Power Adaptation for Goal-Oriented Communication over Fading Channels

This paper proposes a transmit power adaptation design for goal-oriented communication to improve the goal-acquisition performance over fading channels. The specific contribution lies in the power adaptation module which is capable of determining an appropriate transmit power based on the instantaneous channel condition, while satisfying an average power constraint. Moreover, it is a standalone add-on module that can be used in conjunction with different encoders. To illustrate its performance, we consider an image classification task as an example of goal acquisition and adopt the deep-learning-based joint source-channel coding for transmitting the goal-related information. We numerically illustrate the difference between the power adaptation design for goal-oriented communication and the traditional power adaptation design for bit-oriented communication. Our results demonstrate the performance gain from using the proposed power adaptation design and provide insights into conditions under which such a performance gain becomes either negligible or notable.