Virtual Occupancy – Physical Reality
Mark Burke, Bridget Lally and Andrew Kerans paper in IEEE DySPAN 2011 entitled Virtual Occupancy in Cognitive Radios is very interesting. I am going to use the term virtual occupancy in future! I think it is fantastic.
So what is it? Essentially communication systems are operated in the non-linear region, they can generate emissions at frequencies outside of the band in which they are operating. These emissions are generated through the mixing of two or more unwanted signals and are known as intermodulation products. Due to the generation of these intermodulation products, seemingly open channels may appear occupied – hence the term virtual occupancy.
The authors suggest that as part of selecting an operating frequency, a cognitive radio device must use information about its own position and information about the positions and operating characteristics of nearby land mobile transmitters and receivers to determine whether or not a harmful intermodulation product is likely to be produced when operating on a particular frequency.
I have cut and paste a figure from the paper to show the kinds of results that the authors generated. They consider a 20 MHz band in which cognitive radio operation is permitted. The frequencies of operation of the licensed transmitters and receivers are marked in blue and green respectively. These are obtained from a database in the approach used in the paper. The virtually occupied channels are shown as red points on the top line of each graph. An available channel is one which when occupied by a cognitive radio will not cause the generation of harmful intermodulation products falling on an occupied channel. This is the black line. I don’t think the graph used is the best form of representation as it is hard to see the detail but the general point is made. As an aside, I would like more information about the transmission waveforms of the cognitive radios. If there were OFDM waveforms I imagine the intermodulation issues would be worse.
I have only reproduced one plot from the paper but the authors look at different areas and different power settings for the cognitive radio. The make the general conclusion that the likelihood of virtual occupancy arising is directly proportional to the geographical and spectral service density of the area in which the cognitive radio is intending to operate and its intended operating power level. The paper goes on to show that if the licensed service uses a cavity filter the problem is significantly reduced. The emphasis in the paper there is about actions the incumbents can take to protect themselves.
The only other work I have seen in this area is the work by Preston Marshall. Preston’s thesis (he did is PhD in CTVR) was exactly about this and he has carried out a very detailed analysis of the affects of intermodulation products. The thesis was completed in 2009 and is entitled A Generalized Method for Quantification of Cognitive Radio Benefits within Wireless Systems. It is a great read even if I say so myself. He tackles the issue from a very different mindset though. He is interested in using the flexibility of a cognitive radio to mitigate the effects due to intermodulation. One way in which a radio can be well behaved from an intermodulation prespective is to have a really great dynamic range. This means it does not enter the problem non-linear region easily. However this comes at great expense. Preston argues that cognitive radios can avoid this expense by a combination of selecting bands appropriately (as per analysis in the paper on which I am focusing ) as well as setting the parameters of the cognitive radio appropriately. He makes conclusions about the gains that can be made from using cognitive approaches over non-cognitive approaches.
I have reproduced some of the results here. In these results Preston looks at the probability of overload of the frontend of both a non-cognitive and cognitive radio for different pre-selector bandwidths. A high probability of overload of course leads to a high probability of intermodulation products being produced. Note that Preston assumes that the cognitive radio has a set of filter tuning parameters, with discrete center frequencies and relatively constant effective Q factor. He makes the point that sole reliance on high and low pass filters, or fixed band-pass, is not considered, as their performance is generally unacceptable in any sophisticated conventional or cognitive radio, and they provide no options for adaptation.
The two figures I have cut and paste from the thesis show the benefit of the cognitive radio. Figure 4-5 shows the probability of frontend overload for the non-cognitive case. Figure 4-7 shows how the benefit of a cognitive radio which deploys a strategy of pick the quietest band first. This algorithm consists of applying all possible filter selection choices to the incoming (unfiltered) spectrum and measuring the total power and of course selecting the lowest options. By the way Preston has recently written a book that captures some of his ideas.
I have a great interest in this topic in general and would like very much to carry out experimentation to actually see some of these ideas in operation. It can however be difficult to get the kind of platforms which easily facilitate the types of experiments that would make sense here.
In addition I think it would be interesting to combine the knowledge here with investigations into the deployment of spectrum usage rights. So for example it would be interesting to see what kinds of parameters should be set within a network that would ensure that it would comply with different power flux density profile limits as conditions within that network change.