Investigation and development of DSP algorithms/hardware for real time power spectral density estimation

摘要

This research is concerned with the Power Spectrum Density Estimation with em-phasIze on the bigh-resolution algorithms and their real-time implementations.Tl-ie classical PSD estimation methods are fast and robust. but their resolutions maynot be adequate when the record length is short. On the other hand when therecord length is short the autoregressive parametric methods have higher resolutioncapability, but they may have spurious peaks if the order of the model is chosen toohigh in the attempt to increase the resolution when the SNR is low. An algorithmis proposed to combine the spectrum of the classical method and the autoregressivemodel. This allows the overestimation of the order of the autoregressive model. Thespuriot-is peaks that result are then suppressed by the low values in the spectrum of theclassical nict liods. I'lic wide specl ral mairilobe of the classical method, on the otherliand, serves to indicate the area where the true signals are located. This alleviates thedifficult order selection problem of the parametric methods. An adaptive version ofthis method is also proposed. It is based on the adaptive autoregressive and adaptivemaximum eigenvector concept. It can track a slowly changing environment. WithI lie combination of these txN, o methods. it is shown that it. has the high-resolutionperformance of AR method ýN, ith improved performance in the noisy environment.