Modification of the ICI rule-based IF estimator for high noise environments

The nonparametric algorithm for instantaneous frequency (IF) estimation, based on the intersection of the confidence intervals (ICI) rule and the Wigner distribution (WD), is modified in order to produce an accurate IF estimate for a high noise environment. The original approach is developed under the assumption that a small amount of noise can move the WD maxima only within the auto-term. This is not true for high noise environments. The probability that the WD maxima are outside the auto-term is high for narrow windows used in the WD calculation. Estimates obtained with these windows are used as an initial guess in the adaptive algorithm. In this paper, we set the initial estimate as the one produced by the narrowest window for which the probability of error due to high noise is smaller than a threshold. The error probability is estimated based on the estimates of signal amplitude and noise variance. The IF estimates for some windows are additionally improved by applying a median filter directly to the IF estimate.