A flexible subband-based computation scheme for generalized discrete time-frequency distributions

Time-frequency distribution (TFDs) based on Cohen´s class has significant potential for the analysis of nonstationary signals. However, high computational requirement limits its use in applications where hardware resources are costly. This paper proposes an algorithm which provides mechanism to trade-off computational complexity against exactness for the computation of generalized discrete TFDs (GDTFDs). The approach employs a linear signal expansion by means of the wavelet transform to make the signal sparse, thus simplifying subsequent analysis and processing. It then restricts the accuracy of the computation to some predefined measures using the data dependent wavelet shrinkage via hard thresholding. The paper discusses the complexity of the approach and shows its efficiency under some mild constraints. The trade-off flexibility allows the implementation of the high computational tasks in a power-efficient and cost-effective way tailored to specific application demands