WSN optimization for sampling-based signal estimation using semi-binarized variational autoencoder

•A semi-binarized deep learning framework is proposed to optimize the low-dimensional representation for signal estimation.•A Kronecker Delta function is designed to binarize the signal encoding and generate SPOR.•The binarized encoding layer is shown to be backpropagatable, which can optimize SPOR iteratively according to the performance of the estimation.•The discrepancy between the low-dimensional representation and SPOR is minimized to improve the estimation performance.