Channel Estimation of Massive MIMO-OFDM System Using Elman Recurrent Neural Network

Bandwidth limitations in the wireless communication bands have motivated the investigation and exploration of wireless access technologies like massive multiple-input multiple-output (MIMO) networks. The performance of MIMO networks dramatically depends upon the techniques exploited for channel estimations. The existing methods of channel estimation have failed to resolve the inter-symbol interference (ISI) effects efficiently. This paper presents a new channel estimation method based on machine learning. The presence of ISI in the MIMO-orthogonal frequency division multiplexing (MIMO-OFDM) network introduces errors in the decision device at the receiver. This study aims to limit the impact of ISI in the transmitting and receiving filter designs to convey digital data with the lowest error rate. The Elman recurrent neural network (E-RNN) algorithm was employed herein to estimate the channel in MIMO-OFDM considering reliability and scalability. A low peak-to-average power ratio (PAPR), reduced bit error rate (BER), high capacity, high throughput, and improved mean squared error (MSE) performance are achieved using the E-RNN approach. The obtained PAPR value for the proposed E-RNN is 0.1272 after 40 epochs. Variations of cumulative distribution function (CDF) for various channel capacities are plotted. Also, these channel estimation parameters exploiting recurrent neural network (RNN), artificial neural network (ANN), convolutional neural network (CNN), and deep neural network (DNN) methods are compared.