UCLA

Wideband spectrum sensing is a key enabling functionality for Cognitive Radios (CRs) since it detects unoccupied bands and allows throughput increase. Due to prioritized spectrum sharing between CRs and primary owners of the spectrum, it is not only sufficient to detect occupancy but also distinguish among different users in order to manage interference. To realize this functionality in a practical radio, there are several implementation challenges that we address in this thesis: 1) high computational complexity and energy cost for the detection and classification of a broad range of communication signal types over a wideband spectrum, and 2) impact of wideband receiver impairments including nonlinearities, carrier and sampling offsets, and multipath.

The approach we take for joint sensing and classification is based on extracting and processing cyclostationary features of modulated signals. Conventionally, cyclostationary feature detectors are considered as robust detectors under noise uncertainties. However, estimation of cyclic features under constrained sensing time suffers from cyclic frequency offsets resulting from non-synchronous sampling and local oscillator offsets. We propose a new frame-based cyclic feature estimator and optimize its frame length for a given carrier and sampling offset distribution. For identifying signals with unknown parameters in an energy efficient way, we develop a hierarchical reduced complexity cyclostationary-based classification algorithm by optimizing the search of cyclic features. The reduction in complexity and energy cost comes from discretization of feature space based on tolerable frequency offsets for the required classification accuracy. Next, we study the impact of wideband receiver nonlinearities on feature detection and show that performance loss depends on blockers' strengths and modulation types. Based on this result, we devise a compensation algorithm that incorporates modulation classification into intermodulation terms cancellation. Finally, we investigate how the signal sparsity in the cyclic domain can be utilized to reduce the sampling rate requirements via a compressive sensing approach. As a result of the additional sparsity in the cyclic domain, our results show that a compression rate smaller than the conventional Landau rate can be achieved. In summary, this thesis provides a comprehensive analysis of cyclostationarity based sensing and classification under practical signal and radio conditions, and proposes a set of algorithms for a robust performance and energy-efficient implementation.