Goal-oriented Spectrum Sharing: Trading Edge Inference Power for Data Streaming Performance

We study the problem of spectrum sharing between goal-oriented (GO) and legacy data-oriented (DO) systems. For the former, data quality and representation is no longer optimized based on classical communication key performance indicators, but rather configured on the fly to achieve the goal of communication with the least resource overhead. This paradigm can be followed to flexibly adapt wireless and in-network artificial intelligence operations across different nodes (e.g., access points, users, sensors or actuators) to data traffic, channel conditions, energy availability and distributed computing capabilities. In this paper, we argue and demonstrate that computing and learning/inference operation performance strongly affect lower layers, calling for a real cross-layer optimization that encompasses physical and computation resource orchestration, up to the application level. Focusing on a communication channel shared among a GO and a DO user, we define a goal-effective achievable rate region (GEARR), to assess the maximum data rate attainable by the latter, subject to goal achievement guarantees for the former. Finally, we propose a cross-layer dynamic resource orchestration able to reach the boundaries of the GEARR, under different goal-effectiveness and compute resource consumption constraints.