Efficient Spectrum Sensing With Dyadic Tree Partitioning

A low-complexity spectrum-sensing method that is based on dyadic (binary) tree partitioning is proposed. For this purpose, first, a representation of the spectrum occupation with a dyadic tree is developed. Then, the accuracy of this representation is evaluated through a rate-distortion criterion. Using this criterion, a pruning algorithm for the dyadic tree that minimizes the number of sensing operations for a predetermined value of sensing accuracy is conceived. After tailoring the pruning algorithm to the missed-detection and collision requirements of the opportunistic access, an adaptive version that performs efficient sensing in slowly varying environments is proposed. The proposed algorithm is tested on real spectrum measurements taken in the 400-910-MHz range. The sensitivity of the proposed algorithm to sensing imperfections is also evaluated. The obtained results demonstrate that the proposed algorithm provides a mean distortion of 11.4% and a mean collision probability of 5.6% for sensing only 30% of the total spectrum. A comparison with conventional energy-based detection highlights a gain of 50% in terms of the number of required measurements per channel in the low-signal-to-noise-ratio (SNR) region, where conventional energy detection is known to have poor detection performance.