The false positive congestion problem and probabilistic spectrum sensing

We assess the problem of band congestion due to spectrum sensing false positives combined with a guard-band policy. We calculate the size distribution of spectral white spaces to find that large fractions of a band can be wasted even for small probabilities of false positives in single bins. To ameliorate this problem, we propose an algorithm consisting of random bin sampling followed by density clustering. The clustering algorithm yields a dramatic enhancement of the false positive rejection rate due to large differences between the densities of true and false detections. Simultaneously, a reduction in sensing time is possible due to the fact that the number of randomly sampled bins need only be a small fraction of the total number of bins. We explain the theory behind this method, compare to simulation, and derive heuristic guidelines for its implementation.