Channel Prediction and PMI/RI Selection in MIMO-OFDM Systems Based on Deep Learning

Outdated channel information degrade capacity performance in limited feedback precoded MIMO-OFDM system. Channel prediction is a flexible way to combat channel aging without the cost of scarce time-frequency resources. Long short term memory (LSTM) is a class of neural networks, capable of processing data that has long-term dependence in time series. In this paper, an LSTM-based channel predictor is designed to achieve a longer range of predictions. Precoding matrix indicator (PMI) and rank indicator (RI) are two indicators related to precoding and the number of precoding matrices forming the codebook increased with the number of antenna ports. Furthermore, we try to calculate PMI/RI with the aid of the neural network. Through the simulation experiment, we test the performance of the LSTM channel predictor and the PMI/RI selector. The result illustrates that our predictor achieves considerable channel capacity gain and our selector shows high accuracy. Moreover, the generalization ability of the LSTM predictor for different antenna configurations is studied.