Investigation and implementation of a computationally-efficient algorithm for cyclic spectral analysis

The suitability of cyclic spectral analysis for signal detection and identification is investigated. An overview of the theoretical background leading to the cyclic autocorrelation and cyclic spectral density equations is presented. Brute force and computationally-efficient approaches for estimating the cyclic spectrum are examined. Of particular interest in this thesis is the computationally-efficient FFT accumulation method (FAM) for computing cyclic spectrum estimates. Key features of the algorithm which reduce the number of computations are demonstrated. A cyclic spectrum analyser that processes digitally-modulated signals, computes the cyclic spectrum estimates, and displays the results is designed, implemented, and tested. The FAM algorithm is implemented on a DSP board containing a TMS320c51 processor. Cyclic spectra for phase-shift-keyed (PSK), frequency-hopped frequency-shift-keyed (FH/FSK), FH/PSK, and burst PSK waveforms are computed and displayed. Theoretical and experimental results are compared and discussed.