Quasi-Neural Network based Sequence Detection for Single-Carrier Communications
Qinghe Du, Chenye Wang, Yi Jiang, Rong Ran · 2024
This paper proposes a Quasi-neural network-based detection algorithm, namely QNN-detector, for a single-carrier communication system in an inter-symbol interference (ISI) channel, which can be modeled by a trellis diagram. According to the trellis diagram, a quasi-neural network (QNN) is built to acquire the normalized likelihoods to enable the subsequent detection or decoding. The QNN can accommodate non-Gaussian interferences through ingeniously designing its hidden layers. Unlike the artificial neural network (ANN) based algorithms, which are data-driven, the QNN-detector relies on the physical system model and only needs a short pilot sequence for training. Moreover, it requires neither explicit channel state information (CSI) nor statistics of interference and noise. Simulation results illustrate that in a channel under white Gaussian noise, the QNN-detector significantly outperforms the ANN-based algorithms in that its network training requires a far shorter pilot sequence, and can approach the performance limits provided by the Viterbi detector with perfect CSI. The simulations also show that in the presence of non-Gaussian interferences, the QNN-detector can learn the distribution of the interferences and therefore suppress them effectively, while the conventional Viterbi detector fails to.