Hardware Implementation of Deep Learning Based Channel Estimation for OFDM-IM Systems
DOI:
https://doi.org/10.59287/as-proceedings.505Keywords:
Channel Estimation, Deep Learning, OFDM-IM, Hardware Implementation, PYNQ-Z1, Least Squares (LS), Minimum Mean-Square Error (MMSE)Abstract
This paper describes the hardware implementation of deep learning-based channel estimation for index modulated orthogonal frequency division multiplexing (OFDM-IM) systems in Rayleigh fading channel conditions. The channel response is estimated using a deep neural network (DNN) in the hardware implementation. The proposed DNN is trained through the channel coefficient obtained using the least squares (LS) method. Subsequently, channel estimation is performed through the trained DNN. In the proposed DNN, the hidden layer includes the Long Short Term Memory (LSTM) layer. Hardware testing is conducted under various scenarios whereby the DNN model is compared to traditional channel estimation methods. Results indicate that deep learning-based channel estimation vastly outperforms LS and MMSE techniques.