White Space Networking with Wi-Fi like Connectivity

P. Bahl, R. Chandra, T. Moscibroda, R. Murty, M. Welsh, “White Space Networking with Wi-Fi like Connectivity”, ACM SIGCOMM Conference, (August 2009). [PDF]


With the FCC opening up the unused portions of the UHF spectrum, also known as the “white spaces”, researchers have set upon implementing WiFi over the newly acquired territory. Such networking is different from existing WiFi technologies along three axes: spatial variation, temporal variation, and fragmentation of the UHF spectrum. This paper presents WhiteFi, the first WiFi over white spaces that implements adaptive spectrum assignment algorithm to tackle spatial variation and fragmentation, and proposes a low overhead protocol to handle temporal variation. This paper is concerned with the “only one Access Point (AP) with multiple clients” scenario, and unlike the existing solutions, WhiteFi works on variable channel widths and uses SIFT (Signal Interpretation before Fourier Transform) for faster and better detection of data transmissions.

The Three Challenges

As mentioned earlier, UHF white spaces are different from ISM bands used in the existing WiFi in three ways:

  • Spatial variation: The set of channels occupied by TV transmitters and the number of stations operating in an area vary from one place to another with significant variance. As a result, an AP must select a channel after proper consulting with all its clients based on their location.
  • Spectrum fragmentation: Since UHF bands already have incumbent TV channels, they are fragmented irrespective of urban and rural areas. As a result, radios working in white spaces must use variable channel widths, which makes channel assignment and AP discovery harder than that in ISM bands.
  • Temporal variation: Channel occupancy by incumbents also vary based on time. Since the FCC does not allow messing up incumbents, the AP and its clients must move away as soon as they detect a microphone (working on UHF) starts working.

Proposed Solution

The proposed solution consists of an adaptive allocation/assignment algorithm, a lightweight protocol, a specialized hardware to glue them together.

  • Hardware: The authors developed KNOWS, a specialized hardware platform with a PC, a scanner, and a UHF translator, to support WhiteFi. The PC controls the others, the translator translates backs and forth from WiFi, and the scanner samples UHF spectrum to find white spaces.
  • Spectrum assignment: The AP must pick a channel that can be used by all the clients. So once it finds one, it keeps checking periodically to find a better one (voluntary) or to move away from a conflicting channel (involuntary). For finding a channel, the authors propose a channel probing mechanism based on a new metric that can predict channel availability with high accuracy. The AP evaluates all possible channels and picks the one that maximizes the metric and works for all its clients. Then the AP broadcasts the new channel for its clients to switch to. The information between the AP and its clients are passed using a bitmap where each bit says whether a channel is or can be used or not. To avoid moving back and forth between channels without much gain, the authors also used hysteresis process.
  • AP discovery: Since the AP can use more than one channel, finding an AP is computationally harder for its clients in WhiteFi than that in WiFi. The authors introduced a new signal analysis technique SIFT, which basically determines packet widths based on signal amplitudes with small number of false positives. Two different versions of SIFT are proposed:
    1. L-SIFT linearly scans all the channels, and
    2. J-SIFT scans in a top-down manner from higher MHz channels to lower ones in a staggered manner, and once it detects the presence of a WhiteFi transmitter, it identifies the center frequency.
  • Handling disconnections: To cope with sudden appearance of an incumbent transmitter, the authors propose to keep a backup channel where everyone periodically pings about their status. To backup the backup, an arbitrary available channel is used as a secondary backup.


The authors performed most of their evaluation using simulation, because they could not test on real white spaces, citing FCC regulations! And there are no surprising finding either: SIFT is mostly accurate across different channels, J-SIFT does really well in discovering APs, disconnection handling works, and the proposed metric closely follows availability to make the proposed assignment algorithms perform really well in real time.


The problem is really fresh and exciting, as is the solution, except for the gory signal analysis part (fortunately or unfortunately that’s a big contribution). Its great to see the ‘first’ paper in a completely new area! One can predict that this paper will be really big (at least in terms of citation) in the long run if all the hype about white spaces hold up.

There are a few comments. Throughout the evaluation section, the authors keep talking about the optimal omni-potent algorithm that works best for any channel. How did they come up with that, and how it influenced their design? While evaluation the impact of spatial variation, the authors used some minimum and maximum probabilistic values. Where and how did they come up with those values? If there is any other studies of white spaces, they don’t mention it.

Unlike 10MHz and 20MHz packets, 5MHz packets have little or no variation with respect to any parameter. What is the reason? Are they parameter insensitive or the parameters are not fine tuned enough to catch their sensitivity. It would have been better to see some explanation.

Leave a Reply

Your email address will not be published. Required fields are marked *