1. A Measurement Study for Link Rate Modeling and Handover Optimization for Vehicle Communications over IEEE g Infrastructure Network
Kuang-Ching Wang, Rahul Amin, Michael Juang
Department of Electrical and Computer Engineering
Clemson University
Clemson, SC USA
Bastian Migge
Information Technology Research Center
BMW Group
Greer, SC USA
Abstract
Test Scenarios
Scenario 3: Driving Experiment
Figure 3. Throughput with AP min (1 mW) and max (30 mW) power at 20 mph
Figure 4. Throughput withRSSI at 20 mph and 40 mph
Persistent communication has become increasingly important to modern automobiles in today’s heavily utilized road systems.
Currently used satellite or cellular technologies do not provide sufficient bandwidth to cope with increasing vehicle information services.
With the recent deployment of IEEE based municipal network infrastructures, measurement studies have been conducted to assess the following scenarios:
1) IEEE b radios
2) network association and transport layer throughput performance
3) have not studied handover for persistent end-to-end communications.
Further work needs to be done to determine:
1) an accurate mobility aware link rate and quality model
2) the handover behavior across a series of access points (APs)
3) strategies for link rate and handover optimization during vehicle
mobility.
This poster presents results from a measurement study conducted to address these challenges using IEEE g data network.
Scenario 1: Mobile client was stationary at selected locations on the test track.
Scenario 2: Mobile client walked through the track at controlled speeds.
Scenario 3: Mobile client was on the car’s front passenger seat with the car
driven through the 1200 feet test zone at constant speeds of 20 and 40 miles per hour (mph), respectively.
For each scenario,
Mobile client’s transmit power was 16 dBm (40 mW).
Maximum and minimum AP transmit powers (30 mW and 1 mW respectively) were both studied. Graphs displayed are for 1 mW tests unless otherwise indicated (Figure 3).
UDP upload, UDP download, TCP upload, and TCP download experiments were performed with iperf v on the mobile client initiating transfers with the iperf server on the Ethernet backhaul. The data payload of the packets was 1500 bytes.
Wireshark v set up on mobile client monitored all packet transmissions (in Channel 1) via the external Cisco interface.
The received signal strength indication (RSSI), per transmission link rate, and beacon loss from the associated AP were recorded.
Throughput profiles for selected tests are shown in Figure 3.
1) Handover only occurred for 1 mW AP TX power runs
2) After handover, throughput improved before falling again
Figure 4 examines 20 and 40 mph cases, showing RSSI as well.
1) No handover to AP 2 at 400 ft., so low throughput for middle
2) Handover delayed for 40 mph because scanning duration
equal for all speeds, so the vehicle covers more ground
Experiment Setup
The study was conducted on an IEEE g network established at the BMW high speed test track facility at Spartanburg, SC.
Network consisted of three Cisco 1200 series IEEE a/b/g pole-mounted access points (APs) placed 6 feet above ground at 400 feet distance intervals along the test track. They were connected with a common Ethernet backhaul using 3 Cisco Catalyst 2950 switches, one of which acts as a DHCP server.
The APs were configured with the same ESSID and a common channel (Channel 1).
The mobile client was an IBM Thinkpad T60 laptop running Ubuntu 7.04 Linux, a built-in Intel Pro/Wireless 3945ABG Wi-Fi interface (driver: ipw3945 v1.2.1) for data transfer, and an external Cisco Aironet ABG PCMCIA Wi-Fi interface (driver: Madwifi v ) for link events monitoring.
A second laptop was connected to the backhaul as a server for iperf.
Figure 1 below illustrates the network setup.
Figure 1. Organization of vehicle IEEE g infrastructure network testbed (not to scale).
Throughput and link rate (Figure 5)
1) Throughput follows curve of
individual transmission link rate
2) Throughput much lower than link
rate, caused by packet loss and
retransmission
Scenario 1: Stationary Experiment
1) Throughput, RSSI, and link rate decreased monotonically with increasing
distance from the associated AP
2) Handovers occurred only after severe degradation of performance, when
moving between test locations.
The ipw3945 driver initiates handovers under two conditions:
1) 8 consecutive missed beacons:
initiates roaming, sends probe
request on the current channel.
If a valid AP with higher RSSI
than current network is found,
association is attempted
2) 24 consecutive missed
beacons: disassociates with
current AP and probes in all
channels for APs with any
ESSID and a higher than
current RSSI to associate with.
Scenario 2: Walking Experiment
Figure 5. Link Rate and UDP Upload Throughput at 20 mph
AP locations from start of trial (Figure 2):
AP 3: 0 ft. AP 2: 400 ft. AP 1: 800 ft.
1) Achieved throughput decreased with
decreasing RSSI and distance from
connected AP
2) No handoff, retained association with
AP 3 throughout entire experiment
3) Higher RSSI than for driving experiment
because no vehicle body to block signal
Figure 2. Walking Throughput and RSSI
Figure 6. Missing Beacons
Conclusions
Both procedures are depicted in Figure 6 as the roam failed to find a better AP. In other trials the roam was successful.
Acknowledgements
The current wireless driver is ill-suited for providing mobility. In all of the experiments the handover mechanism was too passive. The client did not switch to another AP with much higher RSSI until the overly high missed beacon thresholds were hit. If some time associated with an AP has been utilized for more aggressive scanning, the overall link quality would likely be better.
Achieved throughput after handovers was much greater than throughput before the handover. But then throughput quickly declined again as the newly associated AP was passed.
Given the dynamic and quickly changing wireless environment at vehicular speeds, special care should be taken in evaluating how aggressive and proactive link-layer adaptations should become.
Future Work
The authors would like to thank the following people who have helped in conducting the experiments by providing networking equipment, test track facilities, and building the IEEE g network testbed:
Clemson Computing and Information Technology Staff
Brian Conner, BMW Group IT Research Center
Benjamin Lippert, BMW Group IT Research Center
Steffen Saplata, BMW Group IT Research Center
Florian Preis, BMW Group IT Research Center
Adrian Steinemann, BMW Group IT Research Center
Acquire sufficient insight of the IEEE g radios’
link rate dependency on the distance and speed of a moving vehicle
handover dynamics across a series of infrastructure APs that provide continuous coverage.
Construct quantitative link rate models based on client-AP distances and relative speeds for studying the performance, capacity provisioning, and handover strategies for vehicle infrastructure communication networks.