Stefan T. Sidahmed
Antal A. Sarkady
Dept. of Elec. Eng., U.S. Naval Acad., Annapolis, MD 21402
Power spectral analysis does a reasonably good job of signal detection but fares poorly in characterization because the phase information is ignored in the spectral computations. From spectral analysis, an initial or seed value for the period of the target signal is obtained. The original digitally sampled noisy time signal is resampled (up/down sample rate conversion) at such a rate that precisely an integer number of time samples are obtained on each cycle of the quasiperiodic target signal. These noisy cycles are synchronously averaged in the time domain and an estimate for the shape of the target signal is computed. The sample rate conversion, and the synchronous averaging is repeated until the rms value of the result is maximized (i.e., phase cancellation is minimized). The voltage signal-noise ratio of the reconstructed target signal is proportional to the square root of the number of target cycles averaged. This procedure was tested with laboratory type signals and found to have an excellent performance at signal-to-noise voltage ratios as low as -40 dB.