A flexible approach and VLSI architecture for computing discrete time-frequency distribution

Time-frequency distribution (TFD) based on Cohen's class has significant potential for the analysis of a number of non-stationary signals, including speech, biomedical signals and machine vibration. The implementation of discrete signals has led to the definition of many different discrete TFD. The spectral decomposition of the kernel allows the computation of the generalized discrete-time TFD (GDTFD) as a weighted sum of spectrograms with orthonormal windows. The partial sum offers the user a vehicle to perform trade-off between exactness and computational requirement. We extend this idea by showing that the combination of incremental refinement and approximating structure along with the spectral decomposition-based GDTFD provides more flexibility and a variety of trade-offs between approximation quality and computational cost. The paper discusses aspects of a suitable VLSI architecture featuring modular structure, local connections and reconfigurable logic module for implementation. The result serves as a basis to design a programmable core for computing GDTFD within an integrated mechatronic system targeting fault detection.