MobiSteer: Using Steerable Beam Directional Antenna for Vehicular Network Access

1
MobiSteer Using Steerable Beam Directional
Antenna for Vehicular Network Access

• V. Navda, A.P. Subramanian, K. Dhanasekaran,
  A. Timm-Giel and S.R. Das
• Stony Brook Univ., NY, U.S.A.
• Univ. of Bremen, Germany
• Best Paper Award, MobiSys 2007
• Slides and Presented by Yong Yang, UIUC
• Sept 20, 2007

2
Motivation

• Performance of vehicular access of roadside APs
  is still poor 1
• Median connection duration 13s
• Delivery rate 80
• Goal to improve the duration and quality (data
  rate or SNR) of connectivity to APs.

1 V. Bychkovsky, B. Hull, A.K. Miu, H.
Balakrishman, and S. Madden. A Measurement Study
of Vehicular Internet Access Using In Situ Wi-Fi
Networks, MobiCom06 Best Paper
3
Basic Idea

• Use directional antenna
• Continuously steering the beam appropriately as
  the vehicle moves
• Handoff to another AP when better beam is
  available

4
Overview

• Hardware Setup
• Software Setup
• AP and Beam Selection Algorithm
• Two modes cached or online
• Experiments
• In controlled environment
• In in-situ environment

5
Hardware Setup

• Steerable Phocus Array antenna
• Omni-directional beam
• 16 directional beams (45 width, 22.5
  overlapping with neighbors)
• Use 8 non-overlapping beams

6
Hardware Setup (contd.)

• Soekris net4511 board
• 100/133Mhz AMD proc, 64MB SDRAM
• 802.11 a/b/g miniPCI card based on Atheros
  chipset
• USB-based Garmin GPS receiver
• APs are Soekris net4826 router boards with 802.11
  miniPCI cards and omni-directional antennas

7
Software Setup

• Embedded computer runs pebble Linux
• Madwifi driver for 802.11 interface
• Implement ioctl() for controlling the antenna
  beaming
• Create a virtual interface in monitor mode
• Modify a packet sniffer software, Kismet
• Capture packets received on the virtual interface
• Communicate with a GPS daemon
• Stamp received packets with current time, GPS
  coordinates, beam pattern, channel, PHY rate and
  SNR.
• Log this information as the RF Signature Database

8
Operation Overview

• Cached mode
• when driving in a familiar route
• Use RF signature database to determine optimal
  beam steering and AP selection
• Online mode
• When driving in a previously untraveled route
• Probe the environment and choose the best AP and
  Beam combination based on probing responses.

9
Operation Overview (contd.)
10
Cached Mode Operation

• Discretize the route into segments of length ?
• All RF tuples in the same segment are analyzed to
  compute the best AP and beam combination
• Make selection at the start of a segment i
• Should we hand off from AP k to AP j ?
• Suppose
• Si is the current speed
• h(k, j) hand-off latency from AP k to AP j
• r(i, j) bit rate when associate with AP j in
  segement i
• So useful time t(i,j,k) ?/Si - h(k, j)
• Throughput is r(i, j)t(i, j, k)

11
Cached Mode Operation (contd.)

• Dynamic programming to compute the best AP for
  each segment such that the aggregate throughput
  is maximized
• If best(N, m) is highest for the last step, then
  we trace backaward, selecting prev(N,m) for
  segment N-1, prev (N-1, prev(N,m)) for segment
  N-2,

12
Cached Mode Experiments

• Controlled Scenarios
• In-situ Scenario
• Urban roadways near Stony Brook Univ.

A large empty parking lot
Graduate student apartment complex
13
Parking Lot

• 10 drives on the same route
• 8 such sets of drives on different days
• Location of the peak performance shifts as beams
  are changed
• Omini beam has modest performance
• MobiSteer outperforms others

? 5m, low speed
14
Apt Complex

• Similar result, MobiSteer outperforms others
• Because of multi-path effect
• Shift in the peak point is not that clear
• All beams offer similar performance

15
In-Situ Experiment

• Drove about 5km around campus on different days
• Average speed it 30-40 miles/h and ? 40m
• Discovered 307 unique APs

16
Online Mode

• 1. Probe over all beams and channels
• 2. Associate with APi if ltAPi, beamjgt has highest
  SNR
• 3. After d consecutive packet drops, switch to
  the next best ltAPi, beamkgt
• 4. If no other beam for APi, or the last beam
  failed to transmit all packets, choose next best
  AP
• 5. If no other AP available, restart

17
Apt Complex Experiment

• 39 less packets are received comparing to cached
  mode
• Probing takes 3080ms
• sub optimal since combinations are discovered
  during probing instead of communication

18
In-situ Experiment

• Online mode performs better sometimes due to
  availability of fresher channel estimates
• But in general, scanning all beams and channels
  incurs a large penalty

19
Conclusions

• AP and beam selection algorithms
• How to determine if on the familiar route and for
  how long
• Determine ?
• Short for low speed good granularity for
  optimization
• Long for high speed enough time for probing and
  scanning
• Experiment in controlled and in-situ scenarios
• Use the distance for which AP is heard as the
  measure for connectivity duration

20
Some Other MobiSys07 Papers

• Real-Time Deployment of Multihop Relays for Range
  Extension
• M. Souryal et al (National. Inst. Of Standard and
  Tech.)
• Shooter Localization and Weapon Classification
  with Soldier-Wearable Networked Sensors
• P. Volgyesi et al (Vanderbilt Univ.)
• Context-for-Wireless Context-Sensitive
  Engery-Efficient Wireless Data Transfer
• A. Rahmati et al (Rice Univ.)
• Wireless Wakeups Revisited Energy Management for
  VoIP over Wi-Fi Smartphones
• Y. Agarwal et al (UCSD, Microsoft Research)
• MobiUS Enalble Together-Viewing Video Experience
  across Two Mobile Devices
• G. Shen et al(Microsft Research Asia), Best Demo
  Award