Split Spectrum Processing with Computationally Efficient Sinusoidal Fir Filters

Abstract

Split spectrum processing (SSP) is the process of reducing material noise based on the decomposition of ultrasonic signals by multiple narrow band bandpass filtering and statistical compounding. Conventionally, the filter bank utilized to perform the splitting of the spectrum of the signal is made up of equally spaced Gaussian filters of constant bandwidth. In the frequency domain, the processing time of the technique depends primarily on the number of filters in the bank required for a desired SNR performance. A feasible way to reduce the size of the filter bank (relative to the Gaussian filter bank) is to increase the energy confined within the half-power bandwidth of the filter without substantially increasing the correlation between adjacent or neighboring filters, that is to say, increase the number of effectively uncorrelated filters such that the corresponding increase in correlation is more than offset by the increase of target-signal energy. This paper presents a new filtering scheme or rather a modification to the old scheme that may, under certain conditions, require less filters for a desired SNR enhancement performance than its Gaussian counterpart. The rational for the selection of the processing parameters is established and experimental results from computer-simulated ultrasonic signals are presented.