Optimal Sequence Detection for a One-Bit ADC Receiver Using a Quasi-Neural Network

This letter studies the maximum likelihood sequence detection (MLSD) for a communication receiver equipped with one-bit analog-to-digital converters (ADCs) in an intersymbol interference (ISI) channel. For a conventional receiver, one-bit quantization complicates the channel estimation and the subsequent detection/decoding. We propose the so-termed quasi-neural network (Quasi-NN) for the MLSD, requiring no explicit estimation of the channel state information (CSI). Through the artful design of the inputs, the hidden layers, and the activation functions, we construct the Quasi-NN that can be trained to learn the likelihoods of the 1-bit outputs of the ADC(s) under different state transitions. Similar to a conventional artificial neural network (ANN), the Quasi-NN has the layered structure. But in contrast to a conventional ANN, the Quasi-NN is compactly structured and fully interpretable; thus, it can be well-trained using the backpropagation (BP) algorithm with a pilot of a moderate length. The simulations show that our method outperforms the conventional method based on channel estimation.