Neural network antenna processing for detection and direction finding

Abstract

The goal of neural beamforming is to design neural processing algorithms which adapt to low cost phased array antennas, even when they behave non-linearly, are imperfectly manufactured, or become degraded. Neural beamforming techniques have the potential to decrease antenna manufacturing and maintenance costs, while increasing mission time and performance before repair. The authors introduce a neural network architecture which performs signal detection and direction finding (DF) despite antenna degradations and failures. They first present the neural beamformer's (NBF) detection and DF performance on data from a simulated antenna model with added thermal noise across a range of signal-to-noise ratios (SNRs) and compare its DF accuracy to a classical monopulse technique, with and without calibration. They also test and compare the performance of the neural beamformer and monopulse on simulations of antenna models containing induced degradations and failures. Finally, they retrain the neural beamformer with measured antenna data from a 32-element X-band antenna and test its detection and DF performance, comparing DF results with a calibrated monopulse technique and against simulated antenna model performance for several SNRs.