Abstract:

Accurate channel estimation is a critical component in the design of 5G OFDM communication system receivers, since it can significantly reduce the bit error rate (BER), thus improving wireless communication efficiency and quality.Channel estimation methods based on least square (LS) and minimum mean square error (MMSE) effectively utilize the system’s sparsity, but LS algorithms face low computational precision, while MMSE algorithms suffer from high computational complexity.To promote the estimation accuracy, practitioners have presented several deep learning-based channel estimation methods.However, existing methods often split complex matrices into real and imaginary parts, failing to adequately capture the complex characteristics of the channel, leading to distortion in the estimated channel matrix.A complex-valued generative adversarial network (GAN) model that could fully extract the complex features of the signals was proposed, enabling accurate estimation of the channel matrix for the physical downlink shared channel (PDSCH) in the 5G new radio (NR) standard.To validate the effectiveness of the proposed method, the proposed method was compared with LS algorithms, actual channel estimation, super-resolution neural networks, and residual neural network channel estimation methods.Results show that when the mean square error between the estimated channel matrix and the true channel matrix is 0.01, the proposed method-based communication system has a signal-to-noise ratio (SNR) that is 5 dB higher than existing ones.

Key words: 5G new radio, channel estimation, PDSCH, complex valued neural network, generative adversarial network