| A cyclostationary process is a type of time-series that arises from a periodic phenomenon and whose statistical properties vary periodically with time. Spectral correlation is a characteristic property of cyclostationarity that is defined in terms of spectral correlation density function, which is a generalization of the power spectral density function. Most modulated signals encountered in communication systems are appropriately modeled as cyclostationary signals. This dissertation examines the spectral correlation property for a variety of modulated signals commonly used in communication and investigates its application to cyclostationary signal detection and source location.; The spectral correlation functions of various signal modulation types are derived and analyzed. This includes PM and FM signal for analog modulation types and PSK, FSK, and spread-spectrum signals for digital modulation types. The unextendability of Woodward's theorem to the spectral correlation function for high-index FM is explained. The spectral correlation functions for these signals are graphed, compared, and contrasted. The measured spectral correlation functions for four simulated signals, BPSK, QPSK, SQPSK, and MSK, are presented to corroborate the theoretical results.; Two types of optimum detectors for detecting weak cyclostationary signals buried in Gaussian noise are described, and their deflection, a measure of detector output SNR, is completely specified in terms of the spectral correlation of the signal to be detected. The deflection is evaluated for both types of optimum detectors for BPSK, QPSK, SQPSK, and MSK signals.; Several cyclostationarity-exploiting methods for time-difference-of-arrival (TDOA) estimation of a wavefront received at two separate sensors are presented. These methods include the spectral correlation ration (SPECCORR), spectral coherence alignment (SPECCOA), spectral correlation nulling (SPECCON), cyclic cross correlation, and cyclic phase regression methods. The tolerance of the TDOA estimates produced by these methods to noise and interference is explained theoretically and demonstrated with simulations. The bias and variance performance of these methods is studied by means of simulations, and it is found that the variance for some of the methods is smaller the Cramer-Rao bound, which is the minimum variance that can be achieved when the signal of interest is modeled as a stationary process. |
