Deep learning approaches for channel length estimation in blind multichannel systems

Blind channel estimation and equalization techniques aim to estimate the channel or equalizer using only the received signal without any prior knowledge of the transmitted input signal. However, most of these techniques fail completely when the channel length is misspecified. Hence, channel length knowledge is critical in aiding blind techniques to accurately determine the channel and produce optimal results. This work proposes a novel deep learning approach to estimate channel lengths in blind single-input multiple-output (SIMO) systems. Utilizing the symmetric properties of the covariance matrix, two architectures, covariance regression network (CovRNet) and covariance classification network (CovCNet), are proposed for regression and classification tasks, respectively. Simulations demonstrate significant improvements in accuracy and generalization compared to traditional methods across various signal-to-noise (SNR) levels, highlighting the potential of deep learning for efficiently aiding channel length estimation in blind systems.