Prior Model-based Deep Joint Source-Channel Coding for Wireless Image Transmission over Multipath Fading Channels
Feng Wang, Xuechen Chen, Xiaoheng Deng · 2024
Deep Joint Source Channel Coding (DeepJSCC) achieves remarkable results in wireless image transmission. However, image transmission under multipath fading channels based on DeepJSCC is still a huge challenge. Therefore, we propose a subchannel matching strategy based on prior model and DeepJSCC over multipath fading channels. According to channel state information (CSI), our scheme maps the more important image features obtained from the prior model to higher quality subchannels. In practice, the resources can be fully utilized according to the acquired CSI, resulting in higher image transmission quality. Extensive experiments show that our approach achieves state-of-the-art performance and channel resilience over multipath fading channels compared to existing DeepJSCC schemes. In addition, our scheme has good channel adaptability and does not suffer from the cliff effect in the case of channel signal-to-noise ratio (CSNR) of additive Gaussian white noise (AWGN) mismatch.