Decorrelation in Range of Oversampled Weather Radar Signals Using FIR Filter

Due to rapid advances in digital technology and the demands of the telephone industry, digital receivers suitable for use in weather radars have become readily available. These receivers have the capability to produce samples of weather signals at rates that are several times higher than the reciprocal of the transmitted pulse width τ. It is natural to assume that, given the sampling rate of L/τ (where L is a positive integer) one could improve the estimates’ variance by averaging L sample-time autocorrelations (at lags 0 and 1) in range. Unfortunately, simple averaging does not yield the maximum variance reduction because samples are correlated in range. Torres and Zrnić (2003) have proposed an approach that uses the prior knowledge of autocorrelation along range to decorrelate samples. The scheme operates on blocks (i.e., vectors) of L samples in range whereby each vector is multiplied by an L×L matrix producing a new vector of uncorrelated samples. Thus, each spectral moment is estimated from an L×M block (M is the number of samples along sample-time) of samples, where samples along range-time are not correlated. Consequently, range-time averaging of L autocovariances results in an optimal variance reduction given by the oversampling factor. In this paper, an alternative realization of the same concept is presented. Namely, the use of FIR (Finite Impulse Response) filters to obtain decorrelated samples in range is investigated. The study has been motivated by the fact that many digital receiver processors feature built-in FIR filters with programmable coefficients (e.g., GC4016 digital receiver chip from Texas Instruments).