Channel Estimation in Massive MIMO under Hardware Non-linearities using Stacked BLSTM

Abstract

The fifth and beyond generations of wireless communication technologies utilize an extension of Multiple-Input Multiple-Output (MIMO) principles called Massive MIMO. This technology improves communication system performance with a greater number of base station antennas sending signal beams to users’ locations. However, if optimal channels are estimated without considering the distortion characteristics of the base station and user-side hardware, the overall performance can be reduced. Formerly, the non-linear hardware impairments impact channel estimation since this line of research is in its infancy. To overcome this issue, the proposed model is a recurrent neural network (RNN) particularly Stacked Bidirectional Long Short Term Memory (BLSTM). In addition, this model can handle the sequential nature of channel data and temporal correlations in time-varying channels. Eventually, it is observed that the model can reduce the channel estimation error for Massive MIMO hardware non-linearity scenarios. The proposed model has been outperforming its benchmark, which has utilized Fully-Connected Deep Networks (FCDN), for robust channel estimation by handling the structure of the non-linear distortions. Further, it can be inferred that the BLSTM model can be integrated into physical layer design for future wireless systems within Massive MIMO networks.