UCLA

With the rapid increases in the number of wireless devices, fixed spectrum allocation has shown to be a major limitation to the evolution of wireless technologies. Cognitive radio (CR) allows opportunistic spectrum access by searching and utilizing temporally and spatially unused spectrum, provided that CR users do not cause interference to the primary users of the spectrum. Spectrum sensing over a wide bandwidth increases the probability of finding under-utilized spectrum for cognitive radios. However, the realization of wideband sensing is challenging because strong primary users introduce large dynamic range and spectral leakage to adjacent unused bands.

This work presents an algorithm-architecture co-design framework for wideband spectrum sensing. The suppression of spectral leakage is achieved by multitap-windowed FFT processing, which also enables reduced sensing time. The sensing time and detection threshold are adapted according to channel-specific spectral leakage, enabling reliable wideband detection within constrained sensing time. Power and area cost of the compute-intensive FFT block is minimized by using parallelism, radix factorization, and compact delay lines. A per-channel floating point data processing for large dynamic range signal is employed for power and area saving. A partial PSD estimation scheme that performs energy detection on only the band-of-interest further improves the energy efficiency. Two chips have been designed to demonstrate these concepts. These chips guarantee reliable weak signal detection with short sensing time, and outperform the prior work by at least 22x in power/bandwidth. Techniques developed in this dissertation enable energy-efficient chip implementation of advanced wideband signal processing for cognitive radios.