SNR-adaptive deep joint source-channel coding for wireless image transmission

Mingze Ding,Jiahui Li,Mengyao Ma,Xiaopeng Fan
DOI: https://doi.org/10.48550/arXiv.2102.00202
2021-02-02
Abstract:Considering the problem of joint source-channel coding (JSCC) for multi-user transmission of images over noisy channels, an autoencoder-based novel deep joint source-channel coding scheme is proposed in this paper. In the proposed JSCC scheme, the decoder can estimate the signal-to-noise ratio (SNR) and use it to adaptively decode the transmitted image. Experiments demonstrate that the proposed scheme achieves impressive results in adaptability for different SNRs and is robust to the decoder's estimation error of the SNR. To the best of our knowledge, this is the first deep JSCC scheme that focuses on the adaptability for different SNRs and can be applied to multi-user scenarios.
Signal Processing,Image and Video Processing
What problem does this paper attempt to address?
This paper aims to solve the problem of joint source - channel coding (JSCC) in multi - user wireless image transmission. In particular, when transmitting images over noisy channels, how to design a deep joint source - channel coding scheme that can adapt to different signal - to - noise ratio (SNR) changes. Traditional methods usually optimize source coding and channel coding separately, but in multi - user scenarios or in cases of non - ergodic source or channel distributions, this method is not very effective. Therefore, the paper proposes a novel deep joint source - channel coding scheme based on auto - encoders. This scheme can estimate the signal - to - noise ratio at the decoding end and use this information to adaptively decode the transmitted images, thus improving the adaptability and robustness under different signal - to - noise ratio conditions. Specifically, this scheme introduces pilot signals as additional inputs to assist in the signal - to - noise ratio estimation during the decoding process, enabling the decoder to adjust the decoding strategy according to different signal - to - noise ratios. Experimental results show that the proposed scheme has excellent adaptability under different signal - to - noise ratio conditions and good robustness to the noise in the signal - to - noise ratio estimation at the decoding end. This is the first research on deep joint source - channel coding schemes for different signal - to - noise ratio adaptability and multi - user scenario applications.