Signal identification by orthogonal transforms

The phase of a signal contains most of the essential characteristics for its identification. A technique is proposed to process the phase for noise reduction and signal identification. The identifier models the phase as a linear combination of orthogonal vectors, computes the transform coefficients, discards the insignificant coefficients to reduce noise, and keeps significant coefficients as features for classification. The Cramer-Rao lower bound of the phase and coefficient estimates are derived and analysis of the noise reduction process is given. Initial phase and frequency offsets often degrade the performance of an identifier and a procedure to eliminate them is presented. Simulation results for signals with common modulation types such as continuous wave, linear FM and multifrequency indicate that the performance of the proposed identifier is close to that of an optimum identifier with less than half of its computations.