WiNGS Group's Research Snapshot (Oct 2011)

My group's research in the recent years has focused on various aspects of mobile and wireless systems. Our approach has been to (i) understand the wireless environment better and designing better tools to improve our understanding of the network; (ii) design techniques to improve wireless-based data delivery, and (iii) create new infrastructure-based services to combine lessons learnt above. Here are some examples.

List of Publications

Understanding wireless interference WiScape WiRover Apex

Understanding wireless interference

Understanding interference in a wireless environment is a challenging problem. Our work in this domain started with asking a basic question in WiFi environments --- can we tell apart the reason for a packet loss, e.g., is it due to a weak signal or is it due to a collision? We defined a post-mortem approach to diagnosing the cause for a packet's loss, called COLLIE, which was the first experimental attempt to distinguish the root cause of packet losses. Since then, we have looked at a range of problems that try to infer in real-time which other WiFi as well as non-WiFi sources are causing interference to users in a WLAN environment. The following papers capture some of the main results until date.

Airshark + WiFiNet: How can we detect the existence of non-WiFi transmitters in the unlicensed spectrum, using only off-the-shelf WiFi cards? Examples of non-WiFi transmitters are Bluetooth gadgets, ZigBee units, various game controllers (PS2, Xbox, etc.), analog phones, frequency hopping phones, and even microwave ovens. Past work and various commercial products do this but often use more sophisticated spectrum sensing chips or specialized embedded hardware (e.g., WiSpy). If we do this without using any such additional hardware, then every WiFi Access Point (AP) and client can do this in software we can build great interference awareness in WiFi APs and clients. We designed a technique, called Airshark, for doing this using Atheros 9280 WiFi chipsets, all in software.
[IMC 2011 paper], [YouTube video]
Media coverage: Network World, Slashdot, CRA Highlight of the week, PC Magazine, The Register, BoingBoing, ...

Many wireless devices occupy spectrum.
Our continued work in this domain, called WiFiNet, provides even deeper analysis of non-WiFi transmitters in two ways still using off-the-shelf WiFi cards: (i) it can now quantify the impact each individual non-WiFi interferer has on WiFi traffic, including when there are multiple devices of the same or different type For example, there may be two Bluetooth headsets in operation and one analog phone, WiFiNet tries to separate the contribution of each such device on how much it impacts specific WiFi users (maybe 10%, 12%, and 25% respectively); (ii) it can localize the position of this non-WiFi interferer. The core challenge we have solved in both these cases is how to meet both these goals using only WiFi cards, and hence with the constraint that our system cannot decode the transmissions from these non-WiFi devices in the air.
Paper and video coming soon.

Locate Non-WiFi devices in space.

  • PIE (Passive Interference Estimator): This work demonstrates an online system which captures all WiFi to WiFi interference in an enterprise WLAN as and when it happens. We recognize that interference changes with various communication parameters, e.g., transmit power, PHY rate, packet size, etc., and PIE is able to detect occurence of interference based on current operational choices made by the system. PIE reacts well to mobility --- as users move through the enterprise, the interference patterns change, and PIE picks it up immediately.
    [NSDI 2011 paper], [YouTube video]

    FLUID: This work shows efficient ways to assign flexible channels (channels with arbitrary center frequencies and widths) in a multi-link scenario. We show that even in a simple two link scenario, and with two width choices (say, 20 MHz and 40 MHz) and a single center frequency, the best width assignment varies significantly based on the scenario. This is because altering channel widths alter interference significantly. Some links become hidden terminals and others can be exposed terminals simply due to change in a channel width. The work proposes a systematic model for that can be used to efficiently reason how different flexible channels can be assigned to multiple WiFi transmitters in range, and how such choices impact interference.
    [MobiCom 2011 paper]
    This paper was a best paper nominee at MobiCom 2011, and was one of three papers fast-tracked to the Transactions of Mobile Computing.

  • COLLIE (Collision Interferencing Engine): This work defines a technique which can tell apart whether a (WiFi) frame was lost due to a weak signal or due to collision from another wireless transmission. The key constraint in this work was to use off-the-shelf WiFi cards only (as was available at the time) and no additional software capability. Hence, we were able to examine bit error patterns to achieve this. The work does a fast post-portem of erroneous receptions. A number of subsequent work has since increased the toolkit to tell apart these two common causes of packet losses, although most assume additional support from the PHY layer.
    [Infocom 2008 paper]

    WiScape: Monitoring Wide-area Cellular Networks through Client Assistance

    A large number of clients can collaboratively provide a unique view of performance across space and time for wide-area wireless (cellular) networks. The naive approach would require a measurement server requesting each client to collect a lot of measurements. Clearly, this is inefficient and resource intensive. The WiScape system maps out the wireless landscape of a cellular network by figuring out what is a small amount of measurement needed to get a good, coarse-grained understanding of the network's overall user experience. We partition required measurements into time and space and collect a small number of measurements to gain some statistical insights about multiple large wide-area cellular networks, covering 60 sq. miles in and around Madison, WI, and along more than 120 miles of highways between Madison and Chicago.
    [IMC 2011 paper], [Datasets coming soon]

    Snapshot of one cellular network's performance in Madison.
    WiRover: High Bandwidth Internet Connectivity to Moving Vehicles

    WiRover provides high bandwidth connectivity to moving vehicles. Our targeted applications are to provide Internet connectivity to passengers of public transit buses, trains, limousines, taxis, as well as emergency services including ambulances, police, and fire vehicles. We have provided this as an experimental service since April 2010 to upto 12 vehicles in the Madison area, including multiple Madison Metro Transit buses (the local public transit operator) and Van Galder buses (a long-distance bus operator). The WiRover system aggregates data striping across multiple wireless cellular (3G or 4G) and WiFi networks simultaneously. Core innovations in WiRover lie purely in algorithms that efficiently handles transient failures of each network and provides significant improved performance, reliability, and overall user experience.

    We are always interested in trialing the WiRover system with various vehicle operators (or in other non-vehicular scenarios). If you have a fleet of vehicles (taxis, buses, trains, or other rental cars) that you want to provide Internet connectivity too, we can make WiRover available to you for use. Similarly, if you have a non-vehicular application that requires high-bandwidth connectivity (e.g., high-bandwidth video) in any arbitrary location, you can use WiRover for that as well. Please email Prof. Suman Banerjee.
    [Paper coming soon]
    Media coverage: Wisconsin State Journal, Milwaukee Journal Sentinel WTN News
    This work is also the grand prize winner of the Wisconsin Governor's Business Plan Competition 2011.

    WiRover exploits wireless diversity for improved performance.
    Apex: Cross-layer design for improving wireless media delivery through Value-aware Networking

    In many media delivery systems are bits are not created equal. Some bits are created to be more important (at the time when the media is encoded). The simplest example is found in MPEG streams where the relative importance of I, P, B-frames vary to the receiver. Apex (which stands for APproximate Communication for media EXchange) is a proposed wireless communication model in which we observe received symbols, when in error are typically a good approximation of the transmitted symbols. We exploit this property to provide the abstraction that more important bits (such as I-frames) can be better protected by the wireless channel natively than the less important bits. This design allows for an improved media experience. The system was implemented using the WARP software radio platform.
    [Sigcomm 2010 paper]
    Approximate communication exploits the fact that symbol errors are typically localized around the transmitted symbol.

    Last updated on Oct 2011 
    -- Suman