GRACEFUL DEGRADATION OF AN ADAPTIVE BEAMFORMING PROCESSOR

The phase and amplitude weights of each element in an adaptive, digitally beamformed array antenna can be calculated by a systolic parallel processor. We propose that the entire weight calculation (both QR decomposition, previously discussed by Gentleman and Kung, and the back-substitutions, which we present here for the first time) be performed in a triangular array of processors. This has the highly desirable property of graceful degradation: the loss of one or more component processors does not cause the computed weights to degrade catastrophically. The loss of a processor in the systolic array merely reduces the number of available degrees of freedom. Moreover, the processing array does not need to be globally reconfigured, the functions of all the remaining processors are unaffected and their interconnections are undisturbed. We describe the processing flow in the systolic array, show how its degradation is inherently graceful and present a realistic example of graceful degradation. The specific example demonstrates full null depth even when one processing element in the array is totally nonfunctional This property is extensible to array processors with more than one failed element; null depth is preserved at the expense of degrees of freedom in shaping the main beam.