Channel Estimation in Underwater MIMO-OFDM Systems Using FF-DNN Neural Network

The underwater acoustic channel is known as one of the most challenging communication channels due to its physical nature. In this regard, the use of orthogonal frequency division modulation (OFDM) and multiple input-multiple output (MIMO) systems are effective methods to overcome channel effects and increase the capacity of the underwater channel. In this way, the performance of these telecommunication systems and the achievement of the mentioned benefits are significantly dependent on the estimation of coefficients and the acquisition of channel state information. Considering that in most researches, the channel between the transmitter and the receiver is assumed to be thin, while in practical applications this is not the case; In this article, two feed-forward deep neural network (FF-DNN) models, Net_1 and Net_2, have been used to estimate the coefficients of multi-tap communication channels in underwater channels. The process is as follows: first, the least square (LS) estimate of the channel is obtained, and then it is applied as an input to these two neural network models, the model is trained and learned, and the output of the least mean square estimate is Error (MMSE) of expected channel coefficients. The results obtained from the simulation show that the use of these two deep neural network models with different numbers of hidden layers by overcoming the LS estimation based on the comparative criteria of MSE and BER has a good performance compared to the MMSE estimation. and increases the quality of coefficient estimation. For example, based on the BER criterion, the presented models have improved by 3 and 5.5 dB respectively for an error value equal to 2-10.