Two-Way Semantic Communications Without Feedback
Kaiwen Yu, Qi He, Gang Wu
Abstract
Semantic communications can significantly improve transmission efficiency, especially in the low signal-to-noise (SNR) regime. However, two-way semantic communications still remain an unexplored topic, which is a critical issue in many machine communication scenarios. Simply extending existing semantic communication systems to the two-way situation requires bidirectional information feedback during the training process, resulting in significant communication overhead. To fill this gap, we investigate a two-way semantic communication (TW-SC) system, where the information feedback can be omitted by exploiting the weight reciprocity in the transceiver. Particularly, the channel simulator and semantic transceiver are implemented on both TWSC nodes and the channel distribution is modeled by a conditional generative adversarial network. Simulation results demonstrate that the proposed TW-SC system performs closing to the stateof-the-art one-way semantic communication systems but requiring no feedback between the transceiver in training process.