10/29/03:
Spectrum Reduce, reuse and recycle. - Don't want to Reduce. - Reuse: can use game-theoretic ideas for non-cooperative users. - Recycle: can use assisted interference cancellation. Implementation issues of wireless systems. - Main A/D limitation: sampling clock accuracy - Use non-uniform sampling ideas to compensate for the influence of strong interference + unmodeled channel part. Invite guest speakers in early stages of research. - Bob Brodersen (BWRC) to talk about practical isssues of wireless systems. - Dale Hatfield (U. of Colorado at Boulder) to talk about spectrum policing. Future trends in FCC regulations. - Keep band manager approach for cellular and PCS spectrum. - Commons approach restricted to a limited set of bands.
11/12/03:
Game theoretic ideas for spectrum sharing. (PowerPoint presentation) - Identify cases where power allocation update ordering can make the equilibrium stable or unstable - Determine if mixed strategies can enlarge the set of Nash equilibria.
12/10/03:
FCC conference: --------------- Policy for spectrum sharing: - Won't be grand expansion for commons approach. - Will be some growth for unlicensed spectrum. - Property rights will still be there. - People rejected feasibility of commons approach. How to improve property regimes: - Prevent bad use of licenses to prevent new systems to appear. - Licenses will expire after 10 yrs but there is a renewal assumption. - Problem: determine what is fair for companies and people * How do we limit the rights that have been allocated ? * How do we encourage companies to use the rights in a way that benefits people ? Broadcast spectrum wasteful: - Let owners do whatever with it, use it, sell it, but have it used or FCC takes it back. - 700MHz band blocks (30 MHz of BW) about to be available. (Ch. 50-59,60-69) - Senators against changing property rights and forcing people to sell their spectrum. - Monitor use of resources and force to sell if underused. - Look at sensitivity of receivers. Require that receivers are not too sensitive to noise. - Do not regulate equipment or protocols, just performance. (analogy to fuel efficiency) - Try to make new systems more efficient so that less spectrum is needed. More on property rights: - Future spectrum holders can subdivide and sell spectrum. But holders do not want to risk their supremacy. - Spectrum that was bought can be used for anything (except for conventional broadcasting) even selling it. - Need to set property rights to allow for new heterogeneous, more efficient systems. - Idea: if a part of the spectrum is not very used and a new service has come up that does not interfere, it has priority on that portion of spectrum and use.
1/21/04:
Will meet at 12:30 in Cory 476 and have lunch. NSF Meeting ----------- Most presentations were about test beds for testing existing systems. For example they came up with robots to make measurements repeatable. Test-bed guys seemed serious. Most interesting presentation by Roy Yates. (PowerPoint presentation) Roy Yates talk (see slides): - High level stuff. Not very technical or detailed. - Worth looking at references (slides 19/20) - Some FCC regulations should be made stricter. Radio telescope bands seem to be affected by home appliances that comply with FCC rules. FCC cognitive radio document: ---------------------------- - Deadline for submitting comments is flexible. We should come up with some comments in about 6 months. - 3 strong technical comments could influence FCC decisions. Comments have value even if submitted after deadline. - FCC is really committed to do what the document proposes. - Rural stuff is important for political reasons. - The document rises 6 main questions: 1) Secondary markets for spectrum - Transaction costs ? - Market structure ? 2) Listen and talk: - How to determine that nobody is using the band ? User can be shadowed. Can use feature extraction to make signal strength measurements more sensitive. - Can use location to determine band usage ? Use GPS, TV or radio stations to determine location, and have a centralized database with spectrum usage information. 3) Public safety bands: - How to force people to shut up ? - On what time scale should users release the channels ? - Small cities may consider this to reduce their deficit. - Could also increase allocated BW (with respect to current allocation) in case of emergencies using interruptible spectrum share. 4) Satellite communication. Non geostationary satellites have predictable positions. - Can we ask users to stop transmitting when satellite passes in their direction ? - Can we require the satellite receiver to hand-off to some other satellite instead of forcing other users to shut up ? 5) Rural areas: - What power limits to use ? - How do we determine location (rural/urban) to set power limits ? (GPS, TV or radio stations, etc) 6) Competing UWB standards (impulse radio vs. OFDM) - Get another view from that given by companies.
2/4/04:
Talked about budget issues. Continued discussing NSF Meeting -------------------------------- - Feature extraction/detection problem: want to detect if a signal is present or not before start transmitting. Can be in a shadow, so may need to improve SNR to allow detection. * For what kind of modulations/codes does feature detection make sense ? * What stuff can we put in the signal to make feature extraction easier ? * Could use relaying ideas to help detection: everyone listens and if someone hears something it announces it to some central station. - Secondary markets: * How would the sublicensing work ? * Can we give access to someone and then shut him up ? * There can be a negative result if FCC forces everyone to give out their unused spectrum if in the end that spectrum does not get used. - For next meeting we will analyze the document about feature detection (.pdf)
2/18/04:
Feature extraction/detection ---------------------------- - Radiometric (squaring + detecting DC) not very interesting. - If we can design the signal we can: * Add tone/s * Encode data at low data rate to indicate bands in use, etc. * Use some GSM band for putting beacons to indicate frequencies in use. - Research questions: * Fix allowed time for detection. Assume that signal is BPSK and that we are phase locked to it (very optimistic assumption). How much better can we do with respect to just squating ? This would give us an upper bound on achievable performance. * If we now allow for other modulations we will only do worse. * What if we know the code rate but not the code ? * What if we know everything but the SNR is so low that we cannot decode ? - Bixio offered some simulators/boards for software defined radio.
3/3/04:
Feature extraction/detection ---------------------------- - Analyzed difference in performance between coherent and uncoherent detection: * Uncoherent model: signal to detect is random Gaussian. * Coherent model: signal to detect is a sinusoid (constant in baseband). * Assume that noise and signal statistics are known. * For fixed error probability at low SNR, the required # of samples is: ~ 1/SNR for sinusoidal signal ~ 1/SNR^2 for Gaussian signal * Found some intuitive interpretation for this result - Sinusoid gives performance upper bound, since Tx signal is completely known. - If we do not know what is transmitted, but know the codebook, what can we do ? - Look at Slepian permutation codes to see if we can exploit their properties.
3/17/04:
Feature extraction/detection ---------------------------- - Reviewed findings of previous meeting. - Problem: if the codebook is known how many CW can we tolerate to get required # of samples ~ 1/SNR ? - If we cannot decode correctly, can we still do something ? List decoding: if the codewords in the decoding list are close together, then we can reduce the noise and can achieve good performance. - Q: when do the decoded codewords fall in a small region ?
3/31/04:
- Some spectrum has been allocated, but no-one has bought it yet. - Debate about auctioning all bands: * Do not auction all bands: what if a new system comes out after all bands have been auctioned ? * Auction all bands and allow for leasing. This last approach could unlock the situation. - 2 opposing views: * Some industry people (i.e. large companys such as QComm) say that it is not worth developing technology if the FCC does not give rights to the use of the spectrum. They attribute lack of innovation to lack of guaranties. * Smaller companies claim that lack of innovation may occur because all spectrum is already allocated. - Organization of the FCC: OET, OPP and WCB branches - Bring to our meetings FCC and non-FCC people: * Regarding FCC people, bring one of each branch. Maybe bring all of them simultaneously. * Non FCC people: + Stu Benjamin (Duke): he is against commons approach. + Y. Benkler (Yale): commons guy. + Dale Hatfield (U. of Colorado) * Think about questions to ask them. Some could be: Q: What conditions on techonoly or the physical medium would make their positions (regarding commons/licensed approach) change. Q: What models do they have ? Q: Backwards compatibility. How to deal with it. Social cost of changing technology. Q: Under what conditions (infinite bandwidth for example) property rights do not make sense ? - Regarding the feature extraction/detection problem: * Conjecture: For R>0 and random Gaussian codebook we cannot do better than with the energy detector. - For next meeting: Niels will look at Benkler's paper.
4/14/04:
- Niels presented Bentler's paper. (.ppt) presentation- Some topics discussed: * If we have a very wide bandwidth, why is it important to be efficient ? * Why bother using a particular band if we can go somewhere else ? * Sensing seems to be the main issue. * What if the right of spectrum ownership was shifted to receivers ? - Sectrum inventory from FCC: (link).
4/28/04:
- Interference temperature. Motorola comments (.pdf)* Int. temp. hard to measure. * CDMA systems are very sensitive. Simulation results: 1dB increase in noise floor decreases capacity by 10%. * @ 6GHz need to lower Tx power by 60dB not to affect performance (fishy). - Papers on spectrum pooling to share spectrum: * Divide spectrum in packets that can be assigned as needed. - Future agenda: * No guests will come over summer. * FCC people may come in August. * There is a technology survey in every talk. We should find a good one or do one ourselves.
6/23/04:
- ITR meeting: * Quite useless. Only goal was to show congressmen how much ITR helped achieve NSF goals. * Many 10 minute talks of diverse topics. * Interesting issue: regarding hidden terminal problem, even if one does not hear anything, that does not mean that nobody else is hearing in the same band. This issue arises in weather radar systems that can detect very faint signals. - BWRC retreat: * Qualcomm talk: why not assign all bands to cellular-type systems (based on QComm technology of course :) to achieve a high spectral efficiency ? * Many issues depend on things that are not properly defined, i.e. What is harmful interference ? * To give support to commons approach, need to measure performance in other terms (flexibility, time to deployment, opportunities for small companies, etc.) rather than only spectral efficiency. * No need to speak in favor of software configurable radios: it is already understood that they can be useful in licensed/unlicensed bands. * If commons approach is spectrally inefficient, will everything fall back to licensed spectrum when spectral utilization increases ? If the access to spectrum is free, what prevents users from grabbing large portions of spectrum and use them inefficiently ? * May need to determine whether we will reach a situation of spectrum scarcity, possibly by assuming a maximum bandwidth that a physical person may need and determinig a maximum density of people in a given area. - Discussed other issues: * How do we measure background noise or interference in an absolute way ? * Can punishment strategies work in asymmetric situations ? * Can we assume the existence of an FCC police that would apply punishments everywhere, or that can be called upon to punish ? * Bixio proposed charging each user by the amount of interference generated to other users, weighing the cost depending on the location. This may require signature sequences to identify interferers. * Can everything go wireless ? How do we encourage systems that are naturally wired (i.e. computer monitor) to stay wired ?
8/04/04:
Discussed status of everyone's research: ---------------------------------------- - Andrea: Analyze under what circumstances the primary owner of a band would be willing to negotiate (sell, lease, etc.) the use of the band to a secondary user under very idealized assumptions. - Niels, Rahul: Hidden terminal problem: * If there is a pilot tone for presence detection, required length of transmission for successful detection scales as (1/SNR). * If energy detection is used length scales as (1/SNR)^2 * New result: if transmission is uncoded, and one perfectly knows the constellation and assuming perfect synchronization, if constellation is symmetric about origin still lenght scales as (1/SNR)^2, i.e. one cannot do better than energy detection. * Conjecture: if constellation has zero mean, still lenght scales as (1/SNR)^2. This was observed in simulations. * Want to prove a result in these lines even for the case of coded transmission when the code is known but transmission rate is above capacity of the secondary user. - Krish: talked about Identification via Channels paper: * One can identify N=2^(2^(nR)) (doubly exponentially many) objects in blocklength n with arbitrarily small error probability in a DMC if randomization is used in the coding procedure. * If one is interested in identifying only 2^(nR) objects, then one can reduce the transmission power with the blocklength n. * This could be used to send alarm or beacon signals for spectrum sharing with very low power / high detection probability. * Would be interesting to analyze how to obtain different priorities in the detection probability, i.e. have a high priority alarm that must be detected by every user no matter what, and lower priority beacons that are not so important. - Stark: Use of feedback to reduce system complexity or improve error exponents. Can we relax some of the assumptions on the feedback channel and still obtain improvements ? * Could classify priority of systems according to the requirements on the feedback channel, and allocate resources accordingly. - Raul: Frequency reuse and graph coloring problem for spectrum sharing. - Discussed about property rights on shared resources such as air space for airplanes, fishing quotas, space for each car in a freeway, land, harbors, trying to find analogies to the spectrum sharing problem. * None of these shared resources is similar to spectrum. Some of them are not renewable, some of them can be improved and others are easy to control. * Also most of the sharing strategies do not work or are inefficient: centralized air control is very inefficient, fishing quotas are hard to enforce. * Some ideas could be used in spectrum sharing: carpool lanes give higher priority to users making more efficient use of the resource, metering lights control access to the resource to avoid traffic jams, etc. * People do not necessarily decide in the best way for their interest, i.e. people watch very bad TV shows, or eat unhealthy food knowing that it is not the best choice. Maybe one could provide rules to encourage people in taking better decisions.
10/20/04:
ITR meeting (Anant) ------------------- 1 - No more funding for new radios. 2 - People are making more detailed spectrum measurements: * Histograms of energy in bands show different behaviors for TV bands, cellular bands and empty bands. These results are to be expected - nothing surprising here. * Military uses their bands, at least near its bases. This is somewhat contrary to our intuition that they hog lots of bands without using them. 3 - UWB: * 2 opposing groups: Pulse radio (PPM) and OFDM (most companies) * Most devices are designed to operate in 3-4.9GHz band. Why not use higher frequencies ? * 3-4.9GHz band is practically empty. For now, can use this band w/o complications with cheap radios. If later on the band gets crowded, these systems will not work (they are not designed to deal with interference) * Applications they look at: high data rate bursty traffic, small range. They cannot use high processing gain unless using extremely large bandwidth. Large processing gain requires computation, and results in high cost. Therefore they cannot deal with interf. * All efforts are concentrated in designing a good MAC. * Hope for these guys: in the future 3-4.9 GHz band will be reserved for these devices. They rely on politics to save them. * Bottom line: they are not doing spectrum sharing. 4 - People at UIUC and Philips are looking at the problem of detecting unused bands. 5 - GNU radio: software radio based on PCs and A/D converters. Hardware will be available by Jan '05, at about $500/piece. We will buy some.
11/03/04:
BWRC cognitive radio workshop ----------------------------- - Danijela's cyclostationary results vs. Anant's results. Are they consistent ? * They don't contradict each other. Basically Danijela's performance using cyclostationary detection still performs like O(1/SNR^2), but by exploiting some features of the signal can improve in a constant factor. * Basically what happens is that if the signal to detect uses a very inefficient modulation scheme with a lot of redundancy, the constant factor can be large enough to make detection relatively easy. - Cooperation between nodes to detect primary user. Can we trust messages sent by other nodes ? * It is easier to trust a positive detection. If the guy detecting gets a good reception of the signal, he can decode it and include a portion of it in the message sent to the other node. Then this other node can coherently try to detect the signal to verify the assertion. This should work even if the signal is much weaker at the second node. * It is harder to verify a negative detection. It could be that the node is just lazy and wants to save battery and says always that no signal is present. However if we have many nodes, some of them will really measure and reliability increases. * How can we design the systems/rules so that things work fine at first when there are few nodes out there measuring (and hence there is little reliabilty in the detection of the primary user) and eventually later, when these systems become more popular ? When there are few systems, spectrum is widely available, and we can afford to be very spectrally inefficient by spreading a lot (thus not hurting the primary user, even if he is there). As the band gets more crowded, we need to use higher spectral efficiency (i.e. transmit at higher power spectral density level in some band) in which case we can hurt the primary user. But as we have more systems, primary user detection becomes more reliable. - In Anant's model, he considered noise uncertainty and finite A/D precision as receiver constraints. * One can add processing power as another constraint, and think of situations where we want some other device (like a notebook computer) to do the processing. * Not clear that a general purpose processor (like a PC) can do DSP like operations more efficiently than a cheap ASIC. - For Spectrum Sharing problem (not cognitive radio): * Noise + interference is not white anymore. Moreover, Noise+interference PSD varies fast as packets from neighboring nodes get transmitted. * Traditionally, colored noise is dealt with by whitening, but how do we characterize the PSD when it varies quite fast ? * Also, the new channel as we whiten the noise has a short coherence time. * Maybe this is not the right way to look at the problem. If we deal with interference by efficiently coordinating spectrum use (say avoiding neighboring nodes to share same bands), then interference seen will be average interference from nodes far away, and these effects will be less important.