White Space Networking with Wi-Fi like Connectivity

Summary

This paper proposes the first network prototype demonstrating the feasibility of WiFi-like networking over UHF white spaces.

The authors contribute a new spectrum assignment algorithm that determines variable bandwidth during communication, a new AP discovery mechanism using a new technique called SIFT, and a novel method for seamlessly handling unexpected disconnections.

There are several properties of the UHF white space spectrum that pose problems to networking. First is the spatial spectrum variation, which implies that the AP must get information from clients before deciding the channel to operate on. The second is spectrum fragmentation, which results in several problems related to choosing a channel. And, finally temporal variation, which is due to interference from wireless mics.

The authors built the KNOWS hardware platform to address some of the problems with white space networking. The 2 main features it adds are Variable Channel Widths and Signal Inspection before Fourier Transform (SIFT). Variable channel widths allows the channel bandwidth to be greater than 5 MHz. SIFT addresses the problems with intermittent data transmissions and that transmissions can be sent over multiple channel widths.

They describe the details of the spectrum assignment algorithm which takes an adaptive approach to choosing a channel and periodically re-evaluating this choice based on feedback from the clients.

Finally, they perform an evaluation of WhiteFi, using both simulations and experiments on the prototype. They found that SIFT for pretty accurate, with a worst case loss of 2%. When testing SIFT with low signal strengths, they found that it performs well until it hits a threshold of 96 dB. They evaluated the L-SIFT and J-SIFT to discover APs and found that L-SIFT outperforms J-SIFT for narrow white spaces while J-SIFT is much more efficient for wider white spaces and both outperform the baseline algorithm. They also tried this in more realistic settings and found that J-SIFT still outperformed the baseline by 34%. They found that it took about 4 seconds to reconnect following a disconnection. Simulations of the spectrum assignment algorithm indicate that it yields a reasonably accurate prediction.

Criticism & Questions

I think this was an interesting paper. I was not too familiar with white space networking, so this was a good introduction, especially with all the background information they added about problems.

They explained the setup and results of their evaluations very clearly for the most part. However, when they were talking about handling disconnections, they didn't spend much time on that, only briefly mentioning that there was a lag of at most 4 seconds before reconnection. Without any baseline value to compare that to, I'm unsure of how much of a performance gain that is.