Optimized deep learning based hypernet convolution neural network and long short term memory for joint pilot design and channel estimation in MIMO‐OFDM model

Abstract

AbstractIn multiple input multiple output‐orthogonal frequency division multiplexing (MIMO‐OFDM) systems, efficient pilot design (PD) and channel estimation (CE) greatly influences the reliability and robustness of pilot‐based CE methods. But, accurate estimationof the channel remains a challenge in a high‐mobility environment with non‐linear channel characteristics. Many techniques have been introduced to overcome the pilot contamination and CE problems in the MIMO‐OFDM system to overcome this issue. However, these techniques take multiple paths at the receiver, resulting in delay spread and interference in the communication. Hence, this study presents a novel deep learning (DL) based technique for channel estimation based on channel state information (CSI). A DL technique based on a hyper convolutional neural network (Hyper‐CNN) is introduced for the optimal pilot design. The selection of the pilot position can be made using the tunicate swarm optimization (TSO) approach. Finally, a DL‐based long short‐term memory (LSTM) model is proposed for the CE in the MIMO‐OFDM system. The proposed method is implemented in the MATLAB platform, and the outcome is compared under different metrics like mean square error (MSE) and bit error rate (BER). In the experimental scenario, the proposed method attains the BER of 0.20 and 0.03 for CP (cyclic prefix) and without CP, respectively. In addition, the proposed method attains the MSE of 1.14, 0.99, 1.08 and 0.97 for 8, 16, 48 and 64 pilots, respectively. The performance of a proposed method is compared with the existing method and proves the efficacy of the proposed method.