Array Calibration Using Neural Networks

This paper addresses the problem of calibration for antenna arrays. The performance of array signal processing techniques, like the super-resolution direction finding method MUltiple SIgnal Classification, can significantly degrade if the assumed array model does not match the actual array response. In order to obtain an improved array model that accurately describes the actual array response, both parametric and non-parametric calibration methods have been proposed in the past. However, parametric methods often cannot fully capture the complex array response and non-parametric methods have an increased computational cost. This work proposes a new calibration method for array sensors based on neural networks. It achieves similar performance to non-parametric methods at moderate computational cost and offers a flexible applicability without the need to retain the neural network. The method is comprehensively studied in Monte-Carlo simulations and its performance is exemplarily evaluated for the multi-source direction finding problem.