Communication (DSRC) has gain momentum as the

wireless protocol of choice for vehicle-to-vehicle and vehicle-to-infrastructure

communications. As a potential link for software reprogramming, it is important to

understand the wireless protocol behavior in the mobile vehicular environments.

Meyers, in his Masters' dissertation [8], presents channel characterization of DSRC in

4

suburban driving environment. Ching et al in [9] discusses tunnel propagation channel

characterization of Japanese version of the DSRC. This dissertation extends the

characterization of the DSRC by focusing on the ground reflection effect on the protocol.

Experimental and theoretical results are discussed.

This dissertation is organized as follows. Chapter 1 presents an introduction.

Chapter 2 presents background material. This material covers the work that has been

done for wireless software update. It also includes an overview of the data reduction

protocols for in vehicle networks, followed by an overview of the DSRC protocol.

Chapter 3 presents, in detail, an algorithm for remote software update. Chapter 4

presents an enhanced data reduction algorithm along with experimental and simulation

results showing advantage of the data reduction algorithm. Distributed boot-loading

technique is described in Chapter 5. Experimental characterization of the DSRC and

theoretical analysis are given in Chapter 6. Finally, conclusion is presented in Chapter

7.

5

CHAPTER 2.

BACKGROUND MATERIAL

In recent years, the multicast communication has become necessary in a variety

of mobile services, such as video and audio sharing, software update distribution and

news. A part of this dissertation work presents an overview of wireless multicast routing

concepts and describes a solution for providing multicast content distribution applicable

for software updates with automotive industry.

2.1. MULTICAST DATA DISSEMINATION

Recent studies show great interest in hybrid wireless systems combining different

radio access networks (RANs) to provide efficient data services to mobile users [10]

[11][12]. Keller et al [10] discussed synergy of different RANs. The overall

communication infrastructure consists of many RANs. The type of needed service will

influence the decision on which RAN to use for that particular wireless transmission.

Multicasting intuitively best fits a digital radio broadcasting carrier, digital audio

broadcasting (DAB), digital video broadcasting (DVB) or a multicast capable cellular

channel. Unicasting intuitively belongs to cellular networks such as GSM, IS-95, or IS-

136 (D-AMPS) [13].

Munaka et al [11] presented the Advanced-Joint System for data multicasting in

intelligent transportation system (ITS). This multicast methodology is based on

disseminating data that are location dependent. If a vehicle is in the tower's coverage

area, then that vehicle is a part of the tower’s multicast group. When the vehicle moves

to a different tower, the vehicle will join the next multicast group. The base station that

keeps track of multicast groups will predict, based on the history of the vehicle

6

movement, which multicast group the moving vehicle is likely to join. The current

multicast group will send a preparation message via the wired network to the next

multicast group in the vehicle's path. This additional information is to reduce packet loss

due to coverage handoff and the lack of group membership.

Synergy of unicasting and broadcasting infrastructures into one system is

intuitively justified by two reasons. First, unicasting is an inefficient and very costly

method to deliver common content to many users by point-to-point connections, as in a

traditional cellular system. Second, it is also inefficient to engage multicasting over a

large area for the communication to only a few users. Multicasting is justified when the

number of recipients is very large. Following the same logic, according to Bria [14],

cellular unicasting is a better alternative when number of users is small.

The motivation for proposing remote software upload consists of the desire to

save customers’ and technician’ time, increase customer satisfaction, lower the cost of

software recalls, improve management of the software delivery method, and bring

already existing wireless data delivery technologies into the automotive industry.

2.1.1. EMBEDDED SOFTWARE UPDATE IN AUTOMOBILE TODAY

Software update is usually performed as recall (mandatory or voluntary) or during

scheduled maintenance. The following list illustrates a typical wire-based software recall

in the automotive industry.

2.1.1.1. WIRE-BASED PROCESS STEPS

1. Initially, a vehicle manufacturer finds a problem with vehicle functionality. The

erroneous functionality can be fixed by changing software in one of the vehicle’s

electronic control units (ECU).

7

2. The vehicle manufacturer requests appropriate ECU supplier to provide a new

software release. The manufacturer then tests the software release for quality

assurance and compatibility.

3. The manufacturer notifies the dealers and the owners of the recall via mail. The

manufacturer sends the new version of the software to the dealers on a CD.

4. The owner drops off the vehicle at the dealership.

5. A technician, using serial communications, (a) updates a reprogramming tool with

the content from the CD; (b) connects the tool to the intra-vehicle bus to access

targeted ECU; (c) transfers the software to the targeted ECU; and (d) checks for

downloaded software version and proper functionality.

6. Finally, the customer picks up the updated vehicle.

7. The dealer charges the manufacturer for the recall labor.

2.1.1.2. DISADVANTAGES OF THE WIRE-BASED REPROGRAMMING

PROCESS

The disadvantages come from physical distribution and manual update process.

A manufacturer distributes ECU software updates to all dealers on CDs. This takes time

and resources. It causes delay in getting the latest software to the vehicles. In addition,

all dealers need to maintain resource consuming software version library. It may take

long time from the time the customer is notified to the time the vehicle is actually

updated. Many customers do not respond to the recall notices. Conducting successful

recall depends on the customer cooperation. The download process as well as the

manual setup takes time resulting in increase of overall vehicle costs inconvenience and

customer dissatisfaction. Manual process cannot be scaled or performed in parallel

8

because it involves physical connections. With a recall notice the customer becomes

aware of the problem and overall customer satisfaction decreases.

2.1.1.3. ADVANTAGES IN THE WIRE-BASED REPROGRAMMING

PROCESS

The existing reprogramming method has few advantages over the suggested

wireless method. It will take few years until wireless methodology becomes widely an

adopted solution for the automotive industry. It will take time for a vehicle owner to get

used to getting improvements without a physical service at a dealership. Currently, a

technician reprograms an ECU in the controlled environment. The vehicle is not moving

and it is under constant supervision of the technician. Any problem that may occur has

more chances to be detected immediately. The current methodology is proven and it

has worked in the past. Vehicle wired serial communication protocols and algorithms for

reprogramming are proprietary and closed source by nature. As such, protocols provide

added layer of security against unauthorized software changes.

In this dissertation work author presents vehicular software distribution network

as wireless network where vehicles are connected to the software distributors through

base stations.

As functionality of vehicles increases in complexity, the demands on the in-

vehicle networks increase as well. Maximum bus utilization often becomes the

communication bottleneck. One way to satisfy the high bandwidth requirement for future

vehicles is to use a higher bandwidth bus or multiple buses. However, the use of a

higher bandwidth bus increases the cost of the network. Similarly, the use of multiple

buses increases the cost as well as the complexity of wiring and network handling. Both

9

options are becoming solutions to the high bandwidth demand. An alternative option is

the development of a higher layer protocol that uses data reduction techniques to

reduce the amount of data to be transferred. Its goal is to communicate the same

amount of information using less bus bandwidth. It would be acceptable provided that it

does not increase the message latencies significantly so that the safety of the vehicle is

not compromised. The cost of the protocol is expected to be marginal because it

consists of one-time changes to software. Various data reduction algorithms are

available in the literature, but data reduction technology has not been widely introduced

in in-vehicle protocol standards. This paper presents a unique data reduction

methodology, along with its comparison with other proposed methodologies. The

performance of this new data reduction algorithm is found to be better than that of the

existing data reduction algorithms for a wide range of signal dynamics. The cost as well

as the impact of this protocol on the end-to-end message latency has been found to be

very marginal.

2.2. VEHICLE NETWORK DATA REDUCTION

Over the years, as the number of vehicular electronic components increased

significantly, vehicle multiplexing evolved due to the need for better wiring, diagnosis,

reliability and lower cost. The need for vehicle multiplexing was predicted as early as

1976 [10]. Lupini presented need and advantages of vehicle multiplexing [24]. Lupini

also presented issues related to designing vehicle multiplexing systems and future

trends in the area [25]. Appropriate development tools are also necessary to design and

maintain a network system. Wolfhard [26] presented a detailed description of network

development tools, their characteristics and handling procedures. Computer simulation

10

technique is also a vital tool to study the behavior of a proposed network system and

study the performance of some alternative network design [27]. As the number of

electronic components grows in a system, the probability of system malfunction due to a

faulty component increases as well. Masrur [28] proposed a fault-tolerant multiplexing

network architecture and compared the reliability and cost of the fault-tolerant system

versus those of a non-fault-tolerant system.

During the last three decades, various networking protocols were proposed for

vehicle multiplexing. However, at present CAN (Controller Area Network) is the most

popular protocol and it is widely used. Wolfhard [29] presented valuable information

about CAN protocol, its application layer design, CAN chip implementation and CAN

testing technique. LIN protocol is also used for low-cost and low-speed networks.

Various suppliers are manufacturing Electronic Control Units (ECUs) with built-in CAN

and LIN controllers. Sometimes interpretation of protocol specifications by various

suppliers may be slightly different. Therefore, interoperability of distributed ECUs is

risky. Conformance tests are necessary to reduce the risk of lacking interoperability

among cooperative ECUs. Wolfhard [30] described the process for related conformance

tests and presented the implementation architecture.

As different features such as telematics, multimedia, X-by-wire, etc. are being

added to vehicles, future vehicles will need various types of networks with various types

of protocols. Lupini [31] predicts that at least eight in-vehicle networks may be

necessary mainly on high-end vehicles in the next ten years. Interconnecting those

various types of networks will also be a challenge.

Kassakian and Perreault noted in 2001 [32] that there are up to 70 ECUs

11

scattered throughout the vehicle. Numerous ECUs in today’s vehicles are exchanging

an ever-increasing amount of information. These demands are reaching the bandwidth

limitations of the existing in-vehicle networks. Potential solutions are increasing the

bandwidth of existing buses or using multiple buses. Although these solutions are

effective, they necessarily increase the network cost and complexity. An alternative

solution that comes with a negligible cost increase is the use of a data reduction

technique. Data reduction uses algorithms to represent information more efficiently, thus

using less bus bandwidth to exchange the given amount of information. The cost

increase is negligible because it is limited to one-time software development.

To find an effective data reduction method for in-vehicle networks, it is necessary

to understand the nature of the information flow. Numerous modules connected to the

bus handle distributed functionalities of the vehicle. Modules send measurements of

their internally connected sensors, status of their actuators, states of operation, and so

forth. Generally, messages are periodic or event based. Periodic messages are

transmitted at a fixed time intervals. Event based messages are transmitted on event

occurrences. Frequent periodic and event based messages contribute the most to the

bus utilization. Therefore, author will focus on reducing the data of the most frequent

messages.

The amount of data can be reduced in several ways. In the case of a vehicle

traveling on the highway, the vehicle speed varies very little. Thus, the electronic

module responsible for sending the vehicle speed can send only one bit instead of the

actual speed to inform the recipients when the speed does not change. Similarly, when

the change in speed is very small, the electronic module can send only the amount of

12

change using a few bits, rather than the actual speed using many bits. The protocol can

also be made adaptive by allowing it to select different message formats for different

conditions of the parameter. For example, when the vehicle speed is changing slowly

with respect to time, the message format can be different from the format when the

vehicle speed is changing rapidly.

Data reduction is not only useful in production vehicles, but in development

environment as well. For example, data reduction can be applied in data acquisition

tools that use CAN Calibration Protocol (CCP) [33] to reduce the number of messages

and bus utilization. The advantages for resulting tools are increased data rate and ability

to monitor more parameters. Examples of CCP instrumentation tools are Vector’s

CANape [34] and ATI’s Vision [35] software.

This work explains in detail a new algorithm that utilizes these ideas and

compares its performance with earlier attempts at in-vehicle data reduction. The

performance of this algorithm has been found to be significantly better than that of the

previous algorithms.

The transmission and reception of messages in all data reduction techniques

work as follows: Consider a node sending a periodic message with the ID M every T

milliseconds. The sending node stores all signals at the designated transmitting buffer

TX[M] for message M. The first message M going on the bus at t = tn contains all signals

in their entirety. At the next time interval, t = tn + T, the sending node compares the

current values of the signals to the ones stored in the TX[M], transmitting buffer. The

sending node then assembles the second message M for transmitting based on the

signals’ differences. Similarly, at the reception of the first message M, at t=tn+tmtt (tmtt is

13

message travel time), the receiving node stores all signals to the designated receiving

buffer RX[M] for message M. At the reception of the second message M, at t = tn+ T +

tmtt, the receiving node first decodes the message and then using the receiving buffer

RX[M] and the information from the message, reconstructs the signals. New signals

replace their previous versions in the receiving buffer for message M.

Misbahuddin et al [36] provides a comprehensive overview of previous work

applicable to in-vehicle networks. The examined data reduction techniques are several

variants of Huffman coding [37], several variants of arithmetic coding [38], textual

substitution coding [39], and command data stream reference coding. However, the

applicability of all these methods is limited to textual data. As a remedy, Misbahuddin et

al [36] proposed a general data reduction technique based on repeatability of data bytes

in a message. Ramteke et al [40] suggested a data reduction technique based on signal

stability. Both algorithms are designed to work with the Controller Area Network (CAN)

[41] but can be applied to any serial protocol. Brief descriptions of various existing data

reduction techniques are given below.

2.2.1. DATA REDUCTION (DR)

A CAN message has up to 8 bytes [41]. Data bytes often do not change from one

to the next message transmission. Misbahuddin [36] exploited byte repeatability to

come up with a general form of a data reduction algorithm. In Misbahudin's data

reduction algorithm (DR), if all bytes change from the previously transmitted message to

the current message, the message to be transmitted is sent as is, without data

reduction. If more than one byte remains unchanged, a reduced message is sent.

DR uses the reserve bit (R) in the control field of a CAN message to indicate that

14

data reduction is taking place. In DR, this bit is called the Data Compression Bit (DCB).

The DCB is set to “1” when a message is compressed, and it is cleared to “0” when the

message is not compressed. In a compressed CAN message, the first byte is the Data

Compression Code (DCC). The position of each bit in the compression code

corresponds to the data byte position of the originally intended uncompressed message.

A bit with a value of “1” indicates a byte that has not changed since the previous

transmission, and a bit with a value of “0” indicates that the byte has changed. The

changed bytes are placed after the compression code in the data field of the actual data

frame sent over the multiplexing bus [36].

As an example, Figure 1 shows the data field of the first, original, message with 8

data bytes and the second, Misbahuddin’s reduced message, with only 4 bytes. In the

second message, bytes 0, 2, 4, 5, and 7 have not changed from the first message. Thus

the reduced second message contains only the changed bytes 1,3, and 6. Notice that

the compression code has three zeros corresponding to data bytes 1, 3, and 6

indicating that these three bytes changed in values from the first message to the

second.

15

Figure 1: The First Message and the Consequent Reduced Message of the DR Algorithm.

Figure 2: The First Message and the Consequent Reduced Message of the ADR Algorithm [40].

2.2.2. ADAPTIVE DATA REDUCTION (ADR)

Instead of concentrating on byte value preservation as Misbahuddin did,

16

Ramteke looked at the value of signals in the message to come up with the Adaptive

Data Reduction (ADR) algorithm (Figure 2) [40]. In the ADR algorithm, if the entire

message remains unchanged from one intended transmission to the next, then no

message is transmitted. In this case, there is a provision to still send such a message if

it is deemed critical to do so for synchronization, and if it has been suppressed longer

than a predetermined amount of time. Further, if some of the signals in the message do

not change in value from the previous to the current transmission, those signal values

are not sent. If any signals in the message change beyond the scope of its assigned

difference field (delta), the next message is sent in entirety. Otherwise, if signals change

such that they can be represented with their deltas, then only the differences are sent

over the multiplexing bus. A Data Compression Code (DCC) byte is used to designate

the compression type (delta representation or no change) for each delta signal within a

message. Instead of using a reserved bit to differentiate between reduced and

unchanged messages, ADR uses two different message IDs for the uncompressed and

compressed messages, where the ID of the compressed message is created by

subtracting one from the original message ID.

17

Figure 3: The First Message and the Consequent Reduced Message of the IADR Algorithm [42].

2.2.3. IMPROVED ADAPTIVE DATA REDUCTION (IADR)

Miucic et al [42] addressed shortcomings and improvements over the Adaptive

Data Reduction (ADR) suggested by Ramteke [40]. The Improved Adaptive Data

Reduction algorithm, IADR, uses different management of the reduced and original

message frames by using a convention as depicted in Figure 3. In this convention, the

first bit of the data field serves as the Data Compression Bit (DCB) and all signals in the

unchanged message are shifted to the right by one bit. In the ADR algorithm [40], if the

value of any delta of the signals in the consequent message exceeds the length of the

assigned delta field, the absolute values of all the signals (the original CAN message)

18

are transmitted rather than the delta-compressed version of the message. In the IADR

algorithm, the consequent message allows mixtures of different signal representations.

For example, in a message any signal may be represented as a no-change (signal did

not change from the previous to the current transmission), a delta-change (the signal

difference from the previous to current transmission which does not exceed the length

of the assigned delta field), or a signal-in-entirety (the entire signal, because the

difference from the previous to the current transmission exceeds the length of the

assigned delta field). This is accomplished using three levels of codes as shown in

Figure 3. The first level is the aforementioned DCB, the first bit of the data field,

indicating with 1 that there is at least one compressed signal in the message. The

second level is the Data Compression Code (DCC), which immediately follows the DCB

and consists of one bit for each signal, where 1 indicates that some form of

compression occurs for the corresponding signal, and 0 indicates that the original signal

is included instead. The third level is the bit preceding each compressed signal. It

describes how the signal is compressed: 0 represents the delta representation (only the

difference is included) while 1 represents no change (no signal information included).

In addition, to preemptively prevent potential issues of not having synchronized

signals in transmitting and receiving nodes, IADR uses “Cyclic Refresh”. In this scheme,

the first message carries all signals in their entirety. The second message forcibly sends

the first signal in entirety regardless of the amount of change. The third message

forcibly sends the second signal in entirety, and so on. The signal that is sent in entirety

rotates.

2.3. DISTRIBUTED TECHNIQUE FOR REMOTE PROGRAMMING

19

Programming an ECU's embedded software requires erasing the ECU's flash

memory and then reprogramming the ECU. Erasing the flash memory requires some

time, during which the wireless link will remain idle. However, if the wireless link is

released while the ECU is erasing its flash memory, then it will take some time to

reestablish the link between the vehicle and the remote server. Thus, some bandwidth

will be wasted one way or the other. CHAPTER 5 presents an in-vehicle distributed

technique to reduce the latency of the remote software update process as well as save

the bandwidth of the wireless links.

Every ECU in a vehicle has some RAM buffers to accept blocks of code from an

external device before the code is actually written into the ECU's flash memory. If the

code size is larger than the total buffer size, then the code is sent to the ECU in several

steps. At every step, a part of the entire code is sent to the ECU. Each part of the code

is first saved in the RAM buffer, and then it is written to the ECU's flash memory. This

process is continued until the entire code is written into the ECU's flash memory. During

this process of software update, the link between the external device and the vehicle

remains idle for a significant amount of time, which is not acceptable if the external

device is a remote unit connected to the vehicle by a wireless link. In this work, author

proposes to use the RAM buffers of as many ECUs as needed to keep the entire code

that needs to be updated in a particular ECU. This means that first, the code will be

distributed among RAM buffers of several ECUs and then the code will be written into

the flash memory of a particular ECU. The advantage of this technique is that if the total

size of all the buffers of all the ECUs together is larger than the size of the code that

needs to be updated, then the link between the external device and the vehicle will not

20

be idle. As a result, the performance of the communication system can be improved.

This work presents a detailed description of the in-vehicle distributed algorithm and

compares its performance with that of a non-distributed software update algorithm.

Resource allocation is common among distributed systems. A similar principle

can be used in vehicle systems. A vehicle bus and ECUs connected to it are distributed

within the system. Nowadays, the speed of the vehicle network and the complexity of its

resources are assimilating a distributed system. The workload can be distributed among

available ECU resources, so that results can then be used by one or more ECUs in the

network. First, author describes a typical reprogramming process. Then, author

discusses the proposed solution. At the end, the solution is justified with a

microcontroller and vehicle network survey. Chapter 5 presents a reprogramming

technique using all available RAM resources on the vehicle network.

2.4. DEDICATED SHORT RANGE COMMUNICATION

Dedicated Short Range Communication (DSRC) is gaining momentum as the

protocol of choice for wireless vehicle safety applications by automotive original

equipment manufacturers (OEMs) and road operators. DSRC is a low latency and

reliable vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication

protocol. A wide range of its applications have been conceptualized to support safety,

mobility and convenience, including: cooperative collision avoidance, travel information,

and electronic payment. Nulls in the communication signal strength have been reported

in several recent government sponsored, private, and academia research studies but

have not been analyzed sufficiently to predict their occurrence. In this dissertation work

an expression for predicting the null locations using an interference model based on

21

two-ray ground reflection is derived. This work demonstrate the improvement in

accuracy over existing analytical accounts by applying the derived expression to the

experimental data author collected as well as to data found in literature.

In the last ten years, many automotive OEMs, their tier 1 suppliers, research

universities, and wireless radio manufacturers have been investigating the potential use

of Dedicated Short Range Communication (DSRC) for vehicle-to-vehicle and vehicle-to-

infrastructure communication [47][48][49][50][51][52] to enable primarily safety, but also

other applications. There have already been successful demonstrations of DSRC

technology. A major United States Department of Transportation (USDOT) initiative

called the Vehicle Infrastructure Integration (VII) [53] has included a number of technical

projects: VII Proof-of-Concept testing [68] established the feasibility of the VII system

using 5.9 GHz DSRC technology for improving transportation safety and efficiency,

while Cooperative Intersection Collision Avoidance Systems (CICAS) [54] and Vehicle

Safety Communications - Applications (VSC-A) [55] projects both showcased

automotive collision avoidance applications which are enabled or enhanced by the

deployment of DSRC.

In order to effectively engineer and deploy DSRC applications such as those, it is

necessary to understand the characteristics of the DRSC radio communication channel.

Detailed and dependable knowledge about the behavior of the communication channel

in its expected diverse operating environments is sought to design efficient and reliable

vehicle-to-vehicle and vehicle-to-infrastructure communication systems. The wealth of

recent studies with this aim [48][49][65][66][67][68][75] suggests that this is an important

and still open question explored both by industry and academia. The studies variously

22

used combinations of analytical modeling and experimental measurement to

characterize and predict the performance of DSRC communication in typical use

conditions. The behavior of the channel is highly dependent on these diverse

conditions. This partly influences such spread of research and complicates the modeling

efforts. The conditions are diverse not only due to the variety of geographies where

these systems are expected to operate but also due the variety of dynamic conditions

characteristic to vehicle motion. This influences expanding the existing wireless

communication modeling to include vehicle mobility considerations into existing

modeling paradigms. The effectiveness of analytic modeling can still suffer from

inapplicability of model assumptions to particular vehicular scenarios. This is where

experimental data is beneficial – to identify limitations, and to inspire and validate new

modeling developments. Several of the examined studies included experimental

components where drops were encountered in the received signal strength versus the

vehicle separation distance that were not accurately modeled [65][66][67][68][75].

These nulls are common and severe. A model’s inaccuracy in predicting the nulls

leaves the model inapplicable at the null locations, which typically fall in a region

important for vehicular communication, and thus adds uncertainty to the model. This

reduces model effectiveness in communication system design and analysis. This

dissertation proposes an analytic method for more accurately identifying the null points

versus the separation distance and thus improving the usefulness of the related models.

The method is verified against experimental data presented in this dissertation as well

as data from other publications that encountered the same phenomenon.

23

2.4.1. DSRC OVERVIEW

The goal of DSRC is to provide a low-latency, high-availability, secure, wireless

communication system suitable for communication between moving vehicles in safety

and commercial applications. DSRC messages and messaging schemes are defined in

the draft SAE J2735 standard [56]. They are intended to supply a standard mechanism

for exchanging safety-relevant information between vehicles while allowing proprietary

extensions for other applications. DSRC operates over the Wireless Access in Vehicular

Environments (WAVE) communication system. Its physical layer is defined by the draft

IEEE standard 802.11p [57]. This standard is an amended version of the IEEE 802.11

standard (the common Wi-Fi wireless standard) [58]. WAVE is defined in other layers by

IEEE 1609.1 (resource manager), [59] 1609.2 (security and privacy) [60], 1609.3

(networking services for applications) [61], and 1609.4 (multi-channel access) [62].

WAVE is intended to provide interoperability among vehicles and road infrastructure,

regardless of suppliers or regions of use. DSRC can be used to refer to the messages

and messaging schemes defined to be used over the WAVE communication system, or

it can be used to refer to the overall communication system, the latter being the

approach taken here.

The Federal Communications Commission (FCC) allocated a frequency band for

DSRC from 5.85 to 5.925 GHz, as shown in Figure 4a. DSRC divides this range into 7

channels (Figure 4b) and uses orthogonal frequency-division multiplexing (OFDM) with

52 sub-carriers (-26 to 26 in Figure 4c), including 4 pilot carriers (Ps in Figure 4c), for

each channel. Table I shows the most relevant physical layer parameters. Figure 5

shows the assignment of power levels and usage.

24

Figure 4: a) FCC-allocated DSRC spectrum b) channels 172 through 184 are separated into six Service Channels (SCHs) and one control channel (CCH) c) Extraction of sub-carriers of the

channel 172 (Pi denotes pilot and dj denotes data sub-carriers).

Table I: Selected DSRC Physical Layer Parameters

Description Symbol

Value Unit

Channel used CCH 172 Modulation QPSK Data Rate 6 Mb/s Number of pilot sub-carriers

NSP 4

Number of data sub-carriers

NSD 48

Channel frequency spacing

10 MHz

Sub-carrier frequency separation

Δ 156.25 (= 10 MHz/64)

kHz

Lower channel frequency bound

5.855 GHz

Upper channel frequency bound

5.865 GHz

Wavelength of the mid frequency 5.860 GHz

λMF 0.0511591225

m

25

Figure 5: DSRC power level and use plan

The draft SAE J2735 standard [56] defines DSRC messaging schemes. The

most fundamental message is the basic safety message (BSM). It is periodically sent by

all vehicles and contains parameters defining a vehicle’s dynamic state which are

critical for safety applications, such as speed, heading, and location. Figure 6 shows a

BSM based on the current SAE J2735 draft standard [56]. This message is an example

of information exchanged using DSRC. It also served as the main payload in the

experiments to be described, as well as provided the GPS location data for the remote

vehicle, as encoded in latitude and longitude message parameters, which was used in

the transmitter-receiver (T-R) separation distance calculations.

26

Figure 6: Data content byte-by-byte of the current SAE J2735 Basic Safety Message

2.4.2. EXISTING STUDIES ADDRESSING CHANNEL CHARACTERISTICS

Author analyzed a number of past studies, which experimentally observed null

points and which theoretically attempted to account for them. In this section author

briefly introduces these studies. Later, in Section 6.6, author applies the method derived

in this dissertation to experimental data from those studies and show that the proposed

method offers a large improvement in the prediction of null point distance over those

reported by the studies.

The earliest study author encountered was by Masui et al [66]. That paper looked

at the effect of the antenna height and the amount of traffic on distances at which

breakpoints in channel modeling curves appear. The experiments were performed in an

urban LOS environment. The two different traffic volumes, high and low, were

27

represented by testing during day and night respectively. From their data presentation it

cannot be discerned that they actually had null points in signal reception strength.

However, the breakpoints in modeling curves typically occur at same distances as null

points so that the proposed null point predictions and their breakpoint predictions can be

compared, as the later analysis shows. Masui et al used the first Fresnel zone for

predicting the breakpoints and encountered significant deviations from experimental

data. They ascribed the differences between their experimental data and their

theoretical predictions to the different effective heights of antennas due to different

amounts of traffic. The same authors reported similar data and findings in [65], this time

with additional analysis.

Author next analyzed the study by Cheng et al [67] that presented modeling of

experimental data obtained by narrow-band continuous wave experiments with two test

vehicles in a suburban driving environment. They also modeled their data using two log-

linear line segments where the breakpoint was expected according to the first Fresnel

zone. Similarly, they found large discrepancies between the predicted and

experimentally obtained breakpoints and attributed the differences to reflections from

“vehicles, pedestrians, and other objects”.

The experimental work by Vehicle Infrastructure Integration Consortium (VIIC) for

verifying performance suitability of the DSRC communication channel for vehicle-to-

vehicle and vehicle-to-infrastructure applications also encountered null-points in DSRC

communication [68]. The null points were not theoretically analyzed but were suspected

to be due to destructive interference from multiple paths from the transmitter to receiver.

Their testing covered a variety of operating environments, including the open area

28

environment considered in the analysis and experiments. The predictions correspond

closely with their experimental observations. The Vehicle Safety Communications (VSC)

project by the USDOT and VSC consortium of OEMs also performed testing of DSRC in

an open area environment with similar results and explanations [75], as well as close

correspondence to the proposed predictions.

29

CHAPTER 3.

AN ALGORITHM FOR REMOTE SOFTWARE UPDATE IN INTELLIGENT VEHICLES

Current high end vehicles have many sophisticated systems such as drive-by-

wire, road condition aware navigation, satellite radio information feed, traffic congestion

control, automatic parallel parking, road guidance, adaptive cruise control and advance

crash and pedestrian warning. The Future vehicles will have even more complicated

communication intensive systems. The vehicles will communicate with intelligent

roadside equipment to get current information about accidents, traffic congestion, speed

limit, construction, weather, and nearby commerce. Increased complexity of vehicle

functions requires increasing computational power and huge software loads. From time

to time, an onboard computer needs reprogramming. Currently, dealers manually

reprogram software in onboard computers via a wire-based electrical connection one

car at a time. Having many vehicles updated wirelessly is a preferred alternative to the

manual method because the vehicles need not be physically present at a repair facility

to perform the update and multiple vehicles can be updated at the same time. Wireless

software reprogramming will save customers’ and technicians’ time and money. In this

dissertation, author first presents current directions of the vehicle wireless technologies

and then presents a wireless download algorithm for reprogramming of vehicles’

electronic control units. The algorithm proposes a hybrid infrastructure that combines

wireless unicasting and multicasting data delivery.

3.1. FIRMWARE OVER THE AIR

Firmware Over-the-Air (FOTA) is a term used for wireless upgrades of mobile

phones and personal digital assistants (PDAs). Cellular service provider would typically

30

“push-out’ firmware to the phone sets. Push refers to the delivery of firmware without

the recipient having to request it. Typically software is delivered not in entirety but as

delta code. Delta code is smaller than the original software image and it only contains

the changes. Some aspects such as delta code and version management of this cellular

download technology can be applied in the automotive industry [15].

3.2. WIRELESS UNICASTING VS. MULTICASTING

Widely deployed cellular systems such as GSM and its derivatives GERAN and

GPRS are capable of multicasting data to the subscribed mobile devices. Desiniotis

Kypris and Markoulidakis showed in [16] that using multicasting for common content

distribution is more viable solution than using unicasting. For the needs of the software

download to the large number of targeted vehicles multicasting is the optimal solution.

3.3. VEHICLE WIRELESS TECHNOLOGIES

Currently, three wireless technologies are implemented for the communication to

the vehicles. Table II shows the summary of characteristics of DSRC, satellite, and

cellular communication.

Table II: Summary of characteristics of the DSRC, Satellite, and Cellular characteristics

DSRC Satellite Cellular Locality Location specific Nation wide Location specific /nation

wide Primary Safety Radio Phone Secondary. Convenience, local

commerce, navigation, Navigation, road information,

Remote diagnostic, driver assistance,

Example ITS XM, OnStar Allocation 5.85-5.925 GHz [17] 2332.5-2345.0 MHz

[18] 824-849 MHz 869-894 MHz[19]

Direction V2V, V2I, I2V I2V V2I, I2V

31

Figure 7: Wireless Infrastructure Overview. Note That Unicasting And Multicasting Can Be Different Channels Of A Single Cellular Tower (i.e., MT And UT Can Be One Tower).

3.4. INFRASTRUCTURE COMPONENTS

The infrastructure consists of a central server (CS), regional managers (RMs),

multicasting towers (MTs), and unicasting towers (UTs), as shown in Figure 7. The

automotive company (AC) or the ECU supplier decides that a certain ECU needs

reprogramming. In other words, for example, only the ECUs of a certain vehicle type,

having a certain vehicle configuration, and produced during a certain time period will be

updated. The vehicles in need for the particular software update form a multicast

audience.

3.4.1. CENTRAL SERVER

The CS keeps the list of all vehicle identification numbers (VINs) of the multicast

audience. High speed Internet connects the CS to all RMs. The CS distributes the

software update and the multicast audience information to all RMs.

32

3.4.2. REGIONAL MANAGER, TOWERS AND CONNECTIONS

An RM connects, by wire, to unicasting towers (UTs) and a multicasting tower

(MT) for a particular geographic region. The RM instructs MT to multicast packets of the

software update to the targeted vehicles. The targeted vehicles, in the region,

periodically unicast information about missed packets back to the UT. The RM collects

the information from the UT and prepares a missing packet list for the region. The order

of the packets in the list is based on the OR method as Munaka, Yamamoto, and

Watanabe presented in [11]. In other words, the resulting list is prioritized by the number

of times a particular packet failed to reach the targeted vehicles. Higher number means

higher priority for that packet in the list. The RM frequently updates the list, for the

duration of the downloading process, with the current missing packet information. In the

consequent multicast round the MT multicasts only the packets from the missing packet

list. The MT sends the packet with the highest priority first.

3.5. VEHICLE HARDWARE

A vehicle is equipped with a wireless communication capable device. Duri et al

[20] defined automotive telematics as the information-intensive applications that are

being enabled for vehicles by a combination of telecommunications and computing

technology.

3.5.1. TELEMATICS CONTROL UNIT

The telematics control unit (TCU) is a vehicle’s wireless transceiver and the

gateway to the intra-vehicle network, such as Controller Area Network (CAN). The TCU

performs the following functions: (1) receives multicast packets from the multicast

carrier, (2) accumulates packets creating the software image in the internal buffer, (3)

33

analyzes created image and composes the vehicle's missing packet list, (4) sends and

receives unicast messages on unicast carrier, (5) periodically reports missed packets

via unicast, (6) runs algorithm that reprograms targeted ECU from the complete

software image, and (7) performs diagnostic and functional tests after the ECU is

reprogrammed. The TCU has logic to determine if it is safe to reprogram the targeted

ECU. The TCU checks the battery and other target-specific conditions. The TCU

reprograms the targeted ECU from the buffer when it becomes convenient (i.e. safety

critical ECU is updated when the vehicle and engine stops, and the ignition is off).

3.5.2. MEMORY BUFFER

An internal buffer is at least large enough to store the old and the new software

images of the largest ECU in the vehicle. If the TCU needs to update multiple ECUs at

the same time (to maintain interaction compatibility) the vehicle can have even larger

buffer to accommodate images of multiple ECUs. If the size of the images exceeds the

memory capability of the buffer then the reprogramming is done traditionally at the

repair facility. In addition, the TCU has available memory for the verification test vectors.

The TCU ensures proper vehicle functionality after the ECU is reprogrammed.

Otherwise, the TCU restores the old software image in the targeted ECU.

3.6. REMOTE SOFTWARE UPDATE

The author breaks the process of software update into collecting the wireless

software packets in the vehicle’s internal buffer (downloading) and updating the targeted

ECU from the internal buffer (reprogramming).

3.6.1. DOWNLOADING

Downloading, illustrated in Figure 8, is a process of collecting the wireless

34

software packets in the vehicle’s internal buffer. A broadcasting tower sends packets

over-the-air to the vehicles. A vehicle’s TCU accumulates packets in the reserved

memory buffer. Process of disseminating recall-software to vehicles does not require

high-quality service. Unlike audio and video streaming where delivery in timely matter is

a must, the delivery time for recall-downloading can be very loose. The nature of

reprogramming requires successful delivery of the entire download in a timely manner.

In contrast to best effort delivery in some data-streaming application, a download is only

valuable if it is complete and error free [21]. The software download is not required

immediately after an automotive company decides on software-recall, but it should be

accomplished as soon as possible. It does not matter if one vehicle receives all recall

packets at one time and another vehicle receives at a later time as long as all packets

are received by most vehicles. However, the protocol requires serving a large number of

vehicles over the same period. For cost reduction sake, downloading over slower

communication medium is adequate

3.6.2. REPROGRAMMING

Reprogramming is a process of updating the image of targeted ECU from the

vehicle’s internal buffer, as illustrated in Figure 9. After the TCU successfully downloads

all the recall packets, the TCU composes software image by arranging the packets in

the correct order. Along executable code, the image contains verification and

compatibility information. The image must be complete and compatible with the current

vehicle configuration before it is uploaded into the targeted ECU. Otherwise the old

image remains in the ECU.

35

3.6.3. VERIFICATION

The vehicle needs to verify integrity and compatibility. Portion of the software

image is dedicated to the digital signature. TCU checks integrity of the software image

by verifying the signature of the complete image. The TCU must make sure that

software version is compatible with the rest of the vehicle. In other words, the new

image must provide the same interfaces to the rest of the vehicle as the old image did.

Automotive company (AC) maintains the list of all possible configurations of the ECUs’

software releases for the targeted vehicle.

Figure 8: Software downloading: wireless communication from a tower to a vehicle

36

Figure 9: Software reprogramming: wired transfer of the software from TCU to the targeted ECU.

AC must first verify that particular software update will work for all vehicle

configurations before sending it for distribution.

If any of the verification steps fail, the TCU rejects the software update and

flushes the receive buffers.

3.6.4. SECURITY

All over-the-air download packets are encrypted at the tower and decrypted at

the TCU end. Tower encrypts packets to provide security and protection of the

downloading process. The actual security mechanism for the software update is not

within the scope of this dissertation.

3.6.5. ERROR DETECTION AND CORRECTION

Forward error correction (FEC) is embedded in the physical layer of the wireless

transmission protocol. FEC is accomplished by adding redundancy bits to the

37

transmitted information. Each redundant bit is a result of a function of original

information bits. The original information may or may not appear in the encoded output.

The original information is recovered after decoding received messages. [22]

3.7. SIMULATION DESCRIPTION, RESULTS AND DISCUSSION

Author created time-based simulation program that helped test proposed

algorithm. The program simulates download process on a wide area of the roads,

wireless communication infrastructure and large number of moving vehicles. Vehicles

are randomly traveling on the roads. The vehicle position and direction is determined in

the initialization phase of the simulation. A vehicle may change direction and speed

when it reaches an intersection. For simplicity, if a vehicle reaches a boundary of the

map, it reverses its direction. The vehicle receives software packets from the tower

while it is moving.

The vehicle may miss receiving a packet due to a weak signal, noise, multipath,

or a change in the coverage area. Failure rate due to signal strength has been assumed

to be proportional to the square of the distance between the moving vehicle and the

multicasting tower. Multipath and noise is simulated using a random function.

3.7.1. INPUTS DESCRIPTION

The simulation inputs are three two-dimensional arrays: “road map”, “tower map”

and “signal-strength-map”. For the “road map” author used input grid from the map of

several cities in the greater Michigan area, as shown in Figure 10.

Simulation input, “tower map”, of tower arrangement is shown in the Figure 11. In

addition to maps, remaining simulation inputs are described in Table III.

38

3.7.2. OUTPUTS DESCRIPTION

During the execution of the simulation author monitors performance in terms of

the number of programmed vehicles, active towers, missed packets, etc. A complete list

of the simulation outputs is shown in Table IV.

3.8. SIMULATION SCENARIOS

The simulation was executed with two sets of the input values Scenario A and

Scenario B, as described in

Table V. The scenarios differ in speed and minimum unicast cost. All other

simulation inputs remain the same. Road and tower arrangement for all simulation runs

are described in Figure 10 and Figure 11.

In Retransmit Broadcast mode, the towers interrogate vehicles for missing

packets. Each tower composes prioritized retransmit list of up to cnf_twr_mis_pck size.

Second, the towers send packets from the list based on priority. Due to interrogation

delay, number of active towers reaches a low point in the beginning of Retransmit

Broadcast mode at t = 24 sec. The interrogation of the vehicles and retransmission of

the packets from the list repeats multiple times until a threshold of programmed vehicles

is reached or software time-to-live expired.

39

Figure 10: Greater Michigan map

Figure 11: Representation of the tower arrangement in the Greater Michigan area.

40

Table III Configurable Inputs to the Simulator

name description unit cnf_twr_mis_pck number of packets tower sends before

collecting the missing packets information from vehicles in the Retransmit Broadcasting mode

-

cnf_acqst_time measurement update rate ms cnf_time_instant_us time instant of the simulation ms cnf_ini_dly_twr_fctr initial delay of the multicasting of a

tower in the Retransmit Broadcasting mode

ms

cnf_add_uc_dly_mb_us additional unicast delay per vehicle μs cnf_max_uc_dly_mb_trsh maximum number of unicasting

vehicles before saturation of the unicasting backhaul

-

cnf_min_uc_dly_sec minimum unicast delay regardless of the number of vehicles

s

cnf_twr_actv_aftr_brcst_time total time a tower is active after initial broadcast of the software

s

cnf_num_twr total number of towers - cnf_num_mob total number of vehicles - cnf_num_pck number of packets in the software for

the targeted ECU (size of ECU_SW) -

cnf_max_mob_spd maximum vehicle speed m/s

Table IV Output parameters of the simulator

name description acq_glb_tm time instant of the measurement acq_num_twr_act number of towers currently broadcasting acq_num_prg_pck number of programmed packets acq_num_prg_mob cumulative number of programmed vehicles

(vehicles that received entire ECU_SW) acq_rnd_mis_pck number of missed packets (increments when

vehicle in the coverage area fails to receive) acq_unprg_mob_cvrd_rgn number of unprogrammed vehicles that are in

the coverage area at the current instant of time,

acq_unprg_mob number of unprogrammed vehicles acq_twr_pck_snt number of packets sent from all towers acq_uncst_comm number of unicasting messages from vehicles

to the towers

41

Table V Simulation input values

Scenario A Scenario B cnf_acqst_time 0.2 0.2 cnf_time_instant_us 25000 25000 cnf_ini_dly_twr_fctr 0.5 0.5 cnf_add_uc_dly_mb_us 100 100 cnf_max_uc_dly_mb_trsh 5 5 cnf_min_uc_dly_sec 500 50000 cnf_twr_actv_aftr_brcst_time 500 500 cnf_num_twr 1323 1323 cnf_num_mob 100000 100000 cnf_num_pck 1000 1000 cnf_max_mob_spd 50 17

3.9. SIMULATION RESULTS

3.9.1. NUMBER OF TOWERS ACTIVELY BROADCASTING

The towers work in two modes Initial Broadcast and Retransmit Broadcast, as

shown in Figure 12 and Figure 13. In the beginning of the Initial Broadcast, the towers

gradually become active (start to broadcast). At about t = 5 s, all 1323 towers are

broadcasting. Towers continue to broadcast all 1000 packets of the software package.

After the Initial Broadcast at around t = 24 s, towers start working in Retransmit

Broadcast mode.

3.9.2. TIME DURATION FOR PROGRAMMING VEHICLES

Vehicle mobility with respect to towers is one of the reasons for packet loss. A

vehicle moves from Tower X to Tower Y. While the vehicle was under Tower X packets

coming form Tower X were packets that the vehicle needed. When the vehicle is under

Tower Y the packets arriving from Tower Y are not necessarily the packets that the

vehicle needs. If the vehicle stays long enough under Tower Y then the packets coming

from Tower Y in the subsequent retransmission session will include packets that the

42

vehicle needs. Figure 14 and Figure 15 show how many vehicles are becoming

completely programmed as the time progresses.

3.9.3. UNPROGRAMMED VEHICLES IN COVERAGE AREA

Initially vehicles are randomly distributed on the roads on the map (Figure 10).

Some vehicles are not in the coverage area (Figure 11), and are not receiving packets.

Unprogrammed vehicles enter and leave a towers’ coverage area. In other words, the

audience changes dynamically as shown in Figure 16 and Figure 17. The time values

for the system to reach 90% completely programmed vehicles threshold for different

retransmission buffers is presented in Table VI. From Figure 14, Figure 16 and Table VI

it is clear that system with smaller retransmission buffer size for the scenario A performs

better in terms of reaching sooner the 90% mark of completely programmed vehicles.

Better performance is because of the relatively high speed of the vehicles and low cost

of the unicast reporting of the missing packets.

3.10. ANALYSIS OF THE TWO SCENARIOS

Author picked two scenarios to show different algorithm behaviors for different

retransmission buffer sizes. The algorithm is performing better for smaller buffer sizes,

as shown in Figure 14, when the unicasting cost is low and vehicle speeds are high. A

smaller buffer size enables towers to inquire vehicles frequently and therefore compile

up-to-date missing packet list. Towers have a greater chance to broadcast what is

needed for whom that is needed, as shown in Figure 16.

However, when the average vehicle speed is low and unicasting cost is high

there is less demand for frequent update of the missing packet list. Distribution of the

vehicles under the coverage area does not change often. Unicasting cost is affecting

43

the algorithm performance because there is less significance in having frequent vehicle

inquires. In this case, system with larger buffer sizes performs better, as shown in

Figure 15.

Figure 12: Scenario A: Number of towers actively broadcasting vs. time as the size of the retransmit buffer varies.

44

Figure 13: Scenario B: Number of towers actively broadcasting vs. time as the size of the retransmit buffer varies.

Figure 14: Scenario A: Number of programmed vehicles vs. time as the size of the retransmit buffer varies.

45

Figure 15: Scenario B: Number of programmed vehicles vs. time as the size of the retransmit buffer varies.

Figure 16: Scenario A: Number of unprogrammed vehicles in coverage area

46

Figure 17: Scenario B: Number of unprogrammed vehicles in coverage area

Table VI 90% completely programmed vehicles threshold

Retransmit buff.

Time at >90% of the total vehicles are programmed

Time at <10% unprogrammed vehicles within the coverage area

Scenario A Scenario B Scenario A Scenario B [packet] [second] [second] 50 169.2 312.6 133.6 263.198 100 170.4 240.0 134.4 189.198 200 174.4 209.0 137.6 155.599 400 175.8 195.0 139.8 138.399 600 179.0 191.2 142.8 133.399 800 180.9 191.0 144.8 131.799 1000 180.0 190.6 144.6 130.199

47

CHAPTER 4.

AN ENHANCED DATA REDUCTION (EDR) ALGORITHM FOR EVENT-TRIGGERED

NETWORKS

Unlike other existing data reduction algorithms, this algorithm does not use a

Data Compression Bit (DCB) to indicate whether or not a message has been

compressed. Since all the receivers of a particular message know the length of the

uncompressed data field of the message, the receiver, by looking at the length of the

data field, will know whether or not the message has been compressed. This eliminates

the difficulties associated with earlier solutions to identification of compressed

messages such as the use of the reserved bit [36] or the use of the dedicated message

IDs [40], or made the decoding more involved via the additional bit in the data field [42].

Since every data reduction algorithm has some overhead such as using a Data

Compression Bit (DCB) and/or using a Data Compression Code (DCC), the signal

values could be such that after applying the data reduction algorithm the resulting

message may not be shorter than the original uncompressed message. In the

Enhanced Data Reduction (EDR) algorithm, a message will be compressed provided

the length of the data field of the compressed message is less than that of the original

uncompressed message. This means that, after applying the EDR algorithm on a

particular message if it is seen that the length of the data field of the resulting message

is less than that of the original uncompressed message, then a compressed message

will be sent; otherwise, the original uncompressed message will be sent. Therefore, the

EDR method either reduces the amount of data sent, or in the worst case, neither

results in any savings nor produces any overhead.

48

4.1. SIGNAL TYPES AND SIGNAL REDUCTION MECHANISMS

In the EDR algorithm, the coding of compressed signals is done in a similar way

as it was done for the IADR algorithm. However, author defines stricter rules and some

modifications. Signals can be of any bit in length (BL). Typical signals represent either

continuous functions or discrete states. A signal is continuous if its value does not

change significantly from one message transmission to the next. Examples of some

continuous signals are vehicle speed, engine temperature, engine RPM, etc. Examples

of some discrete signals are engine state, gear state, headlight state, wiper state, etc. In

this dissertation, the discrete signals are also called state signals. In order to reduce the

compression overhead and make the EDR algorithm an effective algorithm, author

proposes to combine several state signals into one signal called the group signal. Since

state signals do not change very frequently, the group signal that is made using a

number of state signals will not change very frequently. As a result, significant reduction

can be made in a message.

In the EDR algorithm, the encoded message can have two types of signals. The

first type of signal is SDN (Signal, Delta, No-change). SDN allows a signal to be

represented in entirety, as a delta-change, or as no-change. The second type of signal

is SN (Signal, No-change). SN allows a signal to be represented in entirety or as no-

change. Continuous signals such as vehicle speed, engine temperature, engine RPM,

etc. can be encoded as SDN signals. Since there is some overhead involved in

encoding an uncompressed message into a compressed message, a continuous signal

must be greater than or equal to a minimum length for it to be encoded into a

compressed signal. Otherwise, the data reduction algorithm will not provide any

49

benefits. For each signal to be compressed, there is a coding overhead of two bits: one

bit of the Data Compression Code (DCC) which indicates whether the signal has been

compressed or not, and another bit, reduction type (RT) bit, that precedes the

compressed signal which indicates whether the signal has been fully compressed or

delta compressed. Since the delta value of a signal needs at least two bits, one bit for

the sign and at least another bit for the value, the total cost of delta compression is 4

bits including the two bits required as coding overhead. Thus, a continuous signal has to

be at least 5-bit long in order get some benefit out of the compression technique. State

signals with data length from 1 to 4 bits can be grouped together and be represented as

an SN signal. Figure 18 shows an example of the vehicle signal and its corresponding

delta representation.

Using a delta size that is half the size of the signal will be more effective on the

smaller size signals than on the larger size signals. For example, let us examine 8 and

16-bit signals. Let us say that the engine ECU keeps track of the vehicle speed. The

ECU publishes the vehicle speed information in one of its periodic messages with the

period T=10 ms. The vehicle speed signal has the range of -100 to 400 km/h, and in the

first case it is represented with 16-bit variable. Each count in the 16 bit variable has a

weight of (500km/h)/(65536 counts)=0.00763km/h/count. The delta size is 8 bits and it

spans smhkm /26915.0/96895.0 ±=± (due to the sign bit plus the 7-bit magnitude). If the

vehicle speed is sent once every 10 ms, then in order to be able to represent the signal

with its delta form, the acceleration of the vehicle should not exceed 2/9152.26 sm± . In

the real world, forward acceleration of a very fast vehicle may go up to 6.4 m/s2 [42]. Of

course, here author assumes that the signal is stable and noise free.

50

Figure 18: Two examples of Vehicle Speed Signal and Its Delta Representation.

Table VII Delta Span versus Signal Size

Signal Size (bits)

Delta Size (bits)

Signal Range Abs. value

Delta Range Abs. value

Delta Span (%)

5 2 0-31 0-1 6.25 6 3 0-63 0-3 6.25 7 3 0-127 0-3 3.125 8 4 0-255 0-7 3.125 9 4 0-511 0-7 1.5625 10 5 0-1023 0-15 1.5625 11 5 0-2047 0-15 0.78125 12 6 0-4095 0-31 0.78125 13 6 0-8191 0-31 0.39062514 7 0-16383 0-63 0.39062515 7 0-32767 0-63 0.19531316 8 0-65535 0-127 0.195313

51

Figure 19: An example of the original message and its mask with SDN and SN type groupings.

In the second case, the vehicle speed signal is represented with an 8-bit variable,

where each count is ( ) ( ) counthkmcountshkm //96078.1255//500 = . Here, the delta size is 4

bits and it spans smhkm /81264.3/72549.13 ±=± (sign plus the 3-bit magnitude). In order to

represent the signal with the delta, the acceleration of the vehicle now should not

exceed 2/2636.381 sm± . In this second case, all realistic consequent vehicle speed

signals can be represented with delta.

Following the proposed logic in choosing the delta size, Table I presents the

delta span for various signals. For example, the size of the delta for a 5-bit signal is 2.

The 2-bit delta spans 6.25 % of the signal. However, the 8-bit delta for 16-bit signal

spans only 0.1953 % of the signal. If all signals have the same changing probability,

from Table I it is obvious that larger size signal will have less chance to be represented

with delta form.

52

Figure 20: An example of the original message and its mask with only SN type groupings.

Figure 19 shows an example of an original message and its message mask. All

signals of size higher or equal to 5 bits (vehicle speed, coolant temp, and engine RPM)

are represented with SDN type masks (SDN_0, SDN_1, and SDN_2). On the other

hand, signals of smaller size than 5 bits (fan RPM, oil sensor, engine state, gear state,

and park status) are grouped into SN type masks (SN_0 and SN_1). Similarly, Figure

20 shows a message with all signals being of size smaller than 5 bits and its message

mask representation.

The proposed rules for creating a reduced message in the EDR algorithm are as

follows:

1. any signal with bit length BL ≥ 5 is masked with SDN type

2. only signals with BL ≥ 5 are represented with delta (∆)

3. size of ∆ = Integer (BL/2)

4. the first bit of ∆ is the sign bit, and the rest of the bits indicate magnitude

5. signals with BL < 5 are bundled into SN groups

6. maximum length of an SN group is 8 bits

4.2. SUMMARY OF SIGNAL REPRESENTATIONS IN THE EDR ALGORITHM

53

In the EDR algorithm, the data field of a compressed message starts with a Data

Compression Code (DCC). The DCC is followed by signal representation fields, where

the values of different signals are expressed in compressed or uncompressed forms

depending upon their current values compared to the values during the previous

transmission of the message. Like the IADR algorithm, the DCC carries information on

whether the signals are encoded in reduced form or in entirety. The DCC contains

compression code bits for both SDN and SN type signals. A DCC bit of a signal

indicates if the signal is presented in reduced form (bit =1) or in its original form (bit =0).

If an SDN type signal is not reduced, then its signal representation field contains

its value in entirety. If an SDN type signal is reduced, then the first bit of the

corresponding signal representation field is the reduction type bit (RT bit). The RT bit

indicates the type of compression: delta compressed (RT=0) or fully compressed

(RT=1). If the change in a signal value can be expressed in delta (∆) form, then the RT

bit of the signal is 0, i.e. for this particular signal, the signal representation field contains

a 0 followed by the value of ∆. If the current value of a signal is the same as the

previous value of this signal, then the signal is fully compressed (RT=1), i.e. no signal

value is sent. For this signal, the signal representation field contains only the RT bit

which is a 1.

If an SN type signal is compressed, the compression type is always fully

compressed. That is no signal value is sent. Thus, no RT bit is necessary in the signal

representation field of an SN signal.

4.3. EDR MESSAGE ENCODING AND DECODING ALGORITHMS

Now we are going to explain the encoding technique of a message and show

54

how a message is converted into a reduced encoded message using the EDR

algorithm. Assume that the message format is the one shown in Figure 19. Note that

this message has three SDN type signals and two SN type signals. Assume that from

the previous transmission to the current transmission of the message, its signals

changed as follows:

1. Vehicle Speed changed significantly. Thus, it can’t be represented in delta (∆) form.

So no data reduction is possible for Vehicle Speed. As a result, the DCC bit for

Vehicle Speed is 0.

2. Coolant Temperature changed slightly, and the change can be represented in delta

form. Thus, the signal is delta compressed. As a result, the DCC bit for Coolant

Temperature is 1. Since the signal is delta compressed, the RT (Reduction Type) bit

which precedes the actual delta signal is 0 for Coolant Temperature.

3. Engine RPM did not change. Thus, no data needs to be sent for Engine RPM which

means the signal is fully compressed. Hence, the DCC bit for this signal is 1. Since

the signal is fully compressed, the RT bit is 1 and it is not followed by any data for

Engine RPM.

4. Fan RPM and/or Oil Sensor State changed. Thus SN_0 signal which is a

combination of Fan RPM and Oil Sensor State can’t be compressed. As a result, the

DCC bit for this signal is 0.

5. Engine State, Gear State and Park Status did not change. Thus SN_1 signal is fully

compressed and no data needs to be sent for this signal. Hence, the DCC bit for this

signal is 1. Since SN type signals do not have any RT bits, this signal does not have

any RT bit.

55

Figure 21 shows the reduced encoded message for the above example. Figure

22 shows the flowchart of the EDR message encoding algorithm. In this figure, the

notation Si and Si-1 are used to indicate the current and previous values of a signal,

respectively. The notation Δmax indicates the maximum change that a signal can have

from its previous value in order for the signal to be represented in delta form. For each

signal of a message, the transmitting node will execute the algorithm shown in Figure

22 in order to create the final message that will be sent through the bus. The receiving

node will have to execute a decoding algorithm in order to extract signals from a

message. Figure 23 shows the EDR message decoding algorithm in detail. For every

signal to be extracted from a message, the receiving node has to execute the decoding

algorithm shown in Figure 23.

Figure 21: An example of the Data Field of a Reduced Message.

56

Figure 22: Flowchart of EDR message encoding algorithm.

4.4. COST OF EDR ALGORITHM

Here we determine the cost of the EDR algorithm in terms of memory required to

store the code and data. The encoding and decoding algorithms can be converted into

two subroutines. A node will repeatedly call these routines every time it wants to encode

and decode signals of a message. There are 12 decision and assignment blocks (high-

level operations) in the encoding algorithm as shown in Figure 22. Similarly, Figure 23

shows that there are 9 decision and assignment blocks in the decoding algorithm. Three

assembly/machine level instructions, LOAD, COMPARE and BRANCH, are necessary

to implement the operation of a decision block. In order to implement the operation of an

assignment block, we need one to three machine level instructions, such as LOAD-

ADD-STORE or BITTEST-BRANCH or STORE and so on. Thus, the maximum number

of machine level instructions required to implement a decision or assignment block is

three. Since altogether there are 21 decision and assignment blocks in the encoding

and decoding algorithms, at most 63 machine level instructions are necessary to

57

implement the EDR algorithm. For a typical microcontroller such as a PIC18 family

microcontroller, the average number of bytes required to implement a machine level

instruction is two. Hence, 126 bytes of ROM will be necessary to keep the code of the

EDR algorithm. Each node will also need some memory for receive and transmit

buffers. A typical CAN node for vehicular applications does not transmit too many

different types of messages and also does not take actions based upon too many

different messages from remote nodes. If we assume that a node transmits 10 different

messages and receives another 10 different messages from remote nodes, then

altogether the node needs 20 buffers. Since the maximum length of the data field of a

CAN message is 8 bytes, a node needs 160 bytes of RAM for transmit and receive

buffers. Hence, the total cost of the algorithm in terms of required memory bytes for

code and data is 126 + 160 = 286 bytes. This cost is insignificant compared to the

amount of ROM and RAM available in today’s microcontrollers.

Figure 23: Flowchart of EDR message decoding algorithm.

58

4.5. IMPACT OF EDR ALGORITHM ON MESSAGE LATENCY

In this subsection we investigate the impact of EDR algorithm on end-to-end

message latency. Both the encoding and decoding algorithms will incur additional

message latency. The total impact on the message latency will be equal to the

summation of the impacts due to encoding and decoding algorithms. The impact of

encoding algorithm on the latency of a particular message depends on the CPU time

necessary to execute the encoding algorithm for all the signals of that particular

message. In order to determine the worst case impact of the encoding algorithm on the

latency, we determine the computation time necessary for the longest path of the

flowchart shown in Figure 22. The first column of Table VIII shows the high-level

statements present in the longest path of the encoding algorithm. The second column

shows the assembly/machine-level instructions necessary to implement each high-level

statement. The third column shows the total number of processor clock cycles

necessary to execute each high-level statement assuming that it takes four processor

clock cycles to execute each machine-level instruction. Table VIII shows that in the

worst-case 48 clock cycles are necessary to encode a signal using the EDR encoding

algorithm.

Table VIII Worst-Case Computation Time to Encode a Signal using EDR Algorithm

High-Level Statement Assembly/Machine-Level Statements

Clock Cycles

1. Signal Type? (SDN or SN) LOAD, COMPARE and BRANCH 12 2.∆ = Si – Si-1= 0? LOAD, COMPARE and BRANCH 12 3. ?maxΔ≤Δ LOAD, COMPARE and BRANCH 12 4. Set DCC bit and clear RT bit of Si. BIT-SET and BIT-CLEAR 8 5. Send ∆ STORE 4 Total clock cycles 48

59

Similarly, we can determine the impact of the decoding algorithm on the

message latency. Table IX shows the high-level statements present in the longest path

of the decoding algorithm, the corresponding assembly/machine-level instructions and

clock cycles necessary to execute each high-level statement. Table IX shows that for

the worst-case, the total number of clock cycles necessary to execute the EDR

decoding algorithm for a signal is 44.

Table IX Worst-Case Computation Time to Decode a Signal using EDR Algorithm.

High-Level Statement Assembly/Machine-Level Statements

Clock Cycles

1. Signal Type? (SDN or SN) LOAD, COMPARE and BRANCH

12

2. DCC bit of Si= 1? BIT-TEST and BRANCH 8 3. RT bit of Si= 1? BIT-TEST and BRANCH 8 4. Accept ∆ from the receiving buffer LOAD 4 5. Si = Si-1 + ∆ LOAD, ADD and STORE 12 Total clock cycles 44

From Table VIII and III we see that the worst-case computation time for encoding

and decoding a signal using the EDR algorithm is 48+44=92 clock cycles. It is

mentioned earlier that SDN type signals are at least 5-bit long, and SN type signals are

generated by combining a number of signals whose lengths are less than 5 bits. Since

the maximum number of data bits in a CAN message is 64 and we need DCC bits for

encoding and decoding signals, the total number of SDN and SN type signals in a CAN

message can’t be more than 11. If we assume that in the worst-case there are 11

signals in a CAN message, then the total impact of the EDR algorithm on the message

latency is 92*11 = 1012 clock cycles. Note that a receiving node checks the data length

60

code of a message to determine whether or not the message has been reduced.

Checking the data length code is necessary only once for each message, and the

implementation of this operation needs three assembly/machine level instructions:

LOAD, COMPARE and BRANCH. Thus, another 12 clock cycles are necessary to

encode and decode the algorithm. Hence, the overall impact of the EDR algorithm on

the message latency is 1012+12 = 1024 clock cycles. Various PIC18 family

microcontrollers have built-in CAN modules, and these microcontrollers can use clocks

up to 40MHz frequency. So if a 40MHz oscillator is used with a PIC18 family

microcontroller, then the worst-case impact on the end-to-end message latency is

1024cycles*0.025 microseconds/cycle = 25.6 microseconds. Even if a 10MHz oscillator

is used, the worst-case impact on the latency is 102.4 microseconds which is

acceptable for any safety applications. Note that if a vehicle moves at 100 miles/hour, it

can move only 0.18 inch in 102.4 microseconds. Thus, if the EDR algorithm is used with

a microcontroller with a 10MHz oscillator, the vehicle will move an additional 0.18 inch

before the receiving node can take actions based on the message contents. As a result,

we believe that vehicle safety will not be compromised due to the use of the EDR

algorithm. Moreover, a higher frequency oscillator can be used with the microcontroller,

and for most cases the vehicle speed will be less than 100 miles/hour which will even

lessen the impact of the EDR algorithm on message latency and consequently on the

safety of the vehicle.

4.5.1. SYNCHRONIZATION

The presented data reduction methodology relies heavily on uninterrupted

synchronous communication. A synchronization problem arises when the receiving

61

node goes through a reset and loses all information on the previous signal values. All

signals in the receiving buffers of the resetting node become invalid. Another loss of

synchronization occurs when one or more messages are incorrectly received and the

receiver disregards the content of the messages and disrupts the ability to reproduce

the signals associated with incorrectly received messages. Encoded messages that the

node receives afterward are not sufficient to reconstruct the signals any more.

It is mentioned earlier that a Cyclic Refresh technique could be used to keep all

receiving nodes synchronized with the transmitting nodes. In this technique, each signal

is sent in entirety in a round-robin fashion. Thus, if a message has S signals, then the

receiving node which went through a reset has to wait for S transmissions of the

message to get the values of all signals of the message. As a result, the receiving node,

after went through a reset, will have to wait for a long time before it can use all the

signals that came from remote nodes. If this long delay for a particular receiving node is

not going to cause any safety problems for the vehicle or any major functionality

problems for the receiving node, then the receiving node can take this reactive

approach to update its messages.

The resetting node can also take a proactive approach to update its messages.

One proactive approach could be considered as a Demand Refresh, where the resetting

node could request all transmitting nodes to send their messages with signals in

entirety. The Demand Refresh could significantly increase the bus utilization for a short

period of time depending upon how many resetting nodes are requesting how many

messages at a time. In order to update all messages at the resetting node, the Demand

Refresh will incur significantly less waiting time compared to that for Cyclic Refresh. A

62

second proactive approach could be as follows: if the functionality of the receiving node

requires that immediately after reset, the node must control some actuators based on

some signal values from a remote node, then the receiving node must request

appropriate remote nodes to send only those required signals in entirety. The remaining

signals could be updated based on the Cycle Refresh technique. This second proactive

approach will incur less bus utilization compared to that for Demand Refresh. If the

receiving node is not required to take any actions immediate after reset, then another

proactive action can be taken as follows: the receiving node can wait until it gets

messages from transmitting nodes according to the schedule of the transmitting nodes,

and if the signals arrive in entirety, then the receiving node can accept the signals and

take actions. However, if the signals arrive in reduced form, then at that time the

receiving node can demand the transmitting node to send the signals in entirety.

In summary, we would like to say that we have suggested a number of

techniques using which a resetting node can be synchronized with the transmitting

nodes. Since the reset event is not a common phenomenon for a node, and not too

many nodes will go through a reset at the same time, except during the initial startup of

the vehicle, we believe that such an event will not have any significant impact on the

performance of the proposed data reduction protocol. Since the reset events are not

periodic, they could be considered as rare time-to-time glitches in the operations of in-

vehicle networks. Moreover, since under normal operations, the loading of an

asynchronous bus like a CAN bus is kept around 30%, occasional rare glitches could be

tolerated by the bus. Thus, we believe that node resetting events will not be major

issues as far as the performance of the EDR algorithm is concerned.

63

4.5.2. HANDLING OF INITIAL TRANSIENTS

Data reduction techniques promise to reduce the amount of bus traffic while

transmitting the same required information content. The practical attractiveness of this

lies in the opportunity it provides to add more messages. In other words, it is the desire

to send more messages through the bus without exceeding the bus load beyond the

acceptable maximum limit that derives the need for data reduction. However, while

assessing the benefits of a data reduction technique, with respect to the room it

provides for additional messages, it is important and sometimes neglected, to consider

the initial conditions.

At the vehicle boot up, the transition from ignition-key off to on, the bus becomes

active. Many ECUs start sending periodic messages. At this initial stage, the receiving

(RX) and transmitting (TX) buffers for the signals are not initialized. Therefore, all initial

messages need to be sent in entirety. This condition creates peak bus load.

To eliminate this peak bus load we propose that all messages be prioritized into

vehicle functionality-critical and non-critical groups. For example, vehicle functionality-

critical group would consist of messages carrying critical information for vehicle

functionality such as engine, torque, and brakes information. Messages carrying leisure

and convenience information such as AC, radio, and display would comprise the non-

critical group. For some initial period, for example 200 ms, only vehicle functionality-

critical messages could appear on the bus. After the initial period, when the RX and TX

buffers of vehicle functionality-critical messages have been filled, the rest of the non-

critical messages would go on the bus. At that time, the vehicle functionality-critical

messages are sent in reduced form while the initial messages of the non-critical group

64

are sent in entirety. This method eliminates the peak load of having all messages being

sent in entirety at the same time.

4.5.3. ADAPTATION OF EDR ALGORITHM TO OTHER IN-VEHICLE BUSES

The EDR algorithm can be used for any asynchronous bus where time slots are

not reserved for messages to use the bus. For a CAN-bus system, the EDR algorithm

requires the receiving node to check the data length code to determine whether or not

the message has been reduced. If the CAN protocol did not have a data length code in

the message, then the first bit of the data field could have been used as a compression

bit to indicate whether or not the message has been reduced. Similar coding techniques

could be designed for those in-vehicle networking protocols where a data length code is

not available within the message. Thus, the proposed EDR algorithm can also be used

for LIN protocol. If the EDR algorithm is going to be used for the LIN protocol, then the

first bit of the response field of a LIN message should be used as a compression bit.

This compression bit will indicate whether or not the LIN message has been

compressed. The compression bit will be followed by the Data Compression Code

(DCC). The DCC will then be followed by the Signal Representation Fields as shown in

Figure 7.

The FlexRay protocol can support a bit rate of up to 10Mps, which is high enough

for any real-time automotive applications. Thus, in the near future, data reduction may

not be necessary for a FlexRay bus. However, EDR algorithm can also be used for the

FlexRay protocol. A FlexRay message frame has two types of segments: static segment

and dynamic segment. The static segment contains a number of static slots, and each

static slot is reserved for a particular time-triggered message that needs deterministic

65

latency. Thus, data reduction techniques should not be applied for the static segment.

The time slots in the dynamic segment are not reserved for any particular messages.

Since a FlexRay frame also has a field called the Payload Length (analogous to the

data length code of CAN), the proposed EDR algorithm can be used in the dynamic

segment of FlexRay in a similar way as it can be used for CAN. The EDR algorithm can

be used with other in-vehicle networking protocols as long as some time slots are

available which are not reserved for any particular messages.

4.6. ADVANTAGES OF EDR ALGORITHM COMPARED TO OTHER EXISTING

DATA REDUCTION ALGORITHMS FOR VEHICULAR APPLICATIONS

In this subsection we compare EDR algorithm with other exist data reduction

algorithms in terms of bit savings and complexity of detecting a reduced message.

4.6.1. ADVANTAGES OF EDR COMPARED TO DR

Bit Savings Comparison: Let us consider a periodic message with an 8-bit

numerical signal and period T. An example signal from Figure 24 occupies the third data

byte of the message. We will examine how the signal is represented in DR and EDR

methodology.

Figure 24: Example of a Message with an 8-Bit Signal.

66

Figure 25: Example of Signal Behavior and Signal Representation in EDR and DR.

Figure 25 shows a hypothetical signal behavior and the savings that can be

achieved with EDR and DR methodologies. According to the rules we presented earlier,

the delta span of the signal is ± 7. The very first message at t = 0 has the signal

represented in entirety. The next message at t = T contains the delta change, followed

by several messages where the signal does not change. All other messages after t =

5T, besides the message at t = 10T, can have the signal represented as a delta. On the

other hand, in DR methodology, only signals for messages at t = 2T, 3T and 4T can be

reduced.

67

Reduced Message Detection: The major drawback of DR method is using the reserved

bit of the CAN message frame as the DCB flag. Existing transceivers and

microcontrollers [43], [44] often assume a set value of the reserved bit. Hence, for the

DR methodology to work, existing hardware would have to change. EDR improvement

is to examine the length of the received message and based on the length to determine

if the message frame is in original or reduced form. For example, consider the original

message Y being 8 bytes long. If a node receives message Y with 5 bytes in length,

then the received message is in some reduced form and has to be decoded to extract

the signals.

4.6.2. ADVANTAGES OF EDR COMPARED TO ADR

Bit Savings Comparison: Reduction fields of the ADR are fixed in length. Let us

consider an example of an original message with four signals. After the original

message has been sent, the data reduction algorithm is in place. A transmitter prepares

the next message. Assume that one signal fails to be represented with a delta value,

and other three can be represented with delta values. In ADR, the original message is

transmitted. However, in EDR, if one signal fails to be represented with a delta value

and other three signals can be represented with delta values, the reduced message is

still transmitted. In addition, in ADR, no-change is represented with one plus size of the

delta in bits, while in EDR, no-change is represented with only two bits.

Reduced Message Detection: In ADR having a sub set of message IDs with the

sole purpose to denote reduced messages, complicates the network. If all messages

are covered with the ADR algorithm, network designer will have only half of the

available IDs to work with. In addition, the network designer can assign only every other

68

message ID since the consequent message ID will denote a reduced message. In EDR,

the network designer can use all message IDs.

4.6.3. ADVANTAGES OF EDR COMPARED TO IADR

Reduced Message Detection: In the IADR algorithm, data content is shifted to

the right by one bit to make room for the Data Compression Bit (DCB) bit. A reduced

message is detected if the DCB, the first bit of the data field, is set. In EDR, offsetting is

not needed because a reduced message is detected by the length comparison of the

received message to the uncompressed message.

Though the first bit of the data field can be used as the DCB flag, it is

inconvenient to shorten the message which is already using the entire data field. For

example, in J1939 protocol [45], SAE defines signals that span the 8-byte data field of

the CAN frame. In this case, IADR is not applicable since IADR reserves one bit out of

data field for DCB.

4.7. PERFORMANCE ANALYSIS

In this section, we present a theoretical analysis of the EDR algorithm to show

how EDR algorithm performs. We also present simulation results which were collected

by running a software simulation using both real and synthetic data.

4.7.1. CAN PROTOCOL

Fig. 12 shows all fields of a standard CAN message. Fig. 13 shows the fields in

detail. The CAN protocol exclusively uses 1/0 (recessive/dominant) edge transitions for

synchronization. To maintain synchronization between all nodes on the bus, sufficient

edge transitions are needed. The stream of all recessive or all dominant bits is

insufficient for synchronization. Therefore, bit stuffing has been introduced in the

69

protocol. After every 5 consecutive identical bits, a complement bit, the stuff bit, is

inserted [46]. The stuff bit gives a forced transition. As a result, the receiving nodes can

synchronize with the transmitting node.

Figure 26: Data frame.

Figure 27: The Data Frame of a CAN Message.

Table IV: List of Notations used in Analysis

Notation Description Unit ∆T Period [sec] BU bus utilization for original message traffic [-] BU* bus utilization for EDR message traffic [-] const_li constant length of the original message i [sec] DL(i,j) length of delta signal j of message i [bit] li(t) length of message i at time t [sec] Li(t) length of unstuffed content of message i at time

t [sec]

i message id [-] j signal index [-] N number of messages in the network [-] ni number of occurrences of message i within ∆T [-]

70

Ni(t) number of data bytes in message i [byte] sbi(t) length due to stuff bits in message i at time t for

the original message traffic, 0 ≤ sbi(t) ≤ 19 [bit]

sbi*(t) length due to stuff bits in message i at time t for EDR message traffic, 0 ≤ sbi(t) ≤ 19

[bit]

SL(t,i,j) length of signal j of message i at time t [bit] Si number of signals in message i [-] OL overhead length [bit] PNC(i,j) probability of signal being represented with no

change during t = 0 ≤ t ≤ ∆T [-]

PES(i,j) probability of signal being represented with in entirety during t = 0 ≤ t ≤ ∆T

[-]

PDS(i,j) probability of signal being represented with delta change during t = 0 ≤ t ≤ ∆T

[-]

PI Performance Improvement [-] wb{} Whole byte operator. Returns the number of bits

of the whole-byte datum. If mod(a,8)=0 then wb{a}=a else wb{a}=a+(8-mod(a,8)).

[bit]

4.7.2. THEORETICAL ANALYSIS OF THE EDR ALGORITHM

All the parameters used in this performance analysis are shown in Table IV.

The bus utilization, BU, of a CAN bus in general, is the sum of the lengths, li(t), of all N

messages in the network divided by the period, ΔT, for which the BU is calculated. The

sum of the lengths is expressed in seconds.

T

tlBU

N

i

T

ti

Δ=

∑ ∑=

Δ

=1 0

)( (1)

For the original messages, the length of messages, const_li, does not change

over time (ignoring changes due to bit stuffing). Thus, the expression for BU becomes

( )

T

lconstnBU

N

iii

Δ=

∑=1

_* (2)

Figure 27 shows the portion of the data frame susceptible to bit stuffing. The

overall data frame length, li(t), is the sum of the unstuffed message length, Li(t), and

71

stuff bits sbi(t), i.e.

)()()( tsbtLtl iii += (3)

The maximum number of bits associated with a standard message including the

stuff bits is 130, as shown in Table V.

Table X Maximum number of bits in a Standard ID (11 bit) CAN message

Number of Bits Description 1 Start of frame 11 Identifier 1 RTR 6 Control bits 64 Data (8 byte) 15 CRC 1 CRC delimiter 1 ACK 1 ACK delimiter 7 End of frame 3 Interframe space 19 Stuff bits (maximum) 130 Total (maximum)

Unstuffed message length, Li(t), is the sum of all signal lengths, SL(t,i,j), of the

data field and overhead bits OL. OL is the sum of all bits in the message frame except

the data field and stuff bits, and that is 47. The sum of all bits in the data field can be

expressed as

∑=

+=iS

ji jitSLOLtL

1

),,()( (4)

Substituting (3) in (1) we get

[ ]

T

tsbtLtBU

N

i

T

tii

Δ

+=

∑ ∑=

Δ

=1 0

)()()( (5)

72

Further substituting Li(t) from (4) in (5) we get

T

jitSLtsbOL

tBU

N

i

T

t

S

jii

i

Δ⎥⎥⎦

⎤

⎢⎢⎣

⎡++

=∑ ∑ ∑

=

Δ

= =1 0 1

),,()(

)( (6)

For the original message traffic, the signal length does not change from one

sending to the next. This means that

),(_),,( jiSLconstjitSL = (7)

Therefore, BU for the original message traffic is

T

jiSLtsbOLn

tBU

N

i

S

jii

i

Δ⎥⎥⎦

⎤

⎢⎢⎣

⎡++

=∑ ∑

= =1 1

),()(

)( (8)

For reduced messages, the signal length of a particular message can be of

different value from one message to another message.

Table XI Possible ),,( jitSL Values

Signal Compression Type SL Value (bits) signal presented with no change 2 signal presented in entirety 1 + SL(i,j) signal presented as delta change 2 + DL(i,j)

Probabilities of different signal behaviors such as no change (PNC), delta of signal

(PDS) and entire signal (PES) of the jth signal of the ith message sum up to 1. Hence,

1),(),(),( =++ jiPjiPjiP ESDSNC (9)

Table XI shows the length (in bits) of signal j of message i for various conditions

of the signals. Using probabilities, the estimated length of the signal can be represented

with a

73

[ ][ ] ⎟⎟

⎠

⎞⎜⎜⎝

⎛+

+++=

),(),(2),(),(1),(2

),,(jiPjiDL

jiPjiSLjiPjitSL

DS

ESNC (10)

Number of bits in the sum of the lengths must be a multiple of 8, whole bytes. We

define operator wb that returns the number of bits of the whole-byte datum. If

mod(a,8)=0, then wb{a}=a, else wb{a}=a+(8-mod(a,8)). Bus utilization (BU*) for the EDR

message traffic becomes

T

jitSLwbtsbOL

tBU

N

i

T

t

S

ji

i

Δ⎥⎥⎦

⎤

⎢⎢⎣

⎡++

=∑ ∑ ∑

=

Δ

= =1 0 1

),,()(*

)(* (11)

Where sbi*(t) is the length of stuff bits of the reduced message. Substituting the

value of SL(t,i,j) from (10) in (11) we get

[ ][ ]

T

jiPjiDLjiPjiSL

jiPwb

tsbOL

tBU

N

i

T

t

S

jDS

ES

NC

i

i

Δ⎥⎥⎥⎥⎥

⎦

⎤

⎢⎢⎢⎢⎢

⎣

⎡

⎟⎟⎟

⎠

⎞

⎜⎜⎜

⎝

⎛

+++

+

++

=

∑ ∑ ∑=

Δ

==

1 01 ),(),(2

),(),(1),(2

)(*

)(* (12)

As long as the EDR bus utilization, BU*, is lower than the bus utilization BU of

the original message traffic, it makes sense having EDR algorithm employed. This

means that, for the EDR algorithm to be useful BU*<BU. Performance improvement, PI,

is the ratio between BU* and BU. Therefore, using (6) and (12) we can write PI as

[ ][ ]

∑ ∑ ∑

∑ ∑

=

Δ

==

= =

⎥⎥⎥⎥⎥

⎦

⎤

⎢⎢⎢⎢⎢

⎣

⎡

⎟⎟⎟

⎠

⎞

⎜⎜⎜

⎝

⎛

+++

+

++⎥⎥⎦

⎤

⎢⎢⎣

⎡++

=

N

i

T

t

S

jDS

ES

NC

i

N

i

S

jii

i

i

jiPjiDLjiPjiSL

jiPwb

tsbOL

jiSLtsbOLn

PI

1 01

1 1

),(),(2),(),(1

),(2)(*

),()(

(13)

74

Let us calculate performance improvement of EDR vs. uncompressed bus

utilization for one message having one 64-bit long signal. From (13), performance

improvement is

⎟⎟⎠

⎞⎜⎜⎝

⎛+++

+++++

=

)},(]322[),(]641[),(2{)(*47

64)(47

jiPjiPjiPwbtsb

tsbPI

DSES

NCi

i

⎟⎟⎠

⎞⎜⎜⎝

⎛+

++++

=

)},(34),(65),(2{)(*47

)(111

jiPjiPjiPwbtsb

tsb

DSES

NCi

i

(14)

Let us consider three boundary examples.

1. For 1),( =jiPNC , 0),( =jiPDS and 0),( =jiPES performance improvement is

( ) ))(*55/()(111 tsbtsbPI ii ++= . Since sbi(t) has a maximum value of 19 and a minimum

value of 0, and sbi*(t) has a maximum value of 8 and a minimum value of 0, the

performance improvement is 36.276.1 ≤≤ PI . Thus, for this particular case, we can

send 1.76 to 2.36 times more traffic if we use EDR.

2. For 0),( =jiPNC , 1),( =jiPDS and 0),( =jiPES performance improvement is

( ) ))(*87/()(111 tsbtsbPI ii ++= where, in this case, sbi*(t) has a maximum value of 14 and

a minimum value of 0. For this case, the performance improvement is 49.11.1 ≤≤ PI .

Hence, for this case, we can send 1.1 to 1.49 times more traffic if we use EDR.

3. For 0),( =jiPNC , 0),( =jiPDS and 1),( =jiPES the EDR algorithm does not send the

reduced message that is equal or larger than the original. Therefore, PI=1.

The above examples show that the performance improvement ranges from 1.00 to 2.36,

for this case of a message consisting of one 64-bit signal, which is an ideal case for

compression performance. Fig. 14 and Fig. 15 give a more comprehensive look at the

range of performance improvement for this algorithm. Fig. 14 takes the message

75

example from above and varies the probabilities to show the entire range of

compression performance for this type of message. Fig. 15, on the other hand,

considers a message which has a composition poorly predisposed for compression.

This message consists of 12 signals of length 5 bits and 1 signal of length 4 bits. Fig. 15

also covers the entire domain of probability combinations. The achieved performance

envelope in Fig. 15 is noticeably shifted toward less achieved compression.

Figure 28: Performance Improvement vs. Signal Type Probability for a Message Containing a 64-Bit Long Signal. Note that ),(),(1),( jiPjiPjiP DSNCES −−= .

76

Figure 29: Performance Improvement vs. Signal Type Probability for a Message containing twelve 5-bit Signals and one 4-Bit Signal. Note that ),(),(1),( jiPjiPjiP DSNCES −−= .

These figures illustrate the strong dependence of the algorithm on the signal

dynamics and message structure, but also demonstrate the large compression potential

for the more prevalent steady-state conditions.

4.8. SIMULATION, TEST RESULTS AND DISCUSSION

4.8.1. TEST METHODOLOGY

We used a real-life vehicle CAN message log collected during vehicle testing, as

well as several logs consisting of synthetic message traffic created by manipulating

signal contents of the real-life message log. The performance of EDR, for the tests

based on the real-life message log, is measured by comparing bus utilization of the

original message log with the message log that is produced from original message log

when EDR is applied. The results using the real vehicle data will be discussed first,

77

followed by the results gathered using synthetic data.

4.8.2. TEST OF REAL-LIFE MESSAGE LOGS

Table VII shows the message distribution of a typical passenger vehicle. The

exact details of the messages are not given because of the proprietary concerns. The

bus baud rate is 250 kb/sec. The messages that have smaller periods and greater

lengths contribute the most to bus utilization. We have run the Network Analyzer

considering all messages (Test 1) and considering only the messages that contribute

the most to the bus utilization (Test 2). From Fig. 16, the bus utilization savings of the

EDR using all messages does not differ much from the bus utilization of EDR using the

most frequent messages. The average difference in bus utilization is 1.113 %. Thus,

targeting only frequent messages in the bus network will achieve good amount of BU

savings and reduce complexity and computing overhead in the nodes.

Performance Comparison of Different Data Reduction Algorithms: Fig. 17 shows

performance comparisons, in terms of bus utilization, of three different reduction

algorithms and uncompressed traffic for the real-life test. EDR bus utilization is the

lowest of the three. The average bus utilization difference between EDR algorithm and

the uncompressed traffic is 13.44 %. The average performance improvement, based on

Equation (15) is 1.462. EDR algorithm performs the best for signals of the log. ADR

algorithm performs poorly because of the lack of adaptability. DR algorithm performs

relatively well because of its low overhead and stable signal behavior in this example.

78

Table XII Configuration of the Messages Used in the Analysis.

(Messages marked with "*" are the biggest contributors to bus utilization.)

Msg. ID (hex)

ID (STD or EXT)

Data Length(byte)

Period(ms)

1* 128 STD 3 10 2* 180 STD 4 10 3* 110 STD 8 12 4* 120 STD 7 12 5* 124 STD 5 12 6* 150 STD 8 12 7* 151 STD 8 12 8 140 STD 8 20 9 144 STD 5 20 10 380 STD 5 20 11 388 STD 5 20 12 280 STD 8 32 13 300 STD 8 100 14 308 STD 7 100 15 320 STD 8 100 16 330 STD 3 100 17 410 STD 5 100 18 520 STD 3 100 19 420 STD 2 106 20 2F0 STD 7 106 21 130 STD 6 200 22 348 STD 4 250 23 170 STD 2 500 24 2D0 STD 1 500 25 510 STD 8 1000 26 670 STD 8 1000 27 674 STD 5 1000 28 38A STD 1 1500

79

25

27

29

31

33

35

37

0 5 10 15 20 25Time (sec)

Bus

util

izat

ion

of th

e ve

hicl

e ne

twor

k (%

)

EDR Test1 EDR Test2

Figure 30: Performance Comparison of EDR Considering All Messages (Test1) and the Most Frequent Messages (Test2).

80

Figure 31: Comparison of EDR, ADR, IADR, DR and Uncompressed Bus Utilization for the Real-Life Message Log.

We must note that this log represents only a snapshot of possible signal

behaviors and message compositions. In the next subsection of this dissertation we

examine DR, ADR, IADR and EDR performance based on synthetic signals.

4.8.3. TEST OF ARTIFICIALLY CREATED MESSAGE LOGS

This subsection of the dissertation compares the performance of EDR algorithm

for different signal behaviors. We also examine the performance of the EDR when bus

reaches saturation by increasing the number of simulated messages.

Diverse Signal Behavior Test: Here we used the message configuration from

Table VIII. The first column of Table VIII shows the ID of a message, the second column

81

shows the period of the message in ms, and the third column shows the lengths of

different signals in the message. For example, the ID of the first message in Table VIII

is 1; its period is 10 ms meaning that the message repeats every 10 ms; and there are

11 signals (8,8,8,8,6,6,6,6,6,1,1) in the message. Out of the 11 signals, four are 8-bit

long, five are 6-bit long and 2 are 1-bit long. The baud rate of the bus is assumed 250

kb/sec. We chose a mixture of signal sizes for a realistic network representation.

The left five columns of Table IX show different probabilities, PES, PDS, and

PNC, for the SDN signal type, and PES and PNC for the SN signal type used for the

test. The right four columns of the table show the resulting number of reduced

messages per second for all algorithms. It is desirable to have greater number of

reduced messages and lower overhead per message.

Table XIII Message Configuration

ID Period (ms)

Message content, signal sizes (bit)

1 10 8,8,8,8,6,6,6,6,6,1,1 2 10 8,8,8,6,6,6,6,6,6,1,1,1,1 3 10 8,8,8,8,8,6,6,6,1,1,1,1,1,1 4 10 16,8,8,6,6,6,6,1,1,1,1,1,1,1,1 5 20 16,8,8,8,8,8,6,1,1 6 20 16,8,8,8,6,6,6,2,1,1,1,1 7 20 16,8,8,8,8,6,6,1,1,1,1 8 20 8,8,8,8,8,8,6,2,1,1,1,1,1,1,1,1 9 20 8,8,8,8,8,8,6,6,1,1,1,1 10 20 16,8,8,6,6,6,6,6,2 11 50 16,8,8,6,6,6,6,6,1,1 12 50 16,8,8,8,6,6,6,1,1,1,1,1,1 13 50 8,8,8,8,8,6,6,6,6 14 50 8,8,8,8,8,6,6,6,6 15 50 8,8,8,8,8,6,6,1,1,1,1,1,1,1,1,1,1,1,1 16 50 16,8,8,8,6,6,6,6 17 100 8,8,8,8,6,6,6,6,2,1,1,1,1,1,1 18 100 16,8,8,8,6,6,6,1,1,1,1,1,1 19 100 16,8,8,8,6,6,6,6

82

20 100 8,8,8,8,8,6,6,6,2,1,1,1,1

Table XIV Number of reduced messages per second for different signal type probability.

SDN Type Signal SN Type Signal Number of reduced messages

PES PDS PNC PES PNC DR ADR IADR EDR 1 0 0 1 0 0 0 0 0 0.8 0.2 0 0.8 0 0 0 3 3 0.6 0.40 0 0.6 0 0 0 37 43 0.4 0.60 0 0.4 0 1 20 199 232 0.2 0.80 0 0.2 0 14 120 520 601 0 1 0 0 0 93 585 809 809 0.8 0 0.2 0.8 0.2 224 0 148 154 0.6 0.2 0.2 0.6 0.2 298 0 321 331 0.4 0.4 0.2 0.4 0.2 343 11 485 503 0.2 0.6 0.2 0.2 0.2 397 122 695 725 0 0.8 0.2 0 0.2 512 581 808 808 0.6 0 0.40 0.6 0.4 564 0 511 524 0.4 0.2 0.4 0.4 0.4 614 19 671 679 0.2 0.4 0.4 0.2 0.4 682 132 761 778 0 0.6 0.4 0 0.4 739 581 806 806 0.4 0 0.6 0.4 0.6 754 13 750 758 0.2 0.2 0.6 0.2 0.6 783 94 793 796 0 0.4 0.6 0 0.6 806 581 806 806 0.2 0 0.8 0.2 0.8 803 102 803 804 0 0.2 0.8 0 0.8 802 577 802 802 0 0 1 0 1 809 583 809 809

83

Figure 32: Number of Bytes Saved vs. Signal Type Probability for EDR, DR, IADR, and ADR Algorithms.

84

Figure 33: BU Comparison Between Uncompressed and EDR Traffic for Varying Number of Messages.

For all combinations of the probabilities except one, EDR shows more or the

same number of reduced messages as the other algorithms. For PES(SDN)=0.8,

PDS(SDN)=0, PNC(SDN)=0.2, PES(SN)=0.8 and PNC(SN)=0.2, DR has higher number of

reduced messages than EDR. For this particular case, EDR requires more overhead

bits than DR due to the adaptability nature of EDR. In this limited case, DR performs

better.

Figure 18 shows byte savings in percents vs. different signal type probabilities for

DR, ADR, IADR, and EDR algorithms. For the PNC=1, the highest no change

probability, the DR algorithm has the lowest overhead because all the reduced

85

messages are one data byte long. As the probabilities of the delta change, PDS, and

entire-signal, PES, increase, the benefits of EDR become more evident. Thus, EDR

saves more data bytes in dynamic environments because of its adaptability. In real-life,

signal behaviors are more dynamic than static. Hence, EDR will be the most appropriate

algorithm for practical applications.

Saturation Test: Even though no vehicle manufacturer would ever have bus

utilization near saturation, we wanted to see how EDR behaves when the bus is being

loaded and becoming saturated. We measured average BU starting with 12 messages

and gradually incremented the number of messages in the simulation by 5 until the total

of 42 messages. Table X shows the configuration of initial and added simulated

messages. Uncompressed message traffic reaches saturation with 32 messages, while

the message traffic compressed with EDR algorithm is getting saturated with 37

messages, as shown in Fig. 19. In other words, with EDR, 5 more messages can be

added to the bus before the bus becomes saturated compared to uncompressed traffic.

86

CHAPTER 5.

DISTRIBUTED TECHNIQUE FOR REMOTE PROGRAMMING

5.1. BOOTLOADER

The bootloader is a part of the embedded software of a typical in-vehicle

programmable (IVP) electronic control unit (ECU). The embedded software of the IVP

ECU consists of a bootloader (permanent) and operational (changeable) software. The

bootloader is the very first code executed on power up or reset. It usually consists of a

reduced serial bus handler, flash erase and write functions, and code verification

methods. The main bootloader functionality is to provide the ability for “changeable”

software to be reprogrammed. Typical embedded software is presented in Figure 34.

Figure 34: Typical Execution Flow of Embedded Software in ECU

5.2. DESCRIPTION OF THE CURRENT REPROGRAMMING PROCESS

The programmer, containing the software for the target ECU, is connected to the

vehicle bus. The programmer talks to the target ECU. First, the programmer informs the

ECU about reprogramming. The ECU prepares for reprogramming by erasing FLASH

block(s) of “changeable” memory. FLASH erasing may take a relatively long time. After

the ECU has erased FLASH block(s), the ECU informs the programmer that it is now

87

ready to accept data. The programmer sends the first packet. The ECU receives the

packet and stores it in the RAM buffer. The ECU responds with a message confirming

that the data has been received in the RAM buffer, but the ECU will need some time to

write data from the RAM buffer to the FLASH block. The programmer waits for another

message from the ECU that will say, "ECU wrote data into FLASH block and it is ready

to receive more data". The programmer sends the second packet, and the previously

described steps are repeated until the ECU is completely reprogrammed. A typical

reprogramming process that is currently used is shown in Figure 35.

88

START

Programmer

Request Download Erase FLASH block(s)

Respond saying that the ECU received data butneeds more time to write the data in the FLASH

Send the first packet of the software

Store received packet in the RAM buffer

Write packet from RAM buffer into FLASH block(s).

Respond saying that the ECU wrote the data in the FLASH and it is ready for next packet

Send the second packet of the software

Send the last packetof the software

END



Respond saying that the ECU received data butneeds more time to write the data in the FLASH

Write packet from RAM buffer into FLASH block(s).

Respond saying that the ECU wrote the data in the FLASH and it is completely programmed

Verify FLASH software image

Positive response

Wait

Wait

Targeted ECU

Time

Figure 35: Current typical reprogramming session.

5.3. PROPOSED IMPLEMENTATION

Erasing FLASH blocks and writing packets of data to FLASH memory takes a

long time. It takes comparatively less time to send packets on the serial bus and for the

ECU to receive packets by storing them in the RAM buffer than the time needed to

erase/write the FLASH memory. Available RAM memory of other ECUs on the vehicle

bus can be used as “cache” memory. The programmer will distribute the software to the

available RAM buffers of the ECUs in the vehicle. After distributing the software to the

89

RAM buffers, the programmer can be disconnected from the vehicle bus.

The programmer initiates the reprogramming session by asking the target ECU

(ECU 0) to prepare for reprogramming. ECU 0 starts erasing FLASH memory block(s)

of “changeable” memory. While ECU 0 is erasing FLASH block(s), the programmer

sends the first packet to the first available ECU (ECU 1) on the bus. ECU 1 stores the

received data in its RAM buffer and responds with a message explaining the status of

the availability of its RAM buffer. If there is more room available in ECU 1, the

programmer will send the second packet of the software to ECU 1. The programmer will

continue to fill the ECU 1 RAM buffers until it receives a message from ECU 1 saying

that ECU 1 RAM buffers are filled. The programmer will continue to fill available ECU

RAM buffers until the entire software for the target ECU has been distributed among the

available RAM buffers or until all available buffers are filled. In this dissertation, we will

consider only the case where the entire software for the target ECU is smaller than the

size of all available RAM buffers together. The case of the software being larger than

the total size of RAM buffers will be considered in future studies. After the programmer

has transmitted the entire software it will then inform ECU 0 or another designated unit

of the packets' whereabouts. The designated ECU accepts this packet whereabouts

information and then later controls the flow of packets. For now, we will assume that the

designated ECU is the target ECU (ECU 0). The process is described in Figure 36,

Figure 37, and Figure 38.

90

Programmer

Request Download Erase FLASH block(s)

Respond saying that the ECU received data and it is ready for next packet

Send the first packet of the software


Send the second packet of the software

Targeted ECU (ECU0)

ECU1



Send the next packet of the software

Respond saying that the ECU receiveddata and the RAM buffer is full


ECU2





Respond saying that the ECU receiveddata and the RAM buffer is full


ECU K

Send the last packet of the software



START

BA

Figure 36: Proposed Reprogramming Session part 1

91

Programmer

ECU 0 erased FLASH blocks

Targeted ECU (ECU0) ECU1 ECU2 ECU K

Sends the information about location of the packet.

Request softwarepackets

Send the first software packet


Respond saying that the ECU received data but needs more time to write the data in the FLASH

Send the last packet from ECU1 RAM buff.



Send the next software packet

Send the last packet from ECU2 RAM buff.




Write the packet in the FLASH

Respond saying that the ECU wrote the data in the FLASH and it is ready for next packet.

Wait


END

Wait

BA

C


92

Programmer Targeted ECU (ECU0) ECU1 ECU2 ECU K

Send the next software packet

Send the last packet from ECU K RAM buff.






Verify software image in FLASH

Respond saying ECU 0 completely reprogrammed and all buffers can be flushed.

END

END

END END

Flush RAM buffer

Flush RAM buffer Flush RAM buffer

C


5.4. FEASIBILITY STUDY

An example of one vehicle bus is shown in Figure 39. There are 16 ECUs in the

vehicle bus. Let us consider a scenario where all ECUs have the same microcontroller

(micro). Let us make several assumptions. The micro has 6 kilobytes of RAM. The size

of the “changeable” code is 64 K. The time needed to erase the FLASH block is 1.5

seconds. The time needed to write one byte to FLASH is 10 microseconds. The serial

bus speed is 125 kilobits per second. In this simple study we are not considering

message overhead, message-processing, latency, and response time and many other

things that add to the reprogramming time.

CURRENT REPROGRAMMING PROCESS – Ignoring the overhead, the time

93

needed to transfer a one-kilobyte packet to the RAM buffer is 65.5 milliseconds. The

time needed to write one kilobyte to the FLASH block is 10.24 milliseconds. The total

time the programmer has to be on the bus is the sum of the time to erase the FLASH

block, the time to transfer data to the RAM buffer and the time to write data to FLASH.

The total time is 1.5 (seconds) + 64 x (10.24 + 65.5) milliseconds = 6.347 seconds.

PROPOSED REPROGRAMMING PROCESS - Currently when the vehicle is

being reprogrammed, none of the ECUs in the vehicle is performing any operations. For

example, say a vehicle is at the dealer and the technician is reprogramming the target

ECU. Let us assume that 5 kilobytes can be used from each ECU for reprogramming

purposes and that 1 kilobyte will be used for storing ECU-critical data. The total time

during which the programmer needs to be on the bus is now only driven by the time

during which the programmer transfers software to the RAM buffers. The time for

erasing the FLASH block and writing data to FLASH is not a part of the total

programmer time. Therefore, the total time the programmer needs to be on the bus is

4.192 seconds. Table I shows several microcontrollers and the estimated programmer's

time savings in the simplified reprogramming model.

94

Figure 39: Programmer Connected to the Vehicle Serial Bus

95

Table XV -Programmer's time saved

Micro RAM [KB]

Time to

Erase 64 K

FLASH [sec]

Protocol speed

[Kb/sec]

Time to write 1 KB to FLASH

[ms]

Current implementation

programmer time [sec]

Proposed implementation

programmer time [sec]

Programmer time saved

[sec]

st9f150 6.00 1.5000 125 10.24 6.349664 4.194304 2.15536 st9f120 4.00 1.5000 125 10.24 6.349664 4.194304 2.15536 PIC18F6620 3.75 2.0000 125 96.00 12.338304 4.194304 8.144 PIC18F8620 3.75 2.0000 125 96.00 12.338304 4.194304 8.144 LPC2119 16.00 0.4000 125 2.00 4.722304 4.194304 0.528 LPC2129 16.00 0.4000 125 2.00 4.722304 4.194304 0.528 68HC912D60A 8.00 0.0180 125 1.00 4.276304 4.194304 0.082

96

CHAPTER 6.

EXPERIMENTAL CHARACTERIZATION OF THE DEDICATED SHORT RANGE

COMMUNICATION

6.1. INTERFERENCE MODEL BASED ON TWO-RAY GROUND REFLECTION

The communication scenario considered here is two vehicles separated by a

certain distance and approaching each other at a certain speed in an open area without

obstacles (apart from the reflective ground). The simplest way to model the signal

strength propagation over distance would be to use the free space propagation model

(15)

2

4)( ⎟

⎠⎞

⎜⎝⎛=

dLGGPdP rtt

r πλ (15)

where Pr(d) is the received power of the separation distance d (m), Pt is the transmitted

power, Gt is the transmitter antenna gain, Gr is the receiver antenna gain, λ is the

wavelength (m), and L is the system loss factor (≥ 1) accounting for losses other than

propagation [70]. This equation applies when there is an unobstructed line-of-sight

between the transmitter and receiver. However, this model does not take into account

the effect of ground reflections. To account for the interference from the ground

reflections, the two-ray ground reflection model will be used here. Table XVI explains

the notation used.

Table XVI - Notation

Symbol Description Unit d Distance between the TX and

RX vehicle [m]

dLOS Distance traveled by the line-of- [m]

97

sight wave component dNLOS Distance traveled by the non-

line-of-sight wave component [m]

Θi Angle of incidence [rad] Θk Critical angle of reflection [rad] ht Height of transmitter antenna [m] hr Height of receiver antenna [m] λi Wavelength of ith sub-carrier [m] λMF Wavelength of mid-frequency

carrier

fi Frequency of ith sub-carrier [Hz] Ti Period (1/ fc) of ith sub-carrier [s] ΔT Reflected wave delay [s] c Speed of light [m/s] ωi Angular frequency of a sub-

carrier i [rad/sec]

i Sub-carrier index Ei Amplitude of the electric field

due to LOS wave component of ith sub-carrier

E i` Amplitude of the electric field due to reflected wave component of ith sub-carrier

┴ Denotes horizontal polarization (normal to the plane of incidence)

|| Denotes vertical polarization (in the plane of incidence)

sLOS Line-of-sight signal sNLOS Non-line-of-sight (reflected)

signal

k Number of wavelengths that reflected wave traveled longer than LOS wave.

dF Fresnel zone distance [m] Г|| Reflection coefficient for

horizontal polarization

Г┴ Reflection coefficient for vertical polarization

εr Relative permittivity The two-ray propagation model for the open road test scenario considered here,

where two vehicles are communicating while approaching each other, is illustrated in

Figure 40.

98

LOS component

D

ht hr

TX RXΘi

Reflected componentΘr

a)

dLOS

d

ht

hr

TX

RXdNLOS

ht

b)

ht-hr

ht+hr

Figure 40: Illustration of: a) two-ray propagation between the two vehicle and b) method of images to find the path difference between the LOS and NLOS component.

In general, a wave traveling through free space and incident on a surface will be

partially reflected from the surface and partially absorbed. The proportion depends on

the conductive properties of the surface material. In case of a perfect conductor, all of

the energy will be reflected and none absorbed. When the material is a dielectric, there

is both reflection and absorption. The proportion between the two behaviors will depend

on the wave polarization, the angle of incidence, and the surface material. For a wave

traveling through free space and incident on a surface with relative permittivity εr, the

reflection coefficients for vertical and horizontal polarizations are given by the following

expressions, where the angle argument was modified from [70] to suit here-used

definition of θi in Figure 40.

irir

irir

i

r

EE

θεθε

θεθε2

2

||sincos

sincos

−+

−+−==Γ (16)

99

iri

iri

i

r

EE

θεθ

θεθ2

2

sincos

sincos

−+

−−=⊥=Γ (17)

Substituting θi = 90° in (16) and (17) gives |Γ||| = 1 and Γ┴ = -1. This shows that

for grazing incidence (θi approaches 90°) the reflection behavior approaches that of

perfect reflection (no absorption) regardless of wave polarization or electric properties of

ground. Based on this, ground is modeled as a perfect reflector in the development that

follows.

Figure 41: Absolute real value of reflection coefficient for vertical and horizontal polarization for εr = 25. Brewster angle is 78.50.

0 10 20 30 40 50 60 70 80 900

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Practical angles→

vertical polarizationhorizontal polarization

Incidence Angle

Ref

lect

ion

Coe

ffici

ent

100

Figure 41 shows absolute values of reflection coefficients of horizontally and

vertically polarized wave for a typical relative permittivity εr = 6 of an asphalt surface

[76]. However, Rustako et al in [77] and Supanakoon et al in [78] for their calculation

use εr = 15 and εr = 25 as relative permittivity of the ground.

||┴

┴

||┴

ΘΘ

Brewster Angle

Incident wave

Reflected Wave900

Transmitted wave

Figure 42: Polarization by reflection

Reflection coefficient for vertical polarization for near grazing angle is much

smaller than for horizontal polarization. Therefore the effect of the vertically polarized

ground reflected wave will be much less for the angles of interest. For the Brewster

angle there will be no vertically polarized waves reflected as illustrated in Figure 42.

101

Even though antenna specification indicates the vertical polarization actual wave has

some horizontal component as well. This is because of the imperfections of the antenna

and the ground plane which is in this case vehicle roof.

In order to characterize the angle of incidence, θi, consider the omni-directional

antenna used in preformed experiments, and in similar applications at its mounting

position on the roof of the test vehicle. The antenna is designed to radiate energy in

360° of the horizontal plane within approximately 0 – 30° of elevation. However, most

waves that travel downward from the transmitter – all except a few degrees down from

the horizontal - will be blocked by the vehicle body, as illustrated in Figure 43. The

receiver vehicle’s body will similarly block incoming waves. Therefore, the line of sight

(LOS) waves traveling almost parallel to the ground (depending on antenna heights),

and the waves within a few downward degrees of this line which graze the ground and

reflect to the receiver from the ground, will be the only expected sources of the received

energy in the open environment considered here. The expected small tilt from the

horizontal for the grazing waves motivates the assumption of θi being close to 90° in

further derivations.

102

Figure 43: Illustration of energy propagation from the antenna mounted on vehicle’s roof.

According to laws of reflection in dielectrics the incident and reflected angles are

the same [70].

ri θθ = (18)

The distance an LOS wave travels between the transmitter and receiver (19) can

be derived from Figure 40b.

( ) 22 dhhd rtLOS +−= (19)

The distance traveled by a non line-of-sight (NLOS) wave reflected from the

ground (20) can be derived using the method of images in Figure 40c.

( ) 22 dhhd trNLOS ++= (20)

The receiver will get the reflected (NLOS) wave with a certain delay ΔT

compared to the LOS wave. This delay will depend on the difference in distances

traveled by the two waves and the speed of light (21).

103

cddT LOSNLOS −

=Δ

( ) ( )c

dhhdhh rttr2222 +−−++

= (21)

The received LOS wave has the following form

( ) ∑−=

=c

ci

tjiLOS

ieEts )(ω (22)

where the summation is over index c spanning all sub-carriers in a DSRC

channel with i ∈ [-26, 26]. The sub-carrier i angular frequency is related to the sub-

carrier frequency fi by

ii fπω 2= (23)

and the sub-carrier i wavelength is

ii f

c=λ (24)

When a horizontally polarized wave reflects from the hard road surface, it

undergoes a π change in phase. For the vertical polarization there is no change in

phase [70]. Therefore, the reflected horizontally polarized wave at the receiver looks like

(25)

( ) ∑−=

+Δ−−⊥ =26

26

))((`

i

Tc

dtj

iNLOS

LOSieEts

πω

∑−=

Δ−−−=

26

26

))((`

i

Tc

dtj

i

LOSieE

ω (25)

and the vertically polarized wave looks like (26).

104

( ) ∑−=

Δ−−=

26

26

)(`||

i

Tc

dtj

iNLOS

LOSieEts

ω (26)

The signal at the receiver, s(t), will be the sum of the LOS and NLOS waves.

Equations (27) and (28) are the expressions for horizontally and vertically polarized

waves, respectively.

( ) ( ) ( )tststs NLOSLOS +=

( ) ∑∑−=

Δ−−

−=

−

⊥ −=26

26

))((`26

26

))((

i

Tc

dtj

ii

cdtj

i

LOSi

LOSi eEeEts

ωω (27)

( ) ∑∑−=

Δ−−

−=

−+=

26

26

))((`26

26

))((

||i

Tc

dtj

ii

cd

tj

i

LOSi

LOSi eEeEts

ωω (28)

From the horizontal polarization expression (27), which undergoes a phase

change of π radians, it can be seen that the NLOS interference will be the most

destructive, resulting in the received strength null points, when (13) is at minimum,

which occurs at

,...]2,1,0[,2 ∈=Δ kkT i πω (29)

giving

ifkT =Δ (30)

This can be related to the geometry using (21):

( ) ( )

( ) ( )c

dhhdhhck

cdhhdhh

fk

rttr

rttr

i

2222

2222

+−−++=

⎟⎠⎞

⎜⎝⎛

+−−++=

λ

105

( ) ( ) 2222 dhhdhhk rttr +−−++=λ (31)

To be able to conveniently iterate k starting with zero for both polarizations

( ,...]2,1,0[∈k ), this is rewritten as

( ) ( ) 2222 dhhdhhk rttr +−−++=+ λλ (32)

Similarly, for vertical polarization from (28), where the ground reflected wave

stays in phase with the LOS wave, the null points are expected at

,...]2,1,0[,2 ∈+=Δ kkT i ππω (33)

and this can be related to the geometry using

( ) ( ) 2222

2dhhdhhk rttr +−−++=+ λλ (34)

Equations (32) and (34) can be solved for dk, which represents the distance

between vehicles at which the reflected wave will be most destructive and the reception

null points for the horizontal (35) and vertical polarization (36) are expected.

( ) ( )222

)(24

rtii

iitrk hh

kkhhd −−⎟⎟

⎠

⎞⎜⎜⎝

⎛+

+−=⊥

λλλλ (35)

( )2

22

||

)2

(2

24

rt

ii

ii

tr

k hhk

khhd −−

⎟⎟⎟⎟⎟

⎠

⎞

⎜⎜⎜⎜⎜

⎝

⎛

+

⎟⎠⎞

⎜⎝⎛ +−

=λλ

λλ

(36)

For k = 0, rthh4<<λ and rt hh ≈ equation (36) becomes that of the distance where

the first Fresnel zone touches ground, dF.

Fi

tr dhhd ==λ

4||0 (37)

We define the critical wave as a ground reflected wave that arrives at the

106

receiver exactly out of phase with the LOS wave. The first critical wave occurs at k = 0,

the second at k = 1 and so on. Solving the critical distance dk for different values of k in

equations (35) and (36) will produce a multiplicity of null points. However, consider the

angle of incidence of critical waves. It can be determined from the geometry of Figure

40 and is given by (38).

k

trk d

hh +−=Θ arctan

2π (38)

This angle is calculated for the experimental antenna heights hr = ht = 1.45 m and

a range of ⊥kd values in Table XVII. Recalling the earlier discussion based on the

vehicle body obstruction which limits the incident angles to those close to 90° implies

that only k = 0 is of practical significance. This means that for the prediction of the null

points only k = 0 needs to be considered.

Table XVII Predicted null points, incident angles in degrees and reflection coefficients for vertical and

horizontal polarization for hr = ht = 1.45 m

horizontal polarization vertical polarization k ⊥

kd Θk Γ┴ ||kd Θk Γ┴

0 82.16 87.98 -0.986 164.38 88.99 0.835 1 41.05 85.96 -0.972 54.76 86.97 0.575 2 27.32 83.94 -0.958 32.81 84.95 0.380 3 20.45 81.93 -0.944 23.39 82.93 0.229 4 16.31 79.92 -0.931 18.15 80.92 0.108

There can also be a multiplicity of null points due to a range of sub-carrier

frequencies. However, since sub-carriers have the frequency spacing of 156.25 kHz,

the differences in wavelengths between sub-carriers are small (a fraction of a

millimeter). Therefore we will only consider the mid-frequency (5.860 GHz) wavelength

107

for the channel 178, which is λMF =0.0511591 m.

Combing those simplifications yields the predicated null points for k = 0 with

antenna heights hr = ht = 1.45 m at ⊥kd = 82.2 m and ||

kd = 164.4 m. The experimental work

presented later in this dissertation shows that the horizontal polarization critical distance

of 82.2 m better matches the experimental data in spite of the fact that the antennas

used have vertical polarization. The reason for this is not know at this time. This

correspondence also carries over to the analysis of previously published experimental

data. Thus, the method developed here suggests using the ⊥kd distance, as derived

above, for predicting the null points in open road scenarios where two vehicles are

approaching each other and using vertically polarized antennas.

6.2. DIFFRACTION

In a general case, in addition to the effects of reflection, the received signal

strength will depend on the effects of diffraction and scattering. Diffraction is present but

not expected to affect null points calculation in significant manner. In this

communication scenario the LOS path is clear of obstacles.

6.3. SCATTERING AND SURFACE ROUGHNESS CONSIDERATION

The possibility of scattering is considered here. Scattering depends on surface

roughness. Landron, Feuerstein, and Rappaport define the commonly used Rayleigh

criterion as a test for surface roughness in [69]. The threshold height hc is given by

ich

θλ

cos8= (39)

where θi is incident angle as shown in Figure 44. A surface is considered to be

rough when its minimum to maximum protuberance height, h, is greater than hc, and

108

smooth otherwise. For the angles θi close to 90°, which are the angles of most interest

to us, the threshold height becomes very large due the cosine of θi approaching

zero. We can assume that the road undulations do not exceed such large values and

that therefore the road is smooth. Scattering can be ignored for smooth surfaces and

thus will not be considered further here.

Figure 44: Surface roughness protuberances

6.4. EXPERIMENTAL METHOD

6.4.1. EXPERIMENTAL VEHICLES

We converted two ordinary vehicles into experimental platforms by equipping

them with additional hardware shown in Figure 45 and Figure 46. As figures show, each

vehicles is equipped with a Wireless Safety Unit (WSU) provided by Denso with

integrated 802.11p-based DSRC radio (Figure 45), an omni-directional roof-mounted

Nippon DSRC antenna (Figure 46), a Novatel Global Positioning System (GPS) receiver

antenna (Figure 46), a GPS receiver (Figure 45), and a software application running on

the WSU processor sending BSM messages in wireless short message (WSM) type at

the period of 100 ms. We have slightly enhanced the content of the BSM to include a

sequence number. Messages are sent at 20 dBm with a ~4 dB cable loss yielding ~16

109

dBm transmit power. The data rate is 6 Mbps. The antenna height for both vehicles is

1.45 m. The antennas are reported as vertically polarized by the manufacturer [74].

Table XVIII lists the specifics of the DSRC and GPS antennas and receivers. Table I

shows the most relevant DSRC physical layer parameters.

Figure 45: Hardware added to trunk to convert original vehicle into experimental vehicle.

GPS Antenna

DSRC Antenna

110

Figure 46: DSRC and GPS antennas mounted on vehicle roof.

Table XVIII: DSRC and GPS Component Specifics

Component Manufacturer or Part

Additional Information

DSRC receiver

Atheros 802.11 WiFi chipset

DSRC antenna

Nippon 802.11a/DSRC DEN-HA001

average gain at horizon: 0 dBi

GPS receiver Novatel OEMV absolute accuracy 1-3 m GPS antenna Novatel OEMV

The experimental additions are interfaced as shown in Figure 47. Most of the

available WSU interfaces are used: 1) a serial port is used to extract raw positioning

data from the GPS receiver, 2) an SMA connector interfaces to the Nippon DSRC

antenna, 3) a CAN port provides dynamic vehicle information from the vehicle network,

4) another serial port, as well as 5) the Ethernet port are used to interface the WSU with

the test laptop (not shown in Figure 47), and 6) a memory flash card is used for storing

log files.

Figure 47: Logical blocks of vehicle equipment.

111

6.4.2. EXPERIMENTAL SETTING

The experiments were conducted on October 1 and October 2, 2008 in the metro

Detroit area (Figure 48). The road is considered an open road due to clear lines of sight

and road edges mostly clear of potential multipath sources. The speed limit of the

arterial road allowed testing at 55 mph.

Figure 48: Location of the open area test. Open road with 55 mph speed limit

6.4.3. EXPERIMENTAL PROCEDURE

Each test pass consisted of droving the two experimental vehicles on the open

road toward each other in a straight line (Figure 49). We refer to the subject vehicle, in

this case also the receiving vehicle, as the host vehicle (HV) and other participating

vehicles (only one in this case) as remote vehicles (RM). The speed of both vehicles

was 55 mph. Vehicles were sending out BSM messages every 100 ms. Vehicles were

completely entering and completely exiting the communication range for each test run.

Specifically, the turnaround points were at a vehicle distance of 2000 m where no

communication between the vehicles was observed. The testing included ten vehicle

passes.

112

Figure 49: Visualization of two vehicles approaching each other in the opposite direction.

6.4.4. DATA LOGGING AND PROCESSING

The BSMs exchanged between the two test vehicles contain the critical

information about vehicle position, speed, acceleration, heading, along with the

message time stamp. BSMs transmitted by the HV and received by the HV from the RV

were logged. Additionally, the Receive Signal Strength Indicator (RSSI) computed by

the WSU as the average power of every received packet was also logged. The RSSI of

the transmitted packet is assumed to be constant at 16 dBm (20 dBm minus the

expected cable loss) and thus was not logged. The logged data was stored to the flash

memory card of the WSU. In post processing the vehicle separation distance is

calculated based on GPS positions of the test vehicles using the Haversine formula

[63][64]. When using the formula, the Earth radius was optimized for locations at

approximately 42 degrees from the equator.

6.5. EXPERIMENTAL RESULTS

The data collected using the experimental method just described is presented in

Figure 50. The data exhibits a reception null point at about 90 m. There is also a

reception floor line at about -95 dB that is caused by the sensitivity limit of the

experimental hardware. When the two vehicles are less than ⊥0d away from each other,

113

receiver power varies dramatically. It is expected that these rapid changes in the power

levels are due to the sum of the line of site wave, different multipath waves, and

diffraction waves of the surrounding environment. However, for the vehicle distances

beyond ⊥0d , the power at the receiver is mainly the sum of the ground-reflection (asphalt-

reflection) wave and the line of site wave. At these distances further than ⊥0d , the

received data follows the inverse power law. The received data can be approximately

modeled using the inverse power law in two curves: the first curve is proportional to 1/d4

and the second curve is proportional to 1/d2.85.

0 50 100 150 200 250 300 350 400-100

-90

-80

-70

-60

-50

-40

-30

Distance [m]

RS

SI [

dBm

]

RSSI AverageRSSI Rawn1=4

n2=2.85

2

1~ nd

1

1~ nd

⊥0d

Figure 50: RSSI versus separation distance for two vehicles approaching each other on open road.

6.6. NULL-POINT PREDICTION APPLICATION TO EXPERIMENTAL DATA

114

6.6.1. ANALYSIS AUTHOR’S EXPERIMENTAL DATA

In the experimental method, both the transmitting and receiving antennas are

vertically polarized. According to the previously presented method for the vertical

polarization, the null point is expected at ⊥kd calculated using (35). Substituting the

parameters of this experiment (hr = ht = 1.45 m, λMF =0.0511591 m) in (35) yields the

value of 82.2 m. As shown in Figure 50, the critical distance ⊥0d of the first null point (k =

0) was experimentally determined to be around 90 m. This experimental value appears

to be very close to the predicted value of 82.2 m. The difference between the predicted

critical distance and the observed critical distance is relatively small considering the

potential sources of inaccuracy: the difference in the elevation of the road (land

contour), the GPS error, the unsynchronized GPS locations of two test vehicles due to

communication and processing timing (BSM updates RV location every 0.1 s and in the

worst case can delay sending of the GPS reading by 0.1 resulting in a source of

inaccuracy between two vehicles that is in the worst case as large as 2 times 0.1 s at 55

mph, or 4.8 m). The observed difference from the prediction (7.8 m) is within the order

of magnitude of the last described source of error (4.8 m). More importantly, we next

apply the null point prediction method to experimental data in other literature, where this

approach provides better correspondence with experimental data then the reported

explanations.

6.6.2. ANALYSIS OF VII-C OPEN AREA TEST

To verify the approach in predicting null points in the DSRC based systems we

have looked into a related communication test performed by the VII consortium [68]. It

measured the RSSI at the vehicle receiver, or on-board equipment (OBE), at various

115

distances from the transmitter, (road side equipment, RSE) sending DSRC messages

on the control channel (Channel 178) (Figure 51). The RSE height was reported to be 5

m, however, the RSE transmitting antenna is pointed downward and the effective height

may have been lower. Test was performed with low and high profile receiving vehicles.

The OBE receiver antenna height for the low profile vehicle was 58 inches (1.47 m) and

for the high profile 70 inches (1.78 m). The mid frequency wavelength for the control

channel was λMF =0.0508985 m. That test was preformed at an open area on the

deserted Dan Derby airport in Ohio [68].

RSE

ht

dk

OBEht

Figure 51: VIIC test setup

Table XIX: Null point predictions for low (hr =1.47 m) and high (hr= 1.78 m) profile vehicles and transmitter

antenna heights ht=5 m.

⊥kd [m] ||

kd [m] k hr=1.47 hr= 1.78 hr=1.47 hr= 1.78 0 287.5 348.2 575.1 696.4 1 143.7 174.0 191.7 232.1 2 95.7 116.0 114.9 139.2 3 71.7 86.9 82.0 99.4

Table XIX shows the calculated critical distances for the transmitter antenna

height of 5 m and two receiver heights, 1.47 and 1.78 m, using both horizontal and

116

vertical polarization derivations. Figure 52 shows the typical signal strength vs. distance

behavior reproduced from [68] with critical distance marking superimposed for this

dissertation. In agreement with the experimental data, the null points predicted in the

Table XIX using horizontal polarization derivation (35) indicate a good fit for the null

points observed experimentally.

||0d

||2d ||

2d

⊥0d

⊥1d⊥

2d

Figure 52: VII Project: Relative average RSSI vs distance between transmitter and receiver. Predicted points d0, d1, and d2 are for the ht = 5 m and hr = 1.47 m. Reproduced from [68] with critical distance marking superimposed for this dissertation. (Note that RSSI in this figure is a

relative value not expressed in dBm.)

6.6.3. ANALYSIS OF VSC OPEN AREA TEST

The VSC project conducted similar testing in an on open area at Milford proving

grounds. The Appendix G of the “Vehicle Safety Communications Project Final Report”

[75] presents results of the RSSI measurements vs. distance between an RSE and an

OBE. The RSE antenna height was 3.04 m and the OBE antenna height was 1.45 m.

The mid-frequency carrier was 5.8 GHz. Figure 53 is reproduced form the report but the

117

marking of null points ⊥0d and ⊥

1d , as calculated using (35), is added for purposes of this

analysis. The figure shows that the predicted null points match experimentally gathered

data.

⊥1d⊥

0d ⊥1d ⊥

0d

Figure 53: VSC Project: Relative average RSSI vs distance between transmitter and receiver. Predicted points d0 and d1, are for the ht = 3.04 m and hr = 1.45 m. Reproduced from [75] with

null points marking added. (Note that RSSI in this figure is a relative value not a dBm.)

6.6.4. ANALYSIS OF BREAKPOINTS FROM MASUI, KOBAYASHI AND

AKAIKE

Masui, Kobayashi and Akaike experimentally characterized the radio

communication channel at the frequencies of 3.35, 8.45, and 15.75 GHz in an urban

LOS environment ([65], [66]). Their environment should not be considered an open area

due to streets being lined by 10 to 15 story high buildings. However, the LOS and

ground reflections should still expected to be primary contributors to signal strength.

They modeled they experimental data using two-slope log-linear lines. The authors

were initially predicting the breakpoint between the slopes using the distance where first

Fresnel zone touches the ground. This yielded values that did not correspond well with

118

their experimental data (Table XX). They attributed the differences to reflections from

objects in the street, such as vehicles and traffic, and stipulated that the presence of

these objects gives an effective increase in the height of the ground, or correspondingly,

an effective decrease in the height of antennas. They use this concept of effective road

height to match the experimental data to the Fresnel zone values, thus making the

effective road height both an indicator of discrepancy between their experimental data

and the Fresnel zone theory and also a contributive result of the experiment that

characterizes the obstacles in the test environment. Table XX shows that the approach

of Equation (35) for horizontal polarization gives a much better correspondence to

experimental data from [66] than obtained by the author using the Fresnel zone

approach.

Table XX Masui Data Breakpoints

Experimental Parameters Breakpoint Distance (m) ht hr f λ Reported Eq. (35) (m) (m) (GHz) (m) Experimental Theoretical Theoretical4 2.7 3.35 0.089490 170 480 241 8 2.7 3.35 0.089490 320 970 483 4 2.7 8.45 0.035478 300 1220 609 8 2.7 8.45 0.035478 760 2430 1220 4 2.7 15.75 0.019034 680 2270 1130

6.6.5. ANALYSIS OF CRITICAL DISTANCE FROM CHENG, HENTY, STANCIL,

BAI, AND MUDALIGE

To explain drops in the RSSI values at the vehicle separation distance of around

100 m, Cheng, Henty, Stancil, Bai, and Mudalige in [67] used the first Fresnel zone of

the vertically polarized rays. Cheng et al found big discrepancies between the

119

theoretical expectations based on the Fresnel zone and the experimental results and

accredited those discrepancies to the multipath from other vehicles, people, and

buildings. They noticed drops in RSSI values at the separation distance of around 100

m. They used two vehicles, one a transmitter and the other receiver with ht = 1.51 m hr

= 1.93 m, for a continuous wave signal at 5.9 GHz (wavelength λ = 0.050812 m).

Equation (35), for k = 0, produces the critical distance of ⊥0D = 113 m which is much

closer to the observed critical distance then using the distance where the first Fresnel

zone (dF) touches the ground at dF = 225 m as [67] suggested.

120

CHAPTER 7.

CONCLUSION

This work proposes an efficient algorithm for remote software update in intelligent

vehicles. To verify the algorithm effectiveness, we developed computer program that

simulates 100,000 vehicles traveling on the map of greater Michigan area and being

served by 1323 towers. We ran the algorithm in two different scenarios and compared

results. We concluded that for different conditions the algorithm can be tuned for the

best performance. Using simulation, we found out that for Scenario A (maximum speed

of vehicles 50 m/s and unicast cost is 500 μs), although the best performance is

achieved using the smallest retransmit buffer, the choice of retransmit buffer does not

affect performance significantly. For Scenario B (maximum speed of vehicles 17 m/s

and unicast cost is 500000 μs), performance significantly improves with the increase of

the retransmit buffer size.

This work also presented an Enhanced Data Reduction (EDR) algorithm for

reducing bus utilization in future vehicles. The proposed scheme does not require any

changes to the existing wired infrastructure. We showed, using simulation results, that

the EDR can be beneficially used. In the simulation, we considered realistic examples of

message traffic. We have proposed an improvement to detection of data reduction

usage by looking at message length. Also, we proposed a method to manage signals of

shorter length (<5 bits) by combining them into groups that are handled as single

signals. A solution is provided for handling the worst-case conditions for compression,

which arise during network initialization. The cost of the algorithm in terms of memory

usage is under 300 bytes, which is very insignificant compared to the amount of ROM

121

and RAM available in today’s microcontrollers. The impact of the algorithm on the end-

to-end message latency is also very insignificant. The algorithm can also be used for

other networking protocols as long as some time slots are available which are not

reserved for any particular messages.

This dissertation work also studied the characteristics of communication media

for V2V and V2I communications. Transmitted waves are bounced off various objects.

The most destructive multipath waves are the ones that travel exactly multiple

wavelengths of signal carrier. In the open area test case, those waves are found to be

the waves reflected off the traveled road with a small angle from the ground plane.

Destructive waves with larger angles will be blocked by the vehicle body and will not

reach the receiver antenna. We derived a formula for predicting DSRC null locations,

with respect to the vehicle separation distance, from the two-ray model. We used

vertically polarized antennas. We found big discrepancies between expected and

experimentally collected Fresnel break-point distances. However we found that break-

point distances calculation better matches distance calculation formula for horizontal

polarization. Predicted vehicle distance values for the nulls are good fit for

experimentally gathered data as well as the data from several previous researches. In

addition, the first prediction null point better fits the breakpoint for the piecewise channel

models than the break point described with the distance the first Fresnel zone touches

ground. Null prediction formula depends on the antenna heights. Therefore this

dissertation work recommends extension of the current SAE J2735 protocol to include

antenna height information in the BSM.

122

Acronyms

AC: Automotive Company

ACK: Acknowledgement

ADR: Adaptive Data Reduction

BCM: Body Control Module

BSM: Basic Safety Message

BU: Bus Utilization

CAN: Control Area Network

CCH: Control Channel

CCP: CAN Calibration Protocol

CD: Compact Disk

CPU: Central Processing Unit

CRC: Cyclic Redundancy Check

CS: Central Server

DAB: Digital Audio Broadcasting

DCB: Data Compression Bit

DCC: Data Compression Code

DR: Data Reduction

DSRC: Dedicated Short Range Communication

DVB: Digital Video Broadcasting

DVD: Digital Versatile Disc

ECM: Engine Control Module

ECU: Electronic Control Module

123

EDGE: Enhanced Data rates for GSM Evolution

EDR: Enhanced Data Reduction

EXT: Extended

FCC: Federal Communications Commission

FEC: Forward Error Correction

FOTA: Firmware Over-the-Air

GERAN: GSM EDGE Radio Access Network

GPS: Global Positioning System

GSM: Global System for Mobile communications

HV: Host Vehicle

HVAC: Heating, Ventilating, and Air Conditioning

IEEE: Institute of Electrical and Electronics Engineers

IVP: In-Vehicle Programmable

ITS: Intelligent Transportation System

LIN: Local Interconnect Network

LOS: Line of Sight

MT: Multicasting Tower

NLOS: Non Line of Sight

OBE: On Board Equipment

OEM: Original Equipment Manufacturer

OFDM: Orthogonal Frequency-Division Multiplexing

PCM: Powertrain Control Module

PDA: Personal Digital Assistants

124

PSD: Power Sliding Door

RAM: Random Access Memory

RAN: Radio Access Network

RM: Regional Manager

ROM: Read Only Memory

RPM: Rotation Per Minute

RSE: Road Side Equipment

RSSI: Received Signal Strength Indicator

RT: Reduction Type

RV: Remote Vehicle

RX: Receive

SAE: Society of Automotive Engineers

SCH: Service Channel

SDN: Signal, Delta, No-change

SN: Signal, No-change

STD: Standard

TCU: Telematics Control Unit

TX: Transmit

USDOT: United States Department of Transportation

UT: Unicasting Tower

VII: Vehicle Infrastructure Integration

VSC-A: Vehicle Safety Communications – Applications

WSM: Wireless Short Message

125

WSU: Wireless Safety Unit

126

References

[1] R.K. Panta, I. Khalil, and S. Bagchi, "Stream: low overhead wireless reprogramming

for sensor networks," 26th IEEE International Conference on Computer

Communications, Anchorage, Alaska, May 6-12, 2007, pp. 928 - 936.

[2] M. Rossi, G. Zanca, L. Stabellini, and R. Crepaldi, A.F. Harris, and M. Zorzi,

"SYNAPSE: A network reprogramming protocol for wireless sensor networks using

fountain codes," 5th Annual IEEE Communications Society Conference on Sensor,

Mesh and Ad Hoc Communications and Networks, June 16-20, 2008, pp. 188 – 196.

[3] J. Jeong, D. Culler, "Incremental network programming for wireless sensors," 1st

Annual IEEE Communications Society Conference on Sensor and Ad Hoc

Communications and Networks, Santa Clara, CA, Oct. 4-7, 2004, pp. 25 – 33.

[4] N. Reijers and K. Langendoen, "Efficient code distribution in wireless sensor

networks," in Proceedings of the 2nd ACM international conference on Wireless

sensor networks and applications, San Diego, CA, 2003, pp. 60-67.

[5] W. Bromley, B.R. Carl, S. Chang, B. Crull, A. Ditchfield, D. Essenmacher, and M.

Kapolka, “System, method and computer program product for remote vehicle

diagnostics, monitoring, configuring and reprogramming,” U.S. Patent No. 7155321,

Dec. 26, 2006.

[6] R.M. Fosler, "A CAN bootloader for PIC18F CAN microcontrollers," Appl. Note 247,

available at ww1.microchip.com/downloads/en/AppNotes/00247a.pdf

[7] http://www.vector.com/vi_flashbootloader_en.html

127

[8] J. J. Meyers "Channel characterization and reliability of 5.8 GHz DSRC wireless

communication links in vehicular ad hoc networks in suburban driving environment,"

M.S. thesis, Carnegie Mellon University, Pittsburgh, PA, 2005.

[9] G.S. Ching, M. Ghoraishi, N. Lertsirisopon, J.-i. Takada, I. Sameda, R. Soma, H.

Sakamoto, T. Imai, "Tunnel propagation channel characterization for DSRC

applications," in Proceedings of 2007 Asia-Pacific Microwave Conference, IEEE,

Bangkok, Dec. 11-14,.2007, pp. 1-4.

[10] R. Keller, T. Lohmar, R. Tönjes, and J. Thielecke, “Convergence of cellular and

broadcast networks from a multi-radio perspective,” IEEE Personal Communications

Magazine, vol. 8(2), pp. 51-56, 2001.

[11] T. Munaka, T. Yamamoto, and T. Watanabe, “A reliable advanced-join system for

data multicasting in ITS networks,” IEEE Trans. Intell. Transp. Syst., vol. 6, no. 4, pp.

424–438, Dec. 2005.

[12] J. Hu, R. German, A. Heindl, R. Kates, M. Unbehaun. "Traffic modeling and cost

optimization for transmitting traffic messages over a hybrid broadcast and cellular

network," in the Proc. of the Intell. Transp. Syst., 2005. IEEE Sept. 13-15, 2005, pp.

210 – 215

[13] M. Schwartz, Mobile Wireless Communications, New Edition. Cambridge:

Cambridge University Press. 2005.

[14] A. Bria, "Cost-based resource management in hybrid cellular-broadcasting

systems," in the Proc. of the IEEE 61st Vehicular Technology Conference, vol. 5, pp.

3183 - 3187, 2005

128

[15] M. Shavit, R. Miucic, A. J. Gryc, "Firmware update over the air (FOTA) for

automotive industry", in the Proc. of the SAE The 14th Asia Pacific Automotive

Engineering Conference, Aug. 5-8, 2007, Hollywood, CA, Paper Number: 2007-01-

3523.

[16] C. Desiniotis, K. Kypris, Y. Markoulidakis, "Performance evaluation of GPRS

MCAST multicast over GPRS solution," in Proceedings of 14th IST Mobile &

Wireless Communication Summit, Dresden, Germany, June 2005, online:

www.eurasip.org/Proceedings/Ext/IST05/papers/394.pdf

[17] Federal Communication Committee: Dedicated Short Range Communications

(DSRC) Service, available at www.fcc.gov

[18] Wikipedia: XM Satellite Radio http://en.wikipedia.org/wiki/XM_Satellite_Radio

Accessed 12/07.

[19] Federal Communication Committee (FCC): Cellular Services, available at

http://wireless.fcc.gov/services/index.htm?job=service_home&id=cellular, accessed

12/07.

[20] S. Duri, M. Gruteser, X. Liu, P. Moskowitz, R. Perez, M. Singh, J. Tang,

"Framework for security and privacy in automotive telematics," in the of the

Proceedings of the 2nd International Workshop on Mobile Commerce, Atlanta,

Georgia, USA., 2002, pp. 25 - 32

[21] Han, C., Kumar, R., Shea, R., and Srivastava, M. 2005. Sensor network software

update management: a survey. Int. J. Netw. Manag. 15, 4 (Jul. 2005), pp. 283-294.

[22] http://en.wikipedia.org/wiki/Forward_error_correction

129

[23] R. Bell, “Multiplexing-Past, Present and Future,” SAE Paper Number 760178,

1976.

[24] C. A. Lupini, “Advantages of integrating a serial data link controller with CPU

[vehicles],” in IEEE 42nd Vehicular Technology Conference, May 10-13, 1992, vol.2

pp. 1051-1055.

[25] C. A. Lupini, Vehicle multiplex communication: Serial data networking applied to

vehicular engineering, SAE International 2004, Warrendale, PA, USA, ISBN 0-7680-

1218-X.

[26] W. Lawrenz, “Network Development Techniques,” IEE Colloquium on Vehicle

Networks for Multiplexing and Data Communication, Dec 19, 1988, pp. 5/1-5/8.

[27] A. Masrur, "Digital simulation of an automotive multiplexing wiring system," IEEE

Transactions on Vehicular Technology, Vol. 38, No. 3, pp. 140-147, August 1989.

[28] A. Masrur, "Studies on some alternative architectures for fault-tolerant automotive

multiplexing networking systems," IEEE Transactions on Vehicular Technology, Vol.

40, No. 2, pp. 501-510, May 1991.

[29] W. Lawrenz, CAN system engineering: From theory to practical applications, 1997

Springer-Verlag, New York, Inc., USA, ISBN 0-387-94939-9.

[30] W. Lawrenz, “Communication protocol conformance testing - Example LIN -,”

IEEE International Conference on Vehicular Electronics and Safety, pp. 155-162,

December 13-15, 2006.

[31] C. A. Lupini, "Multiplex bus progression 2003," in Proc. of the SAE 2003 World

Congress, March 3-6, 2003, Detroit, Michigan, USA, Paper Number: 2003-01-0111.

130

[32] J.G. Kassakian, and D.J. Perreault, "The future of electronics in automobiles", in

Proceedings of the 13th International Symposium on Power Semiconductor Devices

and ICs, 2001. pp. 15-19, Osaka, Japan

[33] H. Kleinknecht, “CAN Calibration Protocol specification”, Version 2.1, Germany,

1999.

[34] Vector CANtech, URL: http://www.vector-cantech.com/[15 August 2005]

[35] Accurate Technologies Inc., URL: http://www.accuratetechnologies.com/corp/ [15

August 2005]

[36] S. Misbahuddin, S. M. Mahmud and N. Al-Holou, "Development and performance

analysis of a data-reduction algorithm for automotive multiplexing," IEEE

Transactions on Vehicular Technology, Vol. 50, No. 1, pp. 162-169, January 2001.

[37] D. Huffman, “A method for the construction of minimum redundancy codes,” Proc.

of IRE, vol. 40, pp. 1098–1101, Sept. 1952.

[38] G. G. Kempf, M. J. Eckrich, and O. J. Rumpf, “Data reduction in automotive

multiplexing systems,” SAE paper 940 135, pp. 45–50.

[39] Storer, J.A. and Szymanski, T.G., “Data compression via textual substitution”,

Journal of the ACM, Vol. 29, No. 4, October 1982, pp. 928-951.

[40] P. R. Ramteke and S. M. Mahmud “An adaptive data-reduction protocol for the

future in-vehicle networks,” SAE Transactions on Passenger Cars: Electrical and

Electronic Systems, pp. 519-530, 2005.

[41] Bosch, “CAN specification Ver 2.0,” Robert Bosch GmbH, Stuttgart, Germany,

1991.

131

[42] R. Miucic and S. M. Mahmud, “An improved adaptive data reduction protocol for

in-vehicle networks,” SAE Transactions on Passenger Cars: Electrical and Electronic

Systems, pp. 650-658, 2006.

[43] Microchip, “dsPIC30F4011/4012 Data Sheet,” Microchip Technology Inc.,

Chandler, Arizona, 2005.

[44] STMicroelectronics, “ST92F124/ST92F150/ST92F250 Data Sheet Ver 1.4,”

Microchip Technology Inc., Geneva, Switzerland, 2003.

[45] “J1939” SAE Standard, SAE International.

[46] “CAN Protocol” Vector Training Material, Vector CANtech Inc., Novi, MI, 2005.

[47] D. Jiang, V. Taliwal, A. Meier, W. Holfelder, and R. Herrtwich; "Design of 5.9 GHz

DSRC-based vehicular safety communication," IEEE Wireless Communications, Vol.

13, Issue 5, Oct. 2006 pp. 36 – 43.

[48] Bai F. and H. Krishnan; "Reliability analysis of DSRC wireless communication for

vehicle safety applications," in IEEE 2006 Intelligent Transportation Systems

Conference, 2006 pp. 355 – 362.

[49] X. Chen, H.H. Refai, and X. Ma, "A quantitative approach to evaluate DSRC

highway inter-vehicle safety communication," in 2007 IEEE Global

Telecommunications Conference, Nov.26-30, 2007 pp. 151 – 155.

[50] S. Biswas, R. Tatchikou, and F. Dion, "Vehicle-to-vehicle wireless communication

protocols for enhancing highway traffic safety," IEEE Communications Magazine,

Vol. 44, Issue 1, Jan. 2006 pp. 74 – 82.

132

[51] M. Mehta and M. Guinan, "The utilization of multi-antenna enhanced mobile

broadband communications in intelligent transportation systems," in 7th International

Conference on ITS Telecommunications, June 2007 pp. 1 – 4.

[52] M. Lott, M. Meincke, and R. Halfmann, "A new approach to exploit multiple

frequencies in DSRC," in IEEE 59th Vehicular Technology Conference, May 17-19,

2004, Vol. 3, pp.1539 – 1543.

[53] Vehicle Infrastructure Integration (VII), USDOT Major Initiative, www.its.dot.gov/vii/

[54] Cooperative Intersection Collision Avoidance Systems (CICAS), USDOT Major

Initiative, www.its.dot.gov/cicas/index.htm

[55] Vehicle Safety Communications Consortium, USDOT Major Initiative, www-

nrd.nhtsa.dot.gov/pdf/nrd-12/CAMP3/pages/VSCC.htm

[56] SAE, “Draft SAE j2735 Dedicated Short Range Communications (DSRC) Message

Set Dictionary,” PA SAE, May 16, 2008

[57] IEEE, "Wireless Access in Vehicular Environments (WAVE)," Draft 802.1lp/D1.0,

New York, NY: IEEE Press, February, 2006.

[58] IEEE, “Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY)

Specifications,” IEEE Press, February, 2007.

[59] IEEE Std P1609.1 (VT/ITS) Standard for Wireless Access in Vehicular

Environments (WAVE) - Resource Manager


Environments - Security Services for Applications and Management Messages


Environments (WAVE) - Networking Services

133


Environments (WAVE) - Multi-Channel Operations

[63] "Distance Calculation: How to calculate the distance between two points on the

Earth" available at http://www.meridianworlddata.com/Distance-Calculation.asp

online 12/19/2008

[64] “Haversine Formula” available at http://en.wikipedia.org/wiki/Haversine_formula,

online 12/19/2008

[65] H. Masui, T. Kobayashi and M. Akaike “Microwave path-loss modeling in urban

line-of-sight environments” IEEE Journal on Communications, Vol. 20, Issue 6, Aug.

2002 pp. 1151 - 1155.

[66] H. Masui, M. Ishii, S. Takahashi, H. Shimizu, and T. Kobayashi; "Difference of

break-point characteristics due to mobile antenna heights in microwave urban LOS

propagation", in the Proceedings of the International Zurich Seminar on Broadband

Communications, 2000.

[67] L. Cheng, B. E. Henty, D. D. Stancil, F. Bai, P. Mudalige, „Mobile vehicle-to-

vehicle narrow-band channel measurement and characterization of the 5.9 GHz

Dedicated Short Range Communication (DSRC) frequency band“ IEEE Journal on

Selected Areas in Communications, Vol. 25, Issue 8, Oct. 2007 pp. 1501 – 1516.

[68] VIIC System Engineering Integration and Test (SEIT) Team, "Test Report for

Phase T1.1 DSRC VIIC Fleet Vehicle Communication Link Testing" 08 October 2008

[69] O. Landron, M.J. Feuerstein, T.S. Rappaport "A comparison of theoretical and

empirical reflection coefficients for typical exterior wall surface in a mobile radio

http://www.meridianworlddata.com/Distance-Calculation.asp�

http://en.wikipedia.org/wiki/Haversine_formula�

134

environment," IEEE Trans. Antennas and Propagation, March 1996, Vol. 44 No.3,

pp.341-51.

[70] T. S. Rappaport, Wireless Communications: Principles and Practice, second

edition, Prentice Hall PTR, Upper Saddle River, NJ 07458, 2002

[71] W. C. Jakes, Microwave Mobile Communications, Wiley, 1974.

[72] D. Tse, P. Viswanath, Fundamentals of Wireless Communication, Cambridge:

Cambridge University Press, 2005

[73] B.-S. Lee, C. S. Yim, D. H. Ahn, and D. G. Oh, "Performance evaluation of the

physical layer of the DSRC operating in 5.8 GHz frequency band," ETRI Journal,

vol.23, no.3, Sept. 2001, pp.121-128

[74] Nippon Antenna. 802.11a/DSRC Radio Antenna DEN-HA001-001 and DEN-

HA001-002. A specification booklet.

[75] “Appendix G: Field Testing and Evaluation of WAVE/DSRC Communications

Functionalities” online 1/14/2009 available at http://www-nrd.nhtsa.dot.gov/pdf/nrd-

12/060419-0843/PDFs/AppendixG.pdf

[76] J. Q. Shang, J. A. Umana, F. M. Bartlett, J. R. Rossiter, "Measurement of complex

permittivity of asphalt pavement materials," J. of Transportation Engineering, Vol.

125, No. 4, July/August 1999, pp. 347-356.

[77] A. J. Rustako Jr., N. Amitay, G. J. Owens, and R. S. Roman, "Radio propagation

at microwave frequencies for line-of-sight microcellular mobile and personal

communications," IEEE Transactions on Vehicular Technology, vol. 40, no. 1, Feb.

1991

http://www-nrd.nhtsa.dot.gov/pdf/nrd-12/060419-0843/PDFs/AppendixG.pdf�

http://www-nrd.nhtsa.dot.gov/pdf/nrd-12/060419-0843/PDFs/AppendixG.pdf�

135

[78] P. Supanakoon, A. Pokang, S. Promwong, S. Noppanakeepong, and J. Takada,

"Regression models of ultra wideband ground reflection path loss based on peak

power loss," in the Proc. of the Asia Pacific Conference on Communication, Oct. 18-

20, 2007, pp. 15-18.

136

ABSTRACT

INTELLIGENT COMMUNICATION FOR FUTURE AUTOMOBILE NETWORKS

by

RADOVAN MIUCIC

December 2009

Advisor: Dr. Syed Masud Mahmud

Major: Computer Engineering

Degree: Doctor of Philosophy

With every new model vehicle, the features and functionality continue to

increase. Increased functionality also increases the size and complexity of the

embedded software. With an increase in software size, ability to modify and change the

original software becomes a necessity for automotive companies. Traditionally and

currently, automotive companies update software in vehicles using old-fashioned, hand-

held, wired programmer. Advances in the wireless communication technologies have

allowed a new approach in updating in-vehicle software. This work presents a unique

proposal to manage software download using wireless communication. In addition, this

work presents improvements in the in-vehicle network for efficient distribution of data

required by software modules of various electronic control units (ECUs). This

dissertation work also developed an efficient bootloader algorithm for programming

vehicle ECUs by sharing available resources, such as memory, within the entire vehicle

system. In addition, this work also examines channel characteristics of the wireless

protocol, commonly known as Dedicated Short Range Communication (DSRC), for

automotive usage such as software uploads, safety and other telematics applications.

137

AUTOBIOGRAPHICAL STATEMENT

Radovan Miucic received the B.Sc. degree in Electrical and Computer Engineering and the M.S. degree in Computer Engineering both from Wayne State University in 2001 and 2002 consecutively. From 1999 to 2007, he has been working as an Embedded Software Engineer and since 2007 as a Wireless Research Engineer in the automotive industry. He is a member of IEEE and SAE society. His research interests include intra-vehicle, vehicle-to-vehicle and vehicle-to-infrastructure communications, embedded microcontroller design, vehicle safety, and software maintenance. His recent publications include: Journal: 1. Radovan Miucic, Syed Masud Mahmud and Zeljko Popovic, “An Enhanced Data

Reduction Algorithm for Event Triggered Networks,” IEEE Transactions on Vehicular Technology, Vol. 58, No. 6, pp. 2663-2678, July 2009.

2. Radovan Miucic and Syed Masud Mahmud, “An Improved Adaptive Data Reduction Protocol for In-Vehicle Networks,” SAE Transactions on Passenger Cars: Electrical and Electronic Systems, pp. 650-658, 2006.

Conference: 1. Radovan Miucic, Zeljko Popovic, Syed Masud Mahmud, “Experimental Characterization

of DSRC Signal Strength Drops,” accepted for publication in 12th International IEEE Conference on Intelligent Transportation Systems, St. Louis, MO, USA, October 3-7, 2009.

2. Radovan Miucic, Tom Schaffnit, “Communication in Future Vehicle Cooperative Safety Systems: 5.9 GHz DSRC Non-Line-of-Sight Field Testing,” Proc. of the SAE 2009 World Congress, Detroit, MI, April 2009, Paper Number: 2009-01-0163.

3. Radovan Miucic and Syed Masud Mahmud, "Wireless Reprogramming of Vehicle Electronic Control Units," Proceedings of the 5th IEEE Consumer Communications and Networking Conference, Las Vegas, NV, January 10-12, 2008, Paper Number: FP1-S2-5.

4. Moshe Shavit, Radovan Miucic, and Andrew J. Gryc, “Firmware Update Over The Air (FOTA) for Automotive Industry,” Proceedings of the 14th Asia Pacific Automotive Engineering Conference, August 5-8, 2007, Hollywood, California, Paper Number: 2007-01-3523

5. Radovan Miucic and Syed Masud Mahmud, “An Improved Adaptive Data Reduction Protocol for In-Vehicle Networks,” Proc. of the SAE 2006 World Congress, April 3-6, 2006, Detroit, Michigan, USA, Paper Number: 2006-01-1327

6. Radovan Miucic and Syed Masud Mahmud “Wireless Multicasting for Remote Software Upload in Vehicles With Realistic Vehicle Movements,” Proc. of the SAE 2005 World Congress, April 11-14, 2005, Detroit, Michigan, USA, Paper Number: 2005-01-0323.

7. Radovan Miucic and Syed Masud Mahmud “An In-Vehicle Distributed Technique for Remote Programming of Vehicles' Embedded Software,” Proc. of the SAE 2005 World Congress, April 11-14, 2005, Detroit, Michigan, USA, Paper Number: 2005-01-0313.

8. Radovan Miucic and Syed Masud Mahmud, “Mobile Multicasting for Remote Software Update in Intelligent Vehicles,” Proceedings of the 4th Annual Intelligent Vehicle Systems Symposium of National Defense Industries Association (NDIA), National Automotive Center and Vectronics Technology, June 22 –24, 2004, Traverse City, Michigan, pp. 199-207.

intelligent communication for future automobile...

Documents