My Talks and Publications

Related Work

Resume 

MobiNets Home

WINLAB Home

My PhD dissertation titled "Cross-Layer Aware transport protocols for time-varying wireless links" was completed in June 2007. The work combined (a) the intricacies of radio communication that has seen 120 years of advancement, with (b) Packet-switched networks which, in less than 40 years, has catapulted human civilization into an Information Age.
We addressed the problem of how file transfer delays can be reduced in the general class of wireless networks (cellular infrastructure, 802.11 wireless LAN, 802.11 mesh networks, etc). Particularly in difficult wireless conditions such as crowded areas (high interference) and remote/mountaineous terrain (fluctuating/low SNR). Algorithms to address time-varying bandwidth and noise at the transport layer were considered. A clean-slate reliable transport protocol called Cross-Layer Aware transport Protocol(CLAP) was developed and validated in the NS2 simulator. It was extensively evaluated in 802.11 single hop, WiFi and mesh environments. Significant reduction in file transfer delays were observed in all the scenarios considered.

Writeup and publication of CLAP results in WLAN and wireless mesh environments are available (here's the single wireless hop results).

Rapid evolution of wireless network technologies A fleet of new discoveries in the past few decades such as OFDM, turbo-coding and MIMO have facilitated today's commercially viable consumer-end wireless technology. The past two decades has seen tremendous advancement in wireless Wide Area Networks (WWANs or Cellular networks), emergence of wireless Local Area Networks (WLANs - 802.11 and HiperLAN) in offices, homes and even entire cities and wireless Personal Area Networks (WPANs - such as with Bluetooth). In Q3 2008, the first 4G mobile WiMAX network in USA was deployed in Baltimore. Test showed speeds upward of 2Mbps in over 90% of locations. LTE that promises over 15Mbps is scheduled for launch in mid 2010.

Layers 4 and above need to catch up Despite the large wireless bandwidth, file transfer over these networks faces steep challenges. Poor TCP performance in time-varying bandwidth and loss, is a major problem. For over two decades, there have been significant efforts to enhance TCP to work well in wireless as well as wired networks. But a universal solution has evaded researchers. Disconnections, route failures, random link loss, protocol susceptibility to latency variations - all have been addressed.

TCP flow control and error control algorithms respond poorly to time-varying bandwidth and high packet loss. These conditions that are rare on wired networks, however, occur frequently in wireless networks because of fluctuating SNR and shared medium interference. Even with a single time-varying link in the route, there are frequent TCP deadlocks and timeouts that degrade overall performance. TCP timeouts and exponential backoff result in a protocol shut down when error rate exceeds 5%, leading to poor bandwidth utilization. In fluctuating bandwidth and loss conditions, TCP flow control is disrupted because the ACK irregularity upsets the protocol clocking.
In this PhD work, we question from a different angle - "With a clean-slate approach, how to reduce file transfer delays in the general class of wireless networks? " To this end, we developed a reliable transport protocol - Cross-Layer Aware transport Protocol (CLAP) , that can address rapidly fluctuating bandwidth and high rate of packet loss. Cross-layer MAC updates, rate-based flow control, decoupled error control and asynchronous NACK-based aggregate feedback, are utilized in the protocol. A software framework to extract status parameters within a node, and between nodes is instrumented. Cross-layer updates end-to-end utilize UDP/IP signalling, and consume less than 1% bandwidth overhead. Together, these design aspects achieve major improvement in channel bandwidth utilization in noise-free as well as noise-prone conditions. Multi-flow fairness closely to MAC-level fairness.

Today the ease of implementing embedded software and networks' feasibility to support at least a 1% control overhead, makes it viable to implement protocols like CLAP that require supplemental bandwidth information.

There is exciting opportunity for Future research.
(a) The cross-layer framework can be extended across a general class of networks so that high-performance transport protocols like CLAP are feasible.
(b) CLAP enables receiver-driven flexibility in error control. New research could apply CLAP-like transport protocols to applications that don't require 100% reliability. For example in video streaming, opportunistic data transfer in sensor networks etc

Briefly about myself: I graduated in 06/2007 after an exciting 4 years 8 months as Graduate Assistant at WINLAB, was supported during this time by a scholarship from Corporate Research, Thomson Inc (Princeton). My advisors were Prof. Dipankar Raychaudhuri and Dr. Sanjoy Paul. Dr. Kumar Ramaswamy (Thomson Inc.) was my co-advisor. I earned a B.E. in Electronics and Communication, Bangalore University (B.M.S. College of Engineering), India, and M.S. in Digital Signal Processing and Software Engineering at CAIP, Rutgers University. My advisors during MS were Dr. Attila Medl, Prof. Rick Mammone and Prof. James Flanagan.