Abstract
In this paper, a novel channel and content preference feedbacks enabled semantic communication framework that enables a transmitter to use limited channel and content for efficient and accurate image data transmission is developed. In the proposed framework, a base station (BS) divides an image into several sub-images according to the image content and extracts the semantic information from sub-images. Then the BS allocates the semantic information to different sub-channels and transmits them to a user. After receiving the semantic information, the user regenerates the image and selects to transmit a feedback (i.e., channel state information (CSI) feedback or content preference feedback) to the BS. The BS will use this feedback information to optimize sub-channel allocation and feedback selection. Meanwhile, since the number of feedback that the receiver can transmit is limited, the BS must determine when to receive a feedback as well as which type of the feedback should be transmitted according to the dynamic wireless environment. We formulate this problem as an optimization problem whose goal is to minimize the mean square error (MSE) between the original image and the regenerated image via optimizing sub-channel allocation and feedback selection. To address this problem, an actor-critic based reinforcement learning (RL) with dynamic neighborhood construction (AC-DNC) scheme is proposed. Compared to the current RL methods, the proposed method is more effective in finding better actions in huge discrete action spaces since the designed method approximates the discrete neighboring action space using a continuous action and selects an action with the maximum Q value from the neighboring action space. Simulation results show that the proposed scheme can improve the performance by up to 27.14% and 26.87% compared to the proposed scheme without CSI and content preference feedback information and the regular Q actor-critic (QAC) method.