Deep Learning For Channel Estimation And Signal Detection

Channel estimation and signal detection are essential steps in communication systems, they are used along with different modulation techniques such as orthogonal frequency-division multiplexing (OFDM) to ensure the success of the system as whole. There are some conventional methods such as least square (LS) that have been widely used in channel estimation for OFDM systems, the problem with LS is that it is not accurate enough especially in low signal-to-noise ratio (SNR) regions. Deep learning (DL) can be a possible solution to solve many problems in wireless communications. This project proposes a DL model for channel estimation and signal detection in OFDM systems. We adopted a deep neural network (DNN) model to provide an effective solution to increase the accuracy with minimum complexity. The performance of the DNN model is evaluated and then compared with LS method simulations in terms of BER under different SNRs. The comparison shows that DL proves to have a better performance compared with LS simulations.