Deep learning based channel estimation for OFDM system

• Main Author: Ahmad Erfan Hilme Haji Bakri
• Other Authors: Teh Kah Chan
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Electrical and electronic engineering::Electronic systems::Signal processing
• Online Access: https://hdl.handle.net/10356/166982
• Institution: Nanyang Technological University

Channel estimation is a critical component in wireless communication systems, including orthogonal frequency division multiplexing (OFDM) systems. Traditional methods for channel estimation often have limitations in terms of accuracy and performance, particularly in complex wireless environments. Throughout many different applications, deep learning (DL) has proven itself to be a reliable tool to be integrated in our technologies. Our study is directed to showcase the performance of using DL techniques, specifically deep neural network (DNN) involving long short-term memory (LSTM) layers that has shown promise over other similar applications. We investigated the efficiency and robustness of this technique in comparison to its predecessors least square (LS) and minimum mean square error (MMSE) techniques. In our constructed approach, we produced the dataset and processes it through training and testing of the model for the OFDM signals with varying number of pilots. Our study demonstrates the effectiveness of deep learning with LSTM layers in improving the adaptability and reliability of channel estimation in OFDM systems. The initial results suggest that this approach is a valuable tool in future wireless communication systems. We further stressed our results by varying parameters such as length of cyclic prefix (CP) as well as varying the modulation constellation between binary phase-shift keying (BPSK) and quadrature phase-shift keying (QPSK). Overall, our extensive study has demonstrated that lower lengths of CP with QPSK modulation produces the most optimum results. In the endeavour of achieving a more concrete result, we recommend further testing and evaluation on other DL variations to ensure the robustness and efficiency of the technique. The amount of data required for a DL channel estimation would be massive thus further iterations to combat this issue would be more ideal. Additionally, incorporating this technique with others would also be a viable option to look into to outweigh the cons of each of the methods used.