1Fundamental tradeoffs in robust spectrumsensing for opportunistic frequency reuseAnant Sahai Niels Hoven Shridhar Mubaraq Mishra Rahul [email protected] [email protected] [email protected] [email protected]. of Electrical Engineering and Computer ScienceUniversity of California, BerkeleyAbstractUnder the current system of spectrum allocation, spectral bands are allocated for time-scales rangingfrom years to decades and spatial scales ranging from counties to continents. Robust operation of devicesis enabled by the use of guard bands in space and frequency. This paradigm has resulted in vastlyunderutilized spectrum bands, even in urban locales. However, advances in wireless technology nowallow devices to operate on much smaller time and spatial scales. Like sand and pebbles poured intothe gaps between larger rocks, such devices have the potential to greatly improve our overall spectrumutilization.The fundamental constraint is robustly guaranteeing non-interference to privileged users of the band.We focus on cognitive radios that perform sensing and adapt their output to avoid interfering. We showthat uncertainty in fading poses a serious challenge by forcing high sensitivity. These challenges areexacerbated by the presence of multiple users, but gains are available through cooperation. However,cooperative gains are limited by trust/reliability and the network’s cooperation footprint. Furthermore,uncertainty regarding noise and interference imposes fundamental limits on how sensitive robust sensingcan be. Local cooperation is necessary to provide fairness by reducing the uncertainty impact of otherusers’ transmissions.2I. INTRODUCTIONThere is no doubt that the development and deployment of wireless technologies is one ofthe great achievements of the past century. Frequency division has allowed different systemsto be deployed in a decoupled way, even when their spatial and temporal extents overlap.Since the filters used to isolate systems are not perfect and radio waves attenuate graduallywith distance, static “guard bands” in both space and frequency are used to accomodate theseimperfections and robustly prevent harmful interference. Both the structure of the guard bands andthe frequency/space allocations themselves are determined by govermental regulatory agenciesoperating on a time-scale ranging from years to decades and with spatial-granularity that rangesfrom counties to continents. These allocation decisions are then enforced by the certification ofequipment and the issuing of licenses for exclusive use.Looking at the NTIA’s chart of these frequency allocations (Figure 1.a), it appears that weare in danger of running out of spectrum [1] since most bands have already been assigned tospecific uses. However, allocation is only half the story since wireless links themselves operateon different, and often much smaller, space and time-scales. Figure 1.b,c show that much of theallocated spectrum is mostly underutilized [2]. This underutilization is typical [3], [4] and uponreflection, is a natural consequence of the distinct scales at which regulation and use occur —a vase filled with rocks still has plenty of room for sand.When the current regulatory model was adopted, radios were statically configured. Conse-quently, device certification was the only way to assure noninterference and robust operation.It was natural for allocations to operate on a timescale compatible with equipment lifespansand on a spatial extent compatible with device mobility and market size. As digital technologyadvanced, dynamic configuration became possible [6] and it became conceivable to exploit thevast spectral gaps that exist at the time and spatial scales of actual use.A. The policy debate and alternativesFrom an economic perspective, spectrum is an unimprovable finite, but infinitely renewable,resource that can be used as an input to provide useful services. The policy focus is thereforeon putting it to the best possible use. The slow time-scale of the existing allocation processis recognized but is usually attributed to its centralized “command and control” nature withlicensees having an exclusive right to do a specific thing only.3(a) The NTIA’s spectrum allocation chart makes availablespectrum look scarce.(b) Usage varies with time(c) Actual intensity of usage varies greatly by frequency.Fig. 1. Measurements from the Berkeley Wireless Research Center show that there is great discrepancy between spectrumallocation and usage. Measurements were performed once every minute for a duration of 1.6µs using a 20GHz ADC. The lightbackground in (c) represents levels in the range of -130dBm. As a comparison, Dynamic Frequency Selection (DFS) techniquesmandated for the 5.3GHz band need to switch frequencies when they detect received power levels in the range of -62dBm [5].4One proposed solution is simply eliminating legacy spectrum rights to a large extent and creat-ing new comprehensive commons [7]. Commons proponents argue that the astonishing success ofWiFi-style devices is proof that “unlicensed” devices can coexist in a largely unregulated way [8].While usually discussed in terms of freedom and flexibility from red-tape, the commons approachis also implicitly a rejection of the separation of time-scales between allocation and actual use.Commons proponents assert that “wireless transmissions can be regulated by a combination of (a)baseline rules that allow users to coordinate their use, to avoid interference-producing collisions,and to prevent, for the most part, congestion, by conforming to equipment manufacturer’sspecifications, and (b) industry and government-sponsored standards” [7]. However, there hasbeen limited research into whether wildly heterogenous1wireless services can coexist using self-interested coordination. [9] reveals that in cases of severe asymmetry, self-interested behavior isnot enough to guarantee either fairness or total utility whereas reasonably symmetric cases canin fact support a wide range of fair and efficient self-enforced equilibria.Less radical solutions preserve a distinction between time-scales, spatial scales, as well aspreserving a role for “primary users” given priority access to the spectrum. The common goalis increased productive utilization of the spectrum at the scale of actual use, but the exact rightsafforded by this priority access depend on the approach chosen from Table I.Interference