Walking down the STAIRS: Efficient Collision Resolution with Constructive Interference
Xiaoyu Ji, Yuan He, Jiliang Wang, Wei Dong, Xiaopei Wu and Yunhao Liu
INFOCOM, 2014, Toronto
Hong Kong University of Science and Technology
2
Motivation• Wireless Sensor Networks (WSNs)
– Event-driven mode– Low duty cycle operating– Large number of nodes
• CSMA-like protocols– Limitations – Backoff...
3
The Recent Art- COMA• COMA- Contend before data transmission
– Contention packets reserve channel for real data packets
– The drawback: dedicated contention packets in each round
DATA 1Receiver
Sender3
Sender2
Sender1
Collision Contention Data
DATA 1
DATA 2
DATA 2
DATA 3
DATA 3
DATA
DATA
Data DataContention Contention
DATA
DATA
Can we resolve the collision in just one round!
Ref: F. Osterlind, et. al, Strawman: resolving collisions in bursty low-power wirelessnetworks,” in IEEE/ACM IPSN, 2012
4
One-round Collision Resolution• The problem:
– Count, identify and schedule– And of course in one round!
• Approach– Active contention– Virtual ID– Fast identification
5
Our weapon: RSSI Stair Pattern• The observation
– Signals can constructively collide– Requirements of Constructive Interference (CI)
• 0.5 μs• Identical signal waveform
6
The Principle
( ) ( 1) 10 20 log , 21CI k CI k
kRSSI k
k
Proposition: Given the superposed signal CI(k) under CI, let A1 = A2 = … = A be the amplitude and τ1 = τ2= … = B denoting the phase offset with respect to the first signal generated by transmitter i = 1. Consider one IEEE 802.15.4 standard based communication system, RSSICI(k) is equal to:
( )1
20log cos cC i
k
I k ii
RSSI A
Where ωc is a constant and τ1 =0
D1 CR
CR
D3 CR
SP
S1
R
S3
Period 1Collision
D2 CRS2
SP D1
Period 2Contention
CP1
CP2
CP3
D2SP
SP
SP
D3
D1
Period 3Data transmission
Time
SP
SP .........
... RSSI ValueD2 SP
SP
D3
CR Contention Request
CP Contention Packet
SP Schedule Packet
D Data
Design of STAIRS• Overview
7
Through intentional contention, senders can be identified from the stair-like pattern of RSSI in one round.
8
Design Challenges• Challenge 1: Synchronization
– Requirement of CI: Δ≤0.5μs• Challenge 2: Falling edge detection
– CP packets with the same length– External interference, e.g., WiFi signals
CP1CP2
CP3
(1) False negatives
False falling edges
(2) False positives
Alignment for CP packets• Receiver-initiated (CR)
– Triggering transmissions of CP packets– Serving as ACK/NACK– Coping with hidden terminals
• Parallelizing receiving and reading
10
10
S-CUSUM Edge Detection• Discrete lengths of CP packets
– Total sender number N, maximum packet size L, increase step ΔL, length of CP is:
• A paradox- how to find a good ΔL?
, 2 ,3 ...l CP L L L m L
Less false edges
Larger CP space
11
• Finding the optimal ΔL– p=1/m: choose any of the m lengths– α: the probability of false positives
• Three cases for a schedule:
1
(1 )
(1 )
1
Ni
Ns
c i s
P p
P Np p
P P P
argmax , ,opti s c
L
L f P P P
12
Implementation• STAIRS
– A plug-in between APP and MAC layer– Invoked when collision happens– Three main components
13
Evaluation• Micro-benchmark
– Synchronization – Edge detection
• Multi-hop testbed– Completion time– Efficiency– Duty cycle
• Large-scale simulation
14
Micro-benchmark
• Offset among arriving packets less than 0.25 μs!• S-CUSUM increases detection efficiency.• Average detection accuracy is > 85%.
15
Testbed Settings
• Multi-flow-multi-hop environment.• ΔL is set to 10 bytes.• 20 TelosB sensor nodes
Flow number
16
Compared with Strawman
• STAIRS beats Strawman, especially with large number of senders.
• Contention overhead of STAIRS is amortized.
17
Duty Cycle Evaluation
• Both sender and receiver duty cycles are improved, as contention time is reduced.
• Energy efficiency is therefore improved.
19
Results
• No degraded performance• CSMA-L beats CSMA-E after the threshold• Backoff time dominates!
Pkt_size = 50 bytes Pkt_size = 100 bytes
20
Summary• Collision resolution with active contention • Observing the RSSI stair-like pattern, we then
look into its principle• Design STAIRS based on the stair pattern and
solve challenges like synchronization and finding optimal ΔL
• Evaluation in both real testbed and large-scale simulation