SNR Estimation Method based on SRS and DINet
Guohua Yao, Zhuhua Hu · 2023
In conventional SNR estimation, the energy of the useful signal cannot be accurately calculated because of the noise in the received signal. At the same time, because of the random nature of noise, how to accurately estimate the noise is a common challenge in the engineering community. To address this problem, this paper proposes a signal-to-noise ratio (SNR) estimation method that combines the sounding reference signal (SRS) with the deep learning network DINet, where DINet is composed of denoising convolutional neural network (DnCNN) and image restoration convolutional neural network (IRCNN) in parallel. To demonstrate the higher estimation performance of our proposed method, we replicate some advanced algorithms, such as Boumard's algorithm, Qun X et al.'s improved algorithm, and M2M4 algorithm, and in the paper we refer to Boumard's algorithm and Qun X et al.'s improved algorithm as algorithm 1 and algorithm 2, respectively. At the transmitter side, to address the randomness of the noise distribution, we map the SRS into a nine-box grid on the resource block, and this arrangement facilitates a more accurate estimation of the noise and signal. At the receiver side, the SNR of each nine-box grid is calculated firstly using algorithm 1. Then the SNR is put into the corresponding resource unit and linear interpolation is performed on the resource block. Finally, the resource blocks are equated to images and input to DINet for denoising to obtain a more accurate value of the SNR estimate. Experiments show that the proposed method in this paper has a significant performance improvement compared with algorithm 2 and M2M4 algorithm.