A Tensor Decomposition-Based Censored Regression Adaptive Filtering Algorithm

The article delves into the challenges posed by censored measurements in adaptive filtering, where only a portion of the system response is directly acquirable. Traditional adaptive estimators struggle with such data, leading to performance degradation or divergence due to sample selection bias. The proposed censored regression tensor least mean-square (CR-TLMS) algorithm addresses these issues by leveraging tensor decomposition to enhance convergence performance. The article provides a comprehensive performance analysis, including mean convergence and steady-state mean-square error (MSE) derivations, offering practical guidelines for implementation. Through numerical experiments involving system identification under censored measurement environments, the article demonstrates the superior convergence rate and performance advantages of the CR-TLMS algorithm compared to existing methods. The integration of tensor decomposition with censored regression models presents a novel approach to handling partial observability, making this article a compelling read for those interested in advancing adaptive filtering techniques.