Improving Vehicular Networking Reliability and Efficiency in the Context of Platooning Applications

Song Gao

Committee:

Alvin Lim (Chair), Saad Biaz, Hari Narayanan, Xiao Qin

Final Oral Exam

University Reader:

Shiwen Mao

Overview

Background

Squirrel: A Wireless Emulator

Interframe Compression Transmission Layer

Future Work

Conclusions

Contributions

  • A wireless emulation infrastructure for testing real-world networking applications and protocols
  • An inter-frame compression method for reducing bandwidth consumption in vehicular networks
    • An adaptive ICTL that achieves 50% bandwidth reduction
    • Reduced bandwidth consumption helps increase delivery ratio

Overview

Background

Squirrel: A Wireless Emulator

Interframe Compression Transmission Layer

Future Work

Conclusions

DSRC Protocol Suite

5.9 GHz / 10 MHz

OCB

1 CCH + 6 SCH

Vehicle Platooning

(Cooperative Adaptive Cruise Control)

CC

ACC

CACC

Vehicular Platooning

(Cooperative Adaptive Cruise Control)

Vehicular Platooning

ACC

CACC

Following Distance 6 feet
Speed 45 mph
Time Interval 0.1 second

Real-World DSRC in Platooning Scenarios

  • Two trucks with trailers installed
  • Dual antenna: one on each side of tractor
  • Front -> Back (broadcast), delivery ratio measured

Real-World DSRC in Platooning Scenarios

Delivery ratio can be easily affected:

  • Front vehicle turning: outside antenna is blocked by the vehicle body and results in poor delivery ratio
  • On straight road: some terrain can affect delivery ratio significantly
  • On hilly road: vehicle needs to rely on terrain’s reflection because antennas are not parallel

Using both antennas alternately significantly increase (pairwise) delivery ratio in all scenarios

  • >= 90% for up to 6 Mbps

Overview

Background

Squirrel: A Wireless Emulator

Interframe Compression Transmission Layer

Future Work

Conclusions

Algorithm and Protocol Validation

Formal Analysis

Experimental Study

Real-world Experiments

Simulation

Emulation

Real-world Experiments

Advantages:

  • Authentic Results
  • Closest to Production Environment

Disadvantages:

  • High cost (hardware, human hours)
  • Not scalable
  • Solution needs to be sufficiently mature

slow vehicle

Simulation

Advantages:

  • Low cost
  • Extremely scalable
  • Support arbitrary maturity level

Disadvantages:

  • Accuracy is determined by model quality
  • Not deployable to real-world

Emulation

Real-world Components

  • Good balance between real-world experiments and simulations
  • Hard-to-scale part is emulated (models)
  • Non-emlulated part is deployable to real-world

Emulated Models

Challenges:

  • Good system design
  • Proper channel interference modeling

Squirrel Concept

Squirrel and DSRC

Squirrel Architecture

CSMA/CA Model

Challenge:

  • Modeling channel occupation and interference properly

Example: 1. Hidden node problem; 2. Exposed node problem

CSMA/CA Model

T_{data}=t_{busy}+t_{DIFS}+t_{BO}\times aSlotTime+t_{data}
Tdata=tbusy+tDIFS+tBO×aSlotTime+tdata T_{data}=t_{busy}+t_{DIFS}+t_{BO}\times aSlotTime+t_{data}
T_{ACK}=t_{SIFS}+t_{ACK}
TACK=tSIFS+tACKT_{ACK}=t_{SIFS}+t_{ACK}

CSMA/CA Model

Solution:

  • One leaky bucket per node
  • Water -> time durations
  • "Water" is added whenever channel used around the node
  • Drop frames when bucket is full

RPC Updated Mobility Manager

  • Position Manager exposed as RPC Service
  • Position Updating delegated to a different process

Effectiveness Test

Same test on Android tablets and squirrel emulated nodes:

Android:

  • 4 Nexus 7 tablets with modified Android kernel
  • placed next to each other
  • iperf UDP throughput test:
    • node 1 to node 2 from 0th sec until 30th sec
    • node 3 to node 4 from 10th sec until 20th sec

Squirrel:

  • 4 emulated nodes running Linux networking stack
  • emulated distance set to 150, with interference range set to 300k
  • iperf UDP throughput test:
    • same test design

Effectiveness Test

Throughput Test

  • Deployed on a single workstation
    • 6-cores
    • no hyper threading
  • 12 (6 pair) of emulated nodes
  • iperf UDP traffic
    • 500 Mbps demanded

Throughput Test

Related Work

  1. Dynamic Switch [20]: impossible to support interference modeling;
  2. EMANE [21]: very similar; complex and no vehicular networking support.

[20] T. Lin, “Mobile ad-hoc network routing protocols: methodologies and applications,” Ph.D. dissertation, Virginia Polytechnic Institute and State University, 2004.

[21] U.S. Naval Research Lab. Extendable mobile ad-hoc network emulator (emane). [Online]. Available: http://www.nrl.navy.mil/itd/ncs/products/emane

Overview

Background

Squirrel: A Wireless Emulator

Interframe Compression Transmission Layer

Future Work

Conclusions

DSRC Congestion

  • Uses 10 MHz channels
  • Needs 2 antennas
  • Longer distance -> bigger interference range
  • Safety applications require continuous real-time data

DSRC will have congestion problem once fully deployed

Case Study: Busy Interstate Segment

Average Following Distance 20 meters
Platooning Message 150 bytes
Number of Lanes 5 per direction
Interference Range 400 meters
MAC Layer Data Rate 6 Mbps
Platooning Message Rate 20 Hz

(I-285)

Assumptions:

Case Study: Busy Interstate Segment

Interference distance: 400 * 2 = 800m

# of cars generating interference per lane: 800m / 20m = 40

# of cars generating interference: 40/lane * 10lanes = 400

Demanded message rate: 20 Hz * 400 = 8,000 Hz

Demanded chan. bandwidth: 8KHz * 150B = 1.2MB/s (9.6 Mbps)

9.6 Mbps >> MAC data rate (6 Mbps)

plus channel switching, CSMA overhead, etc.

Interframe Compression in Video Codec

Introduced in video compression algorithms

Assumptions:

  • Consecutive frames are similar
  • Minor image degradation doesn't affect video quality

Core Ideas:

  • I-Frame: complete frame
  • P-Frame: forward prediction (much smaller)
  • B-Frame: bi-directional prediction (much smaller)
  • Encoding: IBBPBBP...PBBI

Interframe and Platooning

Assumption:

  • Consecutive messages are similar
  • Minor error is not acceptable
  • Transport is usually unreliable

Video interframe compression cannot be applied directly.

Solution:

  • Replace lossy prediction frames with lossless differential frames
  • Less differential frames between complete frames

Interframe Compression Transmission Layer

  • Between applications and WSMP layer
  • Automatically handles encoding and decoding
  • Two frame types: Key Frame & Differential Frame
Reserved Frame Type Compression Options Frame ID Payload
4 bits 4 bits 8 bits 16 bits variable

Frame Layout:

ICTL Differential Frame

Previous:

Current:

Differential:

Lower entropy, more compressible

XOR

ICTL Encoding & Decoding

Prerequisite: Every node needs to keep a list of recently received Key Frames

Encoding (Transmitter):

  • Key Frames:
  1. Add frame to recent KFs list
  2. Prepend ICTL header
  • Differential Frames:
  1. Construct the DF (XOR) from one of recently sent KFs
  2. Compress DF
  3. Prepend ICTL header

Decoding (Receiver):

  • Key Frames:
  1. Parse ICTL header
  2. Add the KF to recent KFs list
  • Differential Frames:
  1. Parse ICTL header
  2. Decompress DF
  3. Construct original message using the DF and referenced KF

ICTL Implementation and Experimental Setup

  • Platooning Trace File Playback
    • GPS locations -> RPC updated with squirrel
    • DSRC transmission -> goes through squirrel
  • Deployed on a single workstation
    • CoreOS / Rkt
  • Trigering Congestion
    • Replicated messages
  • Implemented as a simple interface ictl.Endpoint
    • Encode() / Decode()

ICTL First Glance

ICTL Performance Studies

  • Cycle Length and Application Delivery Ratio
  • Location and Bandwidth
    • curved roads vs straight roads
  • Payload Compression
  • Cycle Length and Bandwidth

Cycle Length and Application Delivery Ratio

  • ICTL Cycle:
    • [KF-DF-DF]-[KF-DF-DF]
  • Cycle Length (CL)
    • # of frames in a cycle. e.g. CL([KF-DF-DF])=3
  • Application Delivery Ratio (ADR)
    • delivery ratio at application layer (above ICTL)
  • Network Delivery Ratio (NDR)
    • delivery ratio below ICTL

Definitions:

Cycle Length and Application Delivery Ratio

With in an ICTL Cycle, [KF-DF-DF-...-DF]:

  • At probability of 1-NDR, the KF is lost:
    • expected # of deliveries in ICTL Cycle: 0
  • At probability of NDR, the KF is delivered:
    • expected # of deliveries in ICTL Cycle: 1+(m-1)*NDR
E=(1-NDR)*0+NDR*[1+(CL-1)*NDR]=(NDR)^2*CL-(NDR)^2+NDR
E=(1NDR)0+NDR[1+(CL1)NDR]=(NDR)2CL(NDR)2+NDRE=(1-NDR)*0+NDR*[1+(CL-1)*NDR]=(NDR)^2*CL-(NDR)^2+NDR
ADR=\frac{E}{CL}=(NDR)^2+\frac{NDR-(NDR)^2}{CL}
ADR=ECL=(NDR)2+NDR(NDR)2CLADR=\frac{E}{CL}=(NDR)^2+\frac{NDR-(NDR)^2}{CL}

Cycle Length and Application Delivery Ratio

ADR=(NDR)^2+\frac{NDR-(NDR)^2}{CL}
ADR=(NDR)2+NDR(NDR)2CLADR=(NDR)^2+\frac{NDR-(NDR)^2}{CL}

Location and Bandwidth

Does ICTL use more bandwidth at curved roads?

Payload Compression

Does it make sense to try each compression algorithm for each individual frame?

Cycle Length and Bandwidth

Longer CL != Less Bandwidth

Adaptive ICTL

  • Global Optimal: infeasible
  • Based on local metric: average frame size
\mu_{cycle}^n=\frac{S_{KF}+\sum^n_{i=2}{S^i_{DF}}}{n}
μcyclen=SKF+i=2nSDFin\mu_{cycle}^n=\frac{S_{KF}+\sum^n_{i=2}{S^i_{DF}}}{n}

Adaptive ICTL

Adaptive ICTL Bandwidth Consumption

Adaptive ICTL ADR: 4 nodes

Adaptive ICTL ADR: 6 nodes

Adaptive ICTL ADR: 10 nodes

Adaptive ICTL ADR: 12 nodes

Adaptive ICTL ADR: aggregated

ICTL Latency

Read 1MB sequentially from memory 0.25 ms
Round trip within same data center 0.5 ms
ICTL < 1 ms
DSRC On-board WSM normally 1ms ~ 5ms
Round trip ICMP ping to AU_WiFi AP ~ 2.5 ms

Related Work

[34] Z. Wang and M. Hassan: “How much of dsrc is available for non-safety use?” VANET ’08

[35] C.-L. Huang, Y. Fallah, et. al.: “Adaptive intervehicle communication control for cooperative safety systems,” Network, IEEE, 2010.

[36] Y. Fallah, C. Huang, et. al.: “Congestion control based on channel occupancy in vehicular broadcast networks,” VTC '10

non-safety use of DSRC may have to be severely restricted during peak hours of traffic.

adapts communication rate and power based on the dynamics of a vehicular network and safety-driven tracking process. 

a closed loop congestion monitoring and control based on limited feedback from the network.

[38] S. International, “SAE J2945.1 On-Board System Requirements for V2V Safety Com- munications,”

inspects channel busy percentage and packet error rate, to make transmission decisions, including transmission schedule and power

Overview

Background

Squirrel: A Wireless Emulator

Interframe Compression Transmission Layer

Future Work

Conclusions

Conclusions - Emulation Infrastructure

  • Squirrel - Wireless Networking Emulation Framework
    • Replaces MAC/PHY with models
    • For testing real-world software components
    • Configurable parameters and models
    • Throughput up to 1200 Mbps on single workstation
  • CSMA/CA Model
    • Properly model interference with leaky bucket
    • UDP Throughput test show same trend as tested on Android devices
  • RPC Updated Mobility Manager

Conclusions - ICTL

  • Interframe Compression Transmission Layer
    • Two frame types: KF & DF
    • A KF is followed by a number of DFs, which encode difference between consecutive messages
  • Behaviors of ICTL are studied
    • ADR mainly affected by NDR
    • Bandwidth consumption not sensitive to curved roads
    • Longer ICTL != Less Bandwidth Consumption
  • Adaptive ICTL
    • Bandwidth optimization with average frame size within an ICTL Cycle
    • Reduces 50% bandwidth consumption
    • Improves delivery ratio in congested networks

Overview

Background

Squirrel: A Wireless Emulator

Interframe Compression Transmission Layer

Future Work

Conclusions

Rebroadcasting?

  • Platooning applications need rebroadcasting:
    • Control messages need to reach whole platoon
    • Large platoons may be too long for one-hop delivery
      • end vehicle is too far from head vehicle
      • packet loss due to traffic and terrain
    • WSMP does not provide routing
  • Useful for disseminating emergency messages

Multipoint Relays

(Introduced by Optimized Link State Routing protocol)

  • MPR nodes are selected one-hop nodes that rebroadcasts routing messages, reducing flooding overhead
  • Broadcasting reaches all 1-hop neighbors; MPRs cover all 2-hop neighbors
  • Discovery: Hello messages
    • Hello -> 1-hop and 2-hop neighbors

MPRs in Vehicular Networks

Different Assumptions:

  • GPS positions are available all the time
  • Vehicle dynamics data is available
    • mobility is predictable

New MPR Design

  • BSM
    • Discover 1-hop neighbors
    • Estimate temporal communication range
  • Hello messages - Removed

2-hop neighbor set: calculated from distance between nodes and estimated communication range.

New MPR Design

Predicting Topology Change:

  • Predicted trajectory of each vehicle
  • Intention-based prediction from platooning messages

Prediction Guided MPR Selection:

  • Predict more suitable MPRs beforehand. e.g.,
    • vehicle with same speed is changing lane, moving towards to the platoon
  • Change MPR before topology breaks. e.g.,
    • vehicle leaves platoon
    • vehicle in next lane accelerates

Q/A