An improved Viterbi algorithm for instantaneous frequency extraction of overlapped multicomponent signals

Instantaneous frequency extraction of multicomponent signals with time frequency overlapped components has been a challenging task in radar, sonar and other applications. Viterbi algorithm on time frequency distribution can be potentially applied to estimate instantaneous frequencies of multicomponent signals, but it may track wrong instantaneous frequency curves when signal components intersect each other in the time frequency domain. In order to suppress this switch problem, this paper assumes the estimated instantaneous frequency curves on a finite interval can be approximated by liner, quadratic or higher-order polynomials, which should not vary seriously after instantaneous frequencies cross the entire time frequency plane. Then, a novel penalty function describing the variation of polynomial instantaneous frequencies is developed. Furthermore, the fitting technique based on least square method is applied to calculate the penalty between the predicted instantaneous frequency point and instantaneous frequency candidates. Performance of the proposed algorithm on several simulated multicomponent frequency modulated signals indicates the proposed algorithm can obtain more accurate instantaneous frequencies than Viterbi algorithm can by suppressing the switch problem.