Total complex kernel risk-sensitive loss for robust DOA estimation

Abstract

Adaptive filtering-based approaches have been proposed for low-complexity direction of arrival (DOA) estimation. Nonetheless, existing methods demonstrate significant performance deterioration in bias compensation models that incorporate impulse noise. To address this challenge, we have introduced a novel similarity measure in kernel space, termed total complex kernel risk-sensitive loss (TCKRSL), which effectively extracts higher-order statistics from the data to mitigate the detrimental effects of outliers caused by impulse noise. Subsequently, we derived a robust adaptive filtering algorithm known as the minimum total complex kernel risk-sensitive loss (MTCKRSL) algorithm based on stochastic gradient descent and applied it to DOA estimation via adaptive nulling array antenna. To further enhance estimation performance, we implemented a variable step size (VSS) mechanism grounded in cumulative instantaneous error and the estimated signal power aimed at balancing the trade-off between steady-state error and convergence speed, resulting in the VSS-MTCKRSL algorithm. Additionally, the convergence properties and computational complexity of the proposed algorithm were analyzed elaborately. Simulation results across various performance metrics demonstrate that the proposed VSS-MTCKRSL algorithm outperforms the state-of-the-art algorithms regardless of the presence of impulse noise or Gaussian noise.