SNR enhancement of narrowband signals in colored noise using adaptive predictors

Use of the Linear Prediction Filter in cascade with an FFT is demonstrated as a sub-optimum realization of the optimum filter for maximization of instantaneous signal-to-rms noise power. Computer generated colored noise and sinusoidal signals are used to compare the suboptimum filter with a conventional matched filter (designed for white noise only) in terms of output SNR for cases in which the signal and noise spectra overlap. Significant improvements are seen with the LPF-FFT cascade. Four algorithms for implementing the LPF are compared for the application. Time - frequency plots are presented that graphically illustrate the filter effectiveness to extract signals masked by colored noise.