Convolutional Neural Networks Based Sparse Channel Estimation Algorithm for OFDM in Television Systems

In OFDM (Orthogonal Frequency Division Multiplexing) and MIMO (Multiple-in Multipleout) systems, obtaining accurate channel information is a crucial issue that needs to be addressed urgently. Only by obtaining accurate channel information can further functionality be achieved. This article adopted the Attention-FC (Attention-based Fully Connected) module and Attention-Conv (Attention-based Convolutional) module based on attention mechanism. Integrating the attention mechanism module into the residual network, the AttRNet-FC channel estimation models and AttRNet-Conv channel estimation models were constructed. Under different channel conditions, the maximum mean square error was 0.59, indicating that the AttRNet (Attention Residual Network) network had good robustness.

I. Introduction

Due to its high bandwidth utilization, effective resistance to multipath interference, narrowband interference, and other advantages, OFDM is considered a key technology in the industry for the new generation of high-speed broadband wireless transmission systems. It has not only been widely used in wireless local area networks, broadband access, and other aspects, but also as a standard modulation technology for high-definition digital television broadcasting systems. OFDM systems are very sensitive to frequency offset, so correlation detection is usually used. In broadband wireless communication, fluctuations in time and frequency domains can seriously affect signal transmission. Therefore, it is very necessary to estimate the parameters of OFDM channels in digital television.