Transfer Learning Model for Joint Semantic and Channel Coding/Decoding in Wireless Systems

With significant increase in data and demand for large bandwidth, the wireless research community has identified the need to focus on 'what' to transmit rather than focusing on only 'how' to transmit. Inspired by this, in the current article a view of wireless systems based on semantic communications is put-forth. Specifically, a semantic wireless system enabled by deep learning is proposed which aims to maximize the system capacity. As opposed to evaluating only the bit/symbol errors, proposed technique is able to recover the meaning of sentences and is hence able to minimize the semantic errors. Further, transfer learning is implemented for accelerating the process of re-training. Extensive simulations to validate performance of the proposed technique demonstrate that it is able to (i) maintain enhanced robustness to variations in the channel, and (ii) achieve higher performance. Overall, the current study makes it evident that the proposed technique is a good candidate for implementation in the semantic wireless systems.