High Rate Communication over One-Bit Quantized Channels via Deep Learning and LDPC Codes

Abstract: This paper proposes a method for designing error correction codes by combining a known coding scheme with an autoencoder. Specifically, we integrate an LDPC code with a trained autoencoder to develop an error correction code for intractable nonlinear channels. The LDPC encoder shrinks the input space of the autoencoder, which enables the autoencoder to learn more easily. The proposed error correction code shows promising results for one-bit quantization, a challenging case of a nonlinear channel. Specifically, our design gives a waterfall slope bit error rate even with high order modulation formats such as 16-QAM and 64-QAM despite one-bit quantization. This gain is theoretically grounded by proving that the trained autoencoder provides approximately Gaussian distributed data to the LDPC decoder even though the received signal has non-Gaussian statistics due to the one-bit quantization.