Radar system processing gain and computational efficiency improvement using wavelet-based signal processing techniques.

摘要

In this research, signal processing techniques are developed to improve detection performance and/or efficiency of the detection process in radar applications. The problem of detecting radar signals in noise often reduces to the problem of detecting sinusoids of unknown frequency spread over a continuous region corrupted by additive white Gaussian noise (1). The optimal solution is to design a matched filter for each signal of interest. This results in maximum processing gain (PG) for each signal of interest. Each signal of interest is a modulated version of a reference signal, and thus each matched filter is a modulated version of the reference signal's matched filter. Together, these matched filters can be viewed as a filterbank. Because of the theoretically infinite number of filters and the limited computational resources available in a radar, the filterbank size must be significantly reduced. For most radar systems, this reduced filterbank is implemented by a fast Fourier transform (FFT) (2). The FFT is a finite set of matched filters equally spaced in frequency. Sinusoids at frequencies different from the FFT frequencies will suffer processing gain degradation. Detection performance is directly related to processing gain. This research offers two new alternative algorithms which improve the average and/or worst case processing gain/detection performance when the optimal solution is unfeasible. These algorithms are demonstrated in a radar system.;The first algorithm involves a complete replacement of the filters in the reduced size filterbank (3). In this new design, none of the individual filters are matched to any possible return signal. The filters are chosen to improve the average and/or minimum detection probability over the range of possible received signals. A new, efficient filterbank design algorithm has been developed to implement this new filterbank.;The second technique is to process the reduced size filterbank output (4). A one level orthonormal discrete wavelet transform using the Haar wavelet is applied to the filterbank output. This algorithm is shown to improve the worst case detection probability while adding little computational cost to the system.;Finally we note that a wavelet based detection and estimation algorithm (5) was also applied to this problem. Before this algorithm could be used for this application, its false alarm probability had to be computed. In its present form, the algorithm's false alarm probability is too high for use in radar systems (6). A modification to this algorithm with encouraging preliminary results has been presented. This is an area of future research.