Eliminating the bottleneck in WiFi networks: New protocol delivers order-of-magnitude increase in performance

Have you ever been to a very busy coffee shop, hotel lobby or conference and been unable to access WiFi? Wi-Fi networks based on the 802.11 set of standards are pervasive, but have a fundamental performance bottleneck that limits the practical number of active users on a network. The existing protocol is based on a random access method and has limitations, such as traffic congestion and the difficulty of implementing quality-of-service (i.e. prioritization).

A team of Harvard researchers, led by Professor H. T. Kung, developed a solution to these limitations; a new protocol based on a compressive sensing statistical method. This protocol takes a centralized approach to the problem, thus enabling efficient scheduling among a number of hosts and easing the implementation of quality-of-service. The invention can also coexist with current 802.11-based products.

The new protocol, called compressive sensing medium access control (CS-MAC), takes advantage of the sparse property of a system, and has demonstrated its capability to acquire and extract critical information efficiently. For a system with a number of hosts, CS-MAC sets up a central coordinator to collect host requests compressively and make multiple grants simultaneously. Concurrent requests from several hosts can be combined in the air with only a few compressive measurements. On the other hand, currently established 802.11 standards take a random access approach. Because the lack of a central coordinator, requests from different hosts are competing with each other, resulting in traffic congestion. And it is also difficult to prioritize resources to different users and applications.

CS-MAC overcomes these two major problems while maintaining low computational complexity. The protocol overhead for CS-MAC is shown to be comparable to basic 802.11 DCF when amortized over multiple packets. Thanks to its efficient scheduling capability, CS-MAC is especially attractive in large-scale distributed systems.

U.S. Patent(s) Issued: US9032075B2