The main objective of this dissertation is to develop an adaptive beamformer which is highly fast and numerically stable, and is capable of nulling out coherent interferences.
Signal reception using an array of sensor elements is currently the subject of considerable interest because an array antenna system affords the means of break-throughs overcoming the directivity and resolution limitations of a single sensor element. The beam pattern of the adaptive array is automatically adjusted to create nulls in the directions of interfering signals, while passing a desired, look-direction signal with minimum distortion.
Conventional beamformers so far studied require the assumption that interfering signals are not correlated with the desired signal. When this underlying assumption is no longer valid, the adaptive beamformer not only generates false nulls, but also tends to cancel the look-direction signal completely, resulting in a poor performance. In order to cope with the performance degradations due to the coherence between the desired signal and interferences, the spatial smoothing technique has been used. Although the method is found to be effective in combatting coherent interferences, it is disadvantageous in that the method significantly reduce the effective array aperture. To use the array aperture efficiently, the modified spatial smoothing technique has been developed. But, the method still forms covariance matrices as is the case of the original spatial smoothing technique, which difficulties when the wordlength is finite. Accordingly, to prevent the numerical instability caused by the spatial smoothing technique and its variations, one must have twice as long wordlength as the original data set particularly when the given data is ill-conditioned.
In this dissertation, we present a data-domain modified spatial smoothing (DMSS) technique, by which the inter-signal correlation can be removed, and the formation of covariance matrices can be avoided. Since the proposed method can readily be combined with the parallel spatial processing scheme or least-squares solving systems using orthogonal transformations, one can take full advantage of low cost, high density, and fast speed very large scale integration (VLSI) applications using a systolic/wavefront array to get the high throughput. Furthermore, with the increase effective array aperture, the proposed method provides an improved nulling capability against near-field interferences.
In adaptive beamforming, what we are ultimately concerned is not determining the adaptive weight vector, but recovering the signal waveform that is as close as possible to the desired signal. Recently, systolic array architectures for adaptive beamforming applications have been investigated. They yield output signals without explicit computation of sample covariance matrices or adaptive weights, resulting in high throughput. Since the proposed DMSS technique can be used in implementing the modified spatial smoothing technique without forming covariance matrices, systolic arrays can easily be adopted for its hardware implementations. We suggest a modified parallel spatial processing scheme which can easily be incorporated with the proposed DMSS preprocessing technique and is highly suitable for systolic array implementations. The resulting beamformer yields high nulling capability, high throughput, and good numerical stability, while nulling out coherent interferences.
The main objective of this dissertation is to develop an adaptive beamformer which is highly fast and numerically stable, and is capable of nulling out coherent interferences.
Signal reception using an array of sensor elements is currently the subject of considerable interest because an array antenna system affords the means of break-throughs overcoming the directivity and resolution limitations of a single sensor element. The beam pattern of the adaptive array is automatically adjusted to create nulls in the directions of interfering signals, while passing a desired, look-direction signal with minimum distortion.
Conventional beamformers so far studied require the assumption that interfering signals are not correlated with the desired signal. When this underlying assumption is no longer valid, the adaptive beamformer not only generates false nulls, but also tends to cancel the look-direction signal completely, resulting in a poor performance. In order to cope with the performance degradations due to the coherence between the desired signal and interferences, the spatial smoothing technique has been used. Although the method is found to be effective in combatting coherent interferences, it is disadvantageous in that the method significantly reduce the effective array aperture. To use the array aperture efficiently, the modified spatial smoothing technique has been developed. But, the method still forms covariance matrices as is the case of the original spatial smoothing technique, which difficulties when the wordlength is finite. Accordingly, to prevent the numerical instability caused by the spatial smoothing technique and its variations, one must have twice as long wordlength as the original data set particularly when the given data is ill-conditioned.
In this dissertation, we present a data-domain modified spatial smoothing (DMSS) technique, by which the inter-signal correlation can be removed, and the formation of covariance matrices can be avoided. Since the proposed method can readily be combined with the parallel spatial processing scheme or least-squares solving systems using orthogonal transformations, one can take full advantage of low cost, high density, and fast speed very large scale integration (VLSI) applications using a systolic/wavefront array to get the high throughput. Furthermore, with the increase effective array aperture, the proposed method provides an improved nulling capability against near-field interferences.
In adaptive beamforming, what we are ultimately concerned is not determining the adaptive weight vector, but recovering the signal waveform that is as close as possible to the desired signal. Recently, systolic array architectures for adaptive beamforming applications have been investigated. They yield output signals without explicit computation of sample covariance matrices or adaptive weights, resulting in high throughput. Since the proposed DMSS technique can be used in implementing the modified spatial smoothing technique without forming covariance matrices, systolic arrays can easily be adopted for its hardware implementations. We suggest a modified parallel spatial processing scheme which can easily be incorporated with the proposed DMSS preprocessing technique and is highly suitable for systolic array implementations. The resulting beamformer yields high nulling capability, high throughput, and good numerical stability, while nulling out coherent interferences.