Wireless Magnetic Based Sensor System

For Vehicles Classification

P. Ghislain(1)

, D. Carona(1)

, A. Serrador(1)

, P. Jorge(1)

, P. Ferreira(1)

, J. Lopes(2)

(1) Instituto Superior de Engenharia de Lisboa, Lisbon, Portugal,

(2)Brisa Inovação e Tecnologia, Lisbon, Portugal

35945@alunos.isel.pt, {dcarona, aserrador, pjorge, pferreira}@deetc.isel.ipl.pt, jlopes@brisa.pt

Keywords: Smart Road Sensor, Magnetometer, ZigBee, Vehicles Classification.

Abstract: This paper presents a full pre-industrial prototype implementation of an innovative system capable to

detects vehicles and classify them. This work study the architecture of a full autonomous road sensor

network that count and classify passing vehicles, reducing the impact of conventional wired systems in

roads, by dramatically reducing their installation and operational costs. The power supply can be supported

either by solar harvest or by none rechargeable batteries. For communications the IEEE 802.15.4 with

ZigBee on top are used to transmit digitalized vehicle’s magnetic analogue signatures up to a vehicles

classification server. This information is acquired using a magnetometer translating in voltage the variation

of the natural magnetic field of the earth caused by ferromagnetic components of a vehicle passing over the

sensor. Results show clearly that it is possible to distinguish vehicles magnetic signatures per model and

class even at different speeds.

1 INTRODUCTION

With the growing number of vehicles travelling on roads, the need of road traffic control and monitoring also increases. The available technologies includes inductive loop, video image processing, weight-in-motion, pneumatic tube, piezoelectric, ultrasonic, acoustic, infrared and magnetic sensors. Inductive loops have high installation and operation costs. Camera image processing requires expensive specialized processor unit per equipment and is sensible to weather conditions. The piezoelectric systems use a transversal tube on the road, subject to a rapid deterioration and not effective for highways. The others techniques have a smaller range of application than magnetic sensors [1]. The sensor technology used in this study is the Anisotropic Magnetic Resistance (AMR) with a magnetic resolution of one part in ten thousand of the Earth magnetic field [2]. With such a good precision, it has been clear during the development that the quality of the electronic design plays an important role to retain this level of precision. For this application the Earth magnetic field has an important advantage on previous magnetic

technologies: its presence anywhere on Earth. Inductive loop requires the production of an artificial magnetic field with significant energy consumption and the connection of an external power supply. The natural magnetic field may vary from 0.2 to 0.7 Gauss and its lines of forces are not horizontal but oblique, changing place to place in strength and inclination. The intensity of this field can vary with the weather and the season. It has not the stability neither the accuracy of the artificial inductive loop [3]. This variability is a challenge that must solve the developed device. This prototype transmits data using a low bitrate wireless communications with low energy consumption. The sensor cost, installation and operation of the proposed road sensor network is expected to be much lower compared with the inductive loop systems solutions. The control centre that processes the vehicles magnetic signatures is a fundamental sub-system in road management to identify traffic load per vehicles behaviour/profiles/classes. Figure 1, show the laboratory prototype.

Figure 1: Developed Prototype.

The research work developed by ISEL laboratory in vehicle classification methods based on many Smart Road Sensors technologies [4][5][6] contributes to the knowledge needed in some aspects of the work presented. The study of magnetic sensor is a natural development of this strategy. Works of other research centres have been studied in order to gain knowhow.[7][8][9].

To describe the work in more detail, this paper is structured as follows: in Section 2, the system architecture and each component implementation overview is provided, in Section 3 results of the performed tests are presented, while in Section 4 conclusions and future work are summarised.

2 ARCHITECTURE

The system architecture proposes for vehicle count and classification the insertion of one sensor in the centre of each motorway lane. On the road side, the access point collects data from the sensors and relays them to the WAN network of the road operator via GPRS/3G or fixed network. The control centre receives the sensor signature and processes it in real time. Then an information system displays the traffic load for each installed road sensor and gives access to a statistical distribution of the detected vehicle’s class.

The low cost of each detection point allows a higher density of these road sensors and a much better visualization of the traffic load.

Figure 2: System Architecture Overview.

The installation of the access point close to sensors (Figure 2) decreases the wireless transmission power to the lowest possible value, extending the sensor battery live time.

2.1 Magnetometer

The selection criteria for the magnetic sensors for this work were based on a set of principles: sensitivity to the Earth magnetic field variation caused by a ferromagnetic objects, the energy consumption and the financial sensor cost.

Another important selection criterion was the ability to restore the sensibility of the sensor via included coils and a Set/Reset operation that only one supplier offers. We chose the B component [2].

Reference A B C

Supply [V] 5-12 5-25 1.8-25

Current [mA] 10 5 1

Set/Reset Current [A] 3 0.5 0.5

Sensibility [mV/V/Gauss] 3.2 1 1

Resolution [µGauss] 27 85 120

Table 1: Comparison between sensors.

The family of Anisotropic Magneto-Resistances (AMR), allows to measure the Earth magnetic field with a precision below 0.5%. This background field is quite permanent, suffering only small and slow changes. Thus, the sensor includes an auto-calibration algorithm that adapts the measures so that this reference can be considered stable during a period of time well suited to vehicle detection.

Vehicles metallic characteristics produces typically a flux variation around 50% of the Earth magnetic field, but the variation can be many times this value with vehicles built with reinforced iron parts. In those rare cases, the magnetic signature will be partially saturated but still distinctive and usable for classification.

The project also wanted to determine the overall parameters offered by the technology and two orthogonal magneto sensitive axes where important to study the relation between the orientation of magnetic field and the vehicle’s movement [2].

Modern vehicles include many ferromagnetic components as show into the representation in Figure 3 [3], for example the wheels axels, motor components, transmission, gas escape are made of steel or iron. Electric motors and alternator creates a magnetic field. All those components concentrate or disperse the Earth magnetic field and thus, create detectable variations at the output of the AMR sensor.

Figure 3: Vehicle magnetic signature [3].

The resistive elements of the sensor are placed in a Wheatstone Bridge shown in Figure 4 that allows to compare two tensions coming from two resistive dividers. Each resistive divider is made of two branches which are composed of 6 Permalloy magneto-resistive elements oriented at 45º. Each branch elements are positioned at 90º of their companion branch elements. When a branch has low magnetic resistance, its companion branch will have a high one. Both dividers are in a mirrored conformation and respond in an opposite way allowing a maximum tension difference between them.

Figure 4: Structure of the sensor [3].

Where Vb is the applied tension, in our case, 3.3V, Out+ and Out-, the tension of the two dividers values Vb/2 when the magneto-resistance are equal, therefore when there is no external magnetic field applied. The output differential tension (Out+, Out-) is equal to the ratio of the variation of the resistance of one branch on the resistance of it, times the applied tension Vb. It is a small voltage value of few mV. On the influence of the Earth magnetic field, a value of 0.5 Gauss creates a 2.5mV of differential tension. The magnetometer requires a correction of the bias created by the local value of the magnetic field of the Earth and an amplification of the differential tension which gain should be between 500 and 2000.

Those values changes at each installation’s location because the orientation of the main magnetic sensor direction must be parallel to the road traffic direction and therefore, the strength of the Earth magnetic field is always different.

Two possibilities are available, the manual configuration in place or the auto-calibration of the sensor with a set of digital potentiometers. As the Earth field can change on a long time scale, the second option is a requirement for an industrial solution.

2.2 Electronics components

The AMR sensor that provides a small tension variation between two outputs must be followed by a stage of OpAmps, operation amplifiers, that provides a bias correction, a difference and an amplification operation.

The 3 operations can be done in one circuit or via many OpAmp in cascade, depending of pretended level of precision. The signal is conditioned to remain into the range of [0, VCC] Volts, where VCC is the power supply voltage.

Those two tensions values are then redirect to a microprocessor that can digitalize them, store the data and after the end of the capture of a magnetic signature, send it to the wireless radio subcomponent.

The prospection of electronics components to build a LR-WPAN, low rate wireless personal area network, was limited essentially to the wireless solution available for the ISM, Industrial Security and Medical, band without licencing procedure.

The IEEE 802.15.4 compliant components offer the better choice and we made a market research in order to select a good and reliable solution. The more advanced components offers a complete System on a Chip solution, including a microprocessor, a radio subcomponent and many other features that target the component to specific markets. The component that we choose accept the digitalization of up to 8 analogue inputs, offer 8 Kbyes of RAM, um transceiver for 2.4 GHz IEEE 802.15.4 compliant, one OpAmp, two power outputs[10]. It enables to shrink the sensor board and its quality has been confirmed during the project.

2.3 ZigBee

To transmit wirelessly the collected information, the IEEE 802.15.4 standard with ZigBee on top is adopted because it was developed to operate in short range communications and low power consumption systems, supporting also high level networking features.

Many protocols use IEEE 802.15.4 as their PHY and datalink layer. WirelessHART, ISO110a, ZigBee, 6loWPAN and a set of other proprietary network layers.

ZigBee define in OSI, Open System Interconnection, from the Network up to part of the Application layer. In our work, we have not found a specific ZigBee profile for sensors data capture through the implementation of duty cycles; so, we have developed a proprietary ZigBee profile and use only the ZigBee Network layer.

Our communication features includes a transport layer with data fragmentation, packet resend, packet acknowledge if well received by the ZigBee coordinator in the access point; the support of many sensors for one coordinator and an asynchronous serial gateway.

The coordinator firmware had also to be developed. From what ZigBee Alliance announced and what one could use without firmware development there was a gap. This problem is attended through the set of specifications that where launch since 2010 until now by the ZigBee Alliance. Fabricants have naturally a big problem with those specifications: they require a lot of manpower to develop them and the market look for very low cost solutions. For the moment, those announced specifications are not always available.

The ZigBee coordinator firmware that we have built, suits for the project but is in a revision process to adopt an architecture that allows the reception of packet burst from ZigBee. A circular list of buffers memorizes the incoming packets. The relay of them is made asynchronously to the serial gateway. The energy cost of the transfer of a signature is maintained to the minimum with maximum speed.

The road sensor activates communications only when it has data to send or after some seconds of inaction. Data from the server and the access point are access only at that moment.

The ZigBee stack is well suited for rapid development project and has parameters that are available to tune it to specific applications. The main features that are offered are the automatic association and re-association to the network knowing that the coordinator chooses a channel that is not known by the terminal equipment.

The ZigBee features used by the project are limited and the use of the ZigBee stack is not mandatory. Another Stack simpler and power saving, may be proprietary, could also fit all the needs, without license costs.

2.4 Orientation and attenuation

The orientation of the road sensor’s antenna maximum radiate power should be in a quarter of hemisphere in the direction of the road side as the sensor is always oriented toward the vehicles direction. Presently, it is plan to use an isotropic ceramic antenna embedded into the cover of the road sensor, allowing the coordinator installation anywhere around. To increase radio link performance it is recommended to use directional antennas to support adverse weather like rain and snow.

The coordinator must be oriented with a vertical angle of a least 30º and must use a directive antenna. For sharp angles the link attenuation rises quickly because there is no line of sight between antennas. In fact, the road sensor in located into a hole that is a kind of wave guide and the isotropic radiation of the ceramic antenna is affected and the majority of the radiation goes vertically. Higher is the angle of the coordinator, better is the reception.

The attenuation caused by the cover material has taken our attention. We have made some attenuation tests and determine that acrylic and glass could be used for this application as well as asphalt which present 1 dB/cm of attenuation. This is surprising but asphalt is made of very long aliphatic chains and is hydrophobic so the result is coherent with other similar material like plastics. Modified asphalt with the substitution of the gravel with adequate low attenuation material can constitute a good cement to embed the device into the road.

2.5 Power management

The sensor was designed to extend the battery live time to the maximum, thus the device architecture is modular and each part it is only active when strictly necessary. The CPU remains always the master and do not depend of external interrupt to wake-up. It is the most convenient architecture and also the more efficient. The CPU uses one output signal to suspend the magnetometer and manage all internal functions.

The alternative architecture use an external interrupt to wake-up the CPU, and to avoid the appearance of an indeterminate state in which the sensor would never wake-up, a more complex electronic circuit is needed which increases the power consumption and therefore this approach was not taken into account.

The power management developed for the sensor device is based on the electronic architecture and on the firmware development.

When ZigBee do not communicate with the network, no power is used to supply the radio subcomponent. The road sensor wakes up

92%

0%

20%

40%

60%

80%

100%

180 x 10 90 x 10 90 x 5 45 x 5 20 x 5

Ene

rgy

savi

ng

rati

o

Maximal velocity x samples per vehicle meter

Normal

POWER_SAVING

periodically to check for incoming data or detect a car, digitalize and send the magnetic signature and check for incoming data. This optimization cut 75% of the energy requirements of the SoC: from 32 mA the consumption falls to 8 mA. The ZigBee Stack provides elements to implement a closed loop TX Power control but ZigBee don’t implement it. Each received packet of 802.15.4 comes with a LQI, line quality indicator whish calculus is based on the ED, Energy Detection, a signal to noise parameter. The received LQI coming from the coordinator is used by the road sensor to manage dynamically the TX Power. This power control is important to ensure the auto adaptation of the road sensor to changing weather conditions because heavy rains attenuate the radio signal and require more TX power. And when weather conditions turns better, the priority must be to diminish TX power in order to minimize transmission energy cost. The link budget between the road sensor and the coordinator must have an important power reserve. ZigBee chip have a sensibility about 95 dBm. The design of the antennas is essential to make a road sensor system adaptable to any weather conditions: traffic jam and accident happens during those situations and the road sensor mission is to guaranty full functioning state and information dispatch without failures at that moment.

As the road sensor should only send data to the right side of the motorway in right hand traffic countries, the gain of its antenna can be better than 6 dBi which correspond to a quarter of hemisphere of the isotropic antenna of 0 dBi. On the coordinator side, it is very important to use an antenna with a gain higher than 14 dBi, which is about a 60º BW, Beam Wide at 3 dB. This combination offers more than 20 dB total antennas gain that can constitute part of the necessary power budget reserve. The space diversity that some fabricants propose is important if the application is sensible to multipath but, for this open air application, few installations may suffer from short distance attenuation due to the lack of reflection of the transmitted signal. In order to achieve better power saving, the firmware has been compiled to activate the sleeping mode automatic switching that ZigBee stack provide. In the Figure 5, the Normal firmware shows a power saving starting only with a low frequency sampling. After an optimization of the firmware, we compiled a version that processes much better this switching and can achieve about 90% of saving, switching from 8 mA without duty cycle to 0.7 mA.

Figure 5: Energy consumption with both firmware.

The research team believes that the firmware can be still tuned to achieve better saving performance.

2.6 Power sources

With a power saving energy mode spending 0.7 mA, this sensor represents a very low drain voltage equipment enabling the use of none rechargeable batteries. A set of 2x2 D Professional Long Duration Alcaline Batteries offers a capacity of 33000 mAh at 2.2-3.3 V. The discharge per year is about 3% [11] and this means about 4,94 years of useful life. A lithium battery offers more than a 1.5 times the capacity of an Alkaline and very low discharge current. Lithium batteries are not stable when exposed to temperatures higher than 60º C and therefore should not be a first choice for a road sensor in Portugal [12].

If the sensor architecture privilege precision and 3 directions orthogonal magnetic sensing, the energy consumption is higher and the sensor must be supplied by solar harvest power.

2.7 Power capture

To determine the size of the solar captor, it is necessary to calculate the expectable energy power received during the month and the place in Portugal of lowest solar radiation. Hopefully, European program PVGIS [13] provides this information for any European Country, including Portugal.

After that, using a set of atenuation parameters, our model gives a value of the minimum daily average power that can be captured with moncristaline or amorphous silicium cell. The ratio of the necessary power on the capturable power per m

2 defines the minimal surface.

c

n

P

PS

24

(1)

where,

Pc : is the daily average solar power per m2

available after the elimination of all

attenuations;

Pn : necessary power per hour for the sensor;

S : minimum surface for the solar captor in m2.

This expression is expanded with the following parameters:

cojindhdirtrcc

n

ffrRrRrP

PS

))((

24 (2)

where,

Pr : daily average solar power radiation of the

location and month of lowest radiation in the

geographical area of installation;

rtrc : conversion percentage (1 minus losses)

accounting for temperature, reflection of the cell

surface and DC-DC conversion losses.

Typically 70%;

Rdir e Rind : percentage of direct and indirect

solar radiation, the sum of both equal 100%.

Used 50% for both. Use PVGIS estimation;

rh: conversion percentage due to the horizontal

position of the sensor. Used 72%;

rj: conversion percentage due to the window of

the sensor. Used 70%;

fo: osculation factor made by the passing vehicle

shadow. Used 92,5% ;

fc: conversion percentage of the solar cell. Used

15% and 8%;

For rh, the PVGIS provides an estimation that can be used. The overall conversion performance is 5% with monocrystalline Silicon and 2.7% with amorphous cells.

The choice of a solar cell is facilitated by the existence of DC-DC converters that accept low voltage sources under 0.3 V, transform and regulate an output tension of 3.3 or 5 V using charges pump with an efficiency around 90% [14] and [15]. The MPP, Maximal Power Point, of the cell can be obtained via a specific electronic circuit but the efficiency is low for a single solar cell and a serial load with the battery may be more effective and simple.

For the rechargeable battery the horizon of 3 years maximum life expectancy is today a reality. The use of super capacitors has been evaluated and it is possible with one DC-DC converter at the output extract about 80% of the capacity.

2.8 Firmware development

The selected component is a SoC that allows inserting code into the processing of the ZigBee stack. The firmware had to achieve a set of goals:

scan the magnetometer axis in the direction of

the traffic in order to detect passing vehicles,

start the scanning of the magnetic signature at a

rate that is adapted to an estimated speed in

order to minimize data size,

detects the end of the vehicle,

insert the start of curve data store in a circular

memory buffer,

adapts the rate of sampling from start to the first

vehicle’s axis,

send the data via ZigBee and finally

set/reset the magnetometer in order to maintain

its high anisotropy and magneto-resistance

sensitivity.

Maintenance processing of the road sensor includes:

the TX power management,

the control of the power of the magnetometer

and operational amplifiers,

auto calibrations and

Earth Magnetic Field detection.

The speed estimation (3) is made by the firmware and based on the average distance between the start of digitalisation and the first wheel of the car. As each car model is different, this estimation will be different among cars and trucks but it is expected to always be similar for the same vehicle. This allows having a reasonable and fixed amount of samples per vehicle’s model, independent of the speed capture. It is also a discriminant factor for the pattern matching engine in the server and can help the discovery of the car model.

t

dVes

(3)

Where:

Ves is the speed estimation;

d the distance between start and first axe;

Δt the duration of the between both;

The coordinator has been also a matter of attention because the reception of packets depends of its availability. The use of an input circular buffer list waiting for the transmission on the serial gateway was decisive for the coordinator radio reception efficiency.

3 RESULTS

The first question to solve was the orientation of the

sensor. The reproducible effect on a magnetic axis is

achieved when the vehicle motion is in its direction.

This means that we do not orient the main magnetic

0

0,5

1

1,5

2

2,5

3

0 20 40 60 80 100Mag

ne

tom

ete

r Y

axi

s te

nsi

on

[V

]

Samples

Signature

EMF

axis in the direction of the north magnetic pole that

would provide the best sensing amplitude, but

always in the traffic direction. The parallel sensing direction, the 2

nd axe, has a

quite different measure of the Earth Magnetic Field and its amplitude is different from the main axe. The bias can be corrected via a supplementary OpAmp for the calibration than the one of the main axis. If the magnetic axes input signals are amplified with different gain, it can be demonstrated that the vector norm of the two orthogonal axes change if we turn the sensor horizontally, for the same car. In other word, using different gain on both magnetic axes can produce different magnetic signature for the same car when the road sensor is install in different location. This must be ovoid in order to guaranty information coherence. The 1-axes magnetometer oriented in the direction of the traffic flux can provide already much information, completely coherent, with a much lower cost in processing and energy.

Figure 6 and Figure 7 shows the forward signature of a BMW and the parallel signature of the same car.

Figure 6: Forward axis scanning.

Figure 7: Lateral axis scanning.

The tests on the algorithm that was designed show a correct estimation of the Earth magnetic field for both axes. The detection and scanning of the magnetic signature is completed and well captured. The vehicle count is therefore correct.

We verify that for a specific car model, at 20 up to 60 km/h, we receive about the same amount of fixed samples validating the speed detection algorithm. With other car model, we had the same result but with a different number of samples for this car model. Those results are encouraging because we have still to turn better de definition of boundaries.

Figure 8: Main axis, vehicle at 20km/h.

Figure 9: Main axis, same vehicle at 40km/h.

Figure 10: Main axis, same vehicle at 60km/h On the other hand, the variability of the curves

indicated that the signature has an important error margin as it can be seen in the Figures 8 and 9.

We believe that a single axe magnetic sensor correctly oriented in the flux traffic direction is sufficient to characterize a vehicle model. We have test with Van and others small cars and the magnetic signature is very different car by car.

0 10 20 30 40 50

0

0,5

1

1,5

2

2,5

3

Samples

Ma

gn

eto

me

ter

X a

xis

te

ns

ion

[V

]

0 10 20 30 40 50

0

0,5

1

1,5

2

2,5

3

Samples

Ma

gn

eto

me

ter

X a

xis

te

ns

ion

[V

]

0 10 20 30 40 50

0

0,5

1

1,5

2

2,5

3

Samples

Ma

gn

eto

me

ter

X a

xis

te

ns

ion

[V

]

The tuning of the parameters of our firmware will allow defining an optimized set and achieving easily a detection of vehicle classes as well as we believe that the model detection is possible.

4 CONCLUSIONS

The firmware that has been developed process well the magnetic signature capture and allow to confirm that for each tested vehicle, there was a very distinctive signature with however important variation between successive tests for the same car.

The speed algorithm that we have developed performs well and provides a discrimination factor for the classification engine.

In our view, two sensors devices are possible: a small sensor expect to measure 30 x 80 mm that would only use one magnetometer axis and would be powered by battery, with the advantage of having a low cost production. It is expect, from the consumption model and the test with POWER_SAVING firmware, to achieve at least 5 years of permanent operation. The use of a special asphalt cement to embed the sensor into the road is also an option that is considered in order to turn very simple, quick and cheap the insertion of the sensor into the road.

Another sensor architecture proposed uses 3-axes magnetometers and high precision digitalization. It is expected to measure 125 x 80 mm. A solar cell would power the sensor with a super capacitor to maintain the operation during the worst month of the year in terms of solar radiation power. Rechargeable batteries last only some years, for this, we opt for super capacitors used at 80% of the maximum charge voltage which provide about 10 years of live spanning. This sensor should be inserted into a permanent chamber embed into the road and be easily maintained, providing the best precision available with this technology.

The next step of the research team will be the study of the classification techniques.

ACKNOWLEDGEMENTS

Authors would like to thank Brisa Innovation (Technological provider of the largest Portuguese motorway operator), through the research and development Smart Road Sensors project and to ISEL’s research groups GIEST and M2A.

REFERENCES

[1] G. Leduc, “Road Traffic Data: Collection Methods

and Applications,” JRC Technical Notes, 2008.

[2] Honeywell, “1- and 2-Axis Magnetic Sensors

HMC1001/1002/1021/1022,” 2011. [Online].

http://www51.honeywell.com/aero/common/documen

ts/myaerospacecatalog-documents/Missiles-

Munitions/HMC_1001-1002-1021-

1022_Data_Sheet.pdf.

[3] Honeywell, “Application note 218,” [Online].

http://www51.honeywell.com/aero/common/documen

ts/myaerospacecatalog-

documents/Defense_Brochures-

documents/Magnetic__Literature_Application_notes-

documents/AN218_Vehicle_Detection_Using_AMR_

Sensors.pdf.

[4] A. Ferreira, “Integrated Vehicle Classification

System,” in IEEE Intelligent Vehicles Symposium,

Baden-Baden, Germany, 2011.

[5] A. Ferreira, “Development and Evaluation of an

AVDC System,” in IEEE International Symposium on

Computational Intelligence for Engineering Systems,

Oporto, Portugal, 2009.

[6] A. Ferreira, “Automatic Vehicle Detection and

Classification,” in IEEE Conference on Intelligent

Transportation Systems, Beijing, China, 2008.

[7] S. Y. Cheung, Traffic Surveillance by Wireless Sensor

Networks: Vehicle Detection, Classification and Re-

Identification by Wireless Sensor Networks, VDM

Verlag, 2008.

[8] L. B. -. C. M. L. T. Davide Brunelli, “An Efficient

Solar Energy Harvester for Wireless Sensor Nodes,”

University of Bologna, Italy - Swiss Federal Institute

of Technology, Zurich, Suiça, 2008.

[9] H. R. Hajimohammadi, “Classification of Data Series

at Vehicle Detection,” University of Uppsala, Suécia,

2009.

[10] Texas_Intruments, “CC2530,” 2011. [Online].

http://www.ti.com/product/cc2530.

[11] Energiser, “Alkaline Handbook,” 2008. [Online].

http://data.energizer.com/PDFs/alkaline_appman.pdf.

[12] Energiser, “Lithium Ion Disulfite,” 2009. [Online].

http://data.energizer.com/PDFs/lithiuml91l92_appma

n.pdf.

[13] PVGIS, “PVGIS, Photovoltaic Geographical

Information System,” 2011. [Online].

http://re.jrc.ec.europa.eu/pvgis/index.htm.

[14] Texas Instruments, “Supplying TPS61200 With a

Single Solar Cell,” 2009. [Online].

http://www.ti.com/lit/an/slva345b/slva345b.pdf.

[15] Linear, “LTC3105,” 2011. [Online].

http://www.linear.com/product/LTC3105.