A New Regularized Adaptive Windowed Lomb Periodogram for Time-Frequency Analysis of Nonstationary Signals With Impulsive Components

This paper proposes a new class of windowed Lomb periodogram (WLP) for time-frequency analysis of nonstationary signals, which may contain impulsive components and may be nonuniformly sampled. The proposed methods significantly extend the conventional Lomb periodogram in two aspects: 1) The nonstationarity problem is addressed by employing the weighted least squares (WLS) to estimate locally the time-varying periodogram and an intersection of confidence interval technique to adaptively select the window sizes of WLS in the time-frequency domain. This yields an adaptive WLP (AWLP) having a better tradeoff between time resolution and frequency resolution. 2) A more general regularized maximum-likelihood-type (M-) estimator is used instead of the LS estimator in estimating the AWLP. This yields a novel M-estimation-based regularized AWLP method which is capable of reducing estimation variance, accentuating predominant time-frequency components, restraining adverse influence of impulsive components, and separating impulsive components. Simulation results were conducted to illustrate the advantages of the proposed method over the conventional Lomb periodogram in adaptive time-frequency resolution, sparse representation for sinusoids, robustness to impulsive components, and applicability to nonuniformly sampled data. Moreover, as the computation of the proposed method at each time sample and frequency is independent of others, parallel computing can be conveniently employed without much difficulty to significantly reduce the computational time of our proposed method for real-time applications. The proposed method is expected to find a wide range of applications in instrumentation and measurement and related areas. Its potential applications to power quality analysis and speech signal analysis are also discussed and demonstrated.