Voice over the Dins: Improving Wireless Channel Utilization with Collision Tolerance Xiaoyu Ji Xiaoyu Ji, Yuan He, Jiliang Wang, Kaishun Wu, Ke Yi, Yunhao.

Voice over the Dins:Improving Wireless Channel Utilization with

Collision Tolerance

Xiaoyu Ji, Yuan He, Jiliang Wang, Kaishun Wu, Ke Yi, Yunhao Liu

ICNP, 2013, Göttingen

Tsinghua UniversityHong Kong University of Science and Technology

2

Motivation• Collisions in wireless networks– Broadcast nature and lack of collision detection– Event-driven mode (WSNs)

• Network performance, e.g., channel utilization is harmed

• Researchers have designed a lot of medium access control protocols (MAC)– A-MAC, B-MAC, …, Z-MAC

3

The long-held philosophy• Collision avoidance– Avoid simultaneous access from multiple users to

a shared channel– Conservative strategy against collisions– Side effect: idle slots, as a result of random

backoff

Collision avoidance limits the wireless channel to be better utilized!!!

4

Why not more aggressive?• Collision tolerance– Allowing collisions to happen, i.e., the overlap of

signals in time domain – Potential gain in channel utilization

A B

C DR

AB

BC

DBA

CB

DC

BC

CB

DC

DBA

AB

DB

DCB

BC

DBA

CB

4 senders with a common receiver

A B D C B D A

Tolerance

Avoidance

Collision tolerance enables one to approach the upper limit of channel utilization!

Formulation• Suppose N senders choose probability p to

transmit to the receiver, then three possible states for any given time slot:

– Idle

– Successful

– Corrupted

• And the utilization ratio:5

The potential gain• For backoff-based approaches, the success

probability is：

• The simulation results• CT- Collision tolerance• CA- Collision avoidance• η: parameter related to

Ts , Tc , Tslot

20% improvement in general case!

6

7

Feasibility of collision tolerance• PHY layer implementation– Redundancy– Stronger signal dominates the weak ones– Capture effect1,2, message in message (MIM)3

• Therefore, even packets collide, the strongest one could still be correctly received under certain conditions!

1, The capture effect in FM receivers, KRIJN LEENTVAAR and JAN H.FLINT, IEEE TOC, 19762, Sniffing out the correct physical layer capture model in 802.11 b, MLA Kochut, Andrzej, et al. , ICNP, 20043, Order Matters: Transmission Reordering in Wireless Networks, Justin Manweiler etc, Mobicom, 2009

8

Investigating collision tolerance (1/2)

• Experiment setup– Indoor, TelosB nodes, Contiki-OS– 2 senders with a common receiver

1. Timing requirement

160μs• Case 1: Strong first. The receiver receives

the strong signal without knowing the weak ones.

• Case 2: Weak first, strong coming within the preamble window of the weak ones.

• Case 3: Weak first, strong coming out of the preamble window of the weak ones.

Strong

Weak

P

P

P

P

Case 1

P

P

Case 3

P

P

Case 2

9

Investigating collision tolerance (2/2)2. Concurrency requirement

• C(k): concurrency of 2, 3 even 4 are beneficial!• Especially, the collision probability with

concurrency 2 is very small.

All packets should come within 160μs;

Concurrency should be proper.

Protocol overview (Coco)• Timing requirement– ACK triggers the transmission

• Concurrency requirement– Transmitting with probability p

9

ACK-triggered transmission• ACK– Triggering transmissions (synchronization)– Piggybacking the probability p– Coping with hidden terminals

The offset can be controlled within 1 μs as shown in the evaluation.

10

12

Transmitting with p • Optimum: How to decide whether p is proper

or not?• Adjustment: If not, how to adjust it?• Convergence: How well can the adjustment

converge?• Errors: The error involved in the adjustment.

Optimum: Is p proper?• The utilization is a function of N and p, and under

some N, the best p should be:

• Claim: when p is proper, which means Util(p) achieves its best value, Pc, Pi, Ps and Util(p) converge when N ∞

• The criteria 12

Adjustment of p• The feedback control algorithm

• Sliding window: recording status of past packets, to calculate Pc

• The dichotomous algorithm

13

Convergence speed of p• p is in the range of [0:1]• The resolution is 0.001 when N=20 (Tab.1)• It takes at most 10 iterations to regulate p

from initial 1 value to 0.001.• With 100 data packets in each iteration, a

maximum 4.096 ms for each packet, the maximum time is 4.096 s

14

Error in the process• The error introduced by ε: ΔUtil(p)• For different ε, we calculate the errors:

15

Evaluation• Metrics – The accuracy of timing– The adjustment algorithm– Overall performance

• Setup– 21 TelosB nodes, Contiki-OS– Single hop, single receiver

16

18

Timing accuracy• 11 senders with a common receiver• The offset between the tested sender and a

reference sender at receiver side

The offset among senders can be as small as nano seconds!

The feedback control algorithm

With N changing dynamically, p can converge to its best value quickly!

18

Overall evaluation (1/2)• Comparing with B-MAC, default in Tiny-OS• Performance evaluation with different

network parametersParameterEnvironment Indoor: Hall, Testbed

Outdoor: groundPacket length 20, 60, 100 (bytes)Topology Line, circle random

19

Overall evaluation (2/2)• Packet length: longer packets are better.• Topology: resistance to hidden terminals.• Compared with B-MAC (linear and exponential)• Avg. 20% improvement in general case.

20

Summary• Question the avoidance principle and improve

channel utilization with collision tolerance• Investigate the sufficient condition in

achieving beneficial collision tolerance• Design a protocol to exploit collision tolerance

for improving channel utilization• Real implementation and extensive evaluation

21

THANK YOU!

Xiaoyu [email protected]

22