Answers are grounded in this paper and cite their sources.
We introduce a novel and elegant modification to the multiple-input multiple-output (MIMO) least-mean-squares (LMS) deconvolution algorithm of Sturm et al. [1], which is based on a linear time-invariant (LTI) framework. Furthermore, an alternative mean-square error (MSE) convergence criterion is proposed, which overcomes the difficulties of the criterion given by Sturm et al. [2] for multiple output-class systems. The modified algorithm exploits both parallelisation and segmentation to address challenges associated with non-stationary signals, as well as those with time-varying signal-to-noise ratio (SNR), where local MSE gradients in the LMS estimation can differ significantly from the global MSE gradient, and hence Sturm’s algorithm is often unable to resolve temporal regions where SNR is low, or where impulsive noise is present. By parallelising the algorithm, the time complexity is reduced by approximately a factor of 𝑅 − 1 , where R is the number of signal partitions. Through formal analysis, a synthetic experiment over a wide range of parameters, as well as a practical case study, we demonstrate that the modified algorithm provides an order of magnitude reduction in compute time, whilst augmenting estimation accuracy in most cases, making application to systems of practical complexity much more accessible. Furthermore, the proposed convergence criterion proves to be effective in the analysis of a single-input, multiple-output (SIMO) system.
No takes yet. Share an insight, caveat, or question.
Jesse Mccarthy (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: