FedISAC: A Federated Learning Framework with Integrated Sensing and Communication for 6G mmWave Networks

The 6G communication networks are designed to provide secure communication, higher data throughput, minimized power consumption, improved system performance, and enhanced device integration, paving the way for more intelligent and efficient networking systems. To achieve these goals, an AI framework is proposed that integrates the integrated sensing and communication (ISAC) with the federated learning (FL) scheme, where local devices send their sensing information to the global base station (GBS) after local training. The GBS then aggregates this local sensing information and allocates the desired aggregated average power to the local devices based on the aggregated sensing information of the local devices. An optimization problem is formulated to minimize the global model loss, ensure desired power allocation, and maintain improved signal-to-interference-plus-noise ratio (SINR) and achievable rate (AR). An AI framework is proposed, utilizing a federated averaging (FedAvg) algorithm to address the formulated problem and allocate the necessary power to local devices based on their sensing information. Simulation results reveal that our FedAvg-based AI framework achieves cumulative SINR improvements of $1.02 \mathrm{~dB}, 1.01 \mathrm{~dB}$, and 1.01 dB, and $A R$ enhancements of 1.71 $\mathrm{bps} / \mathrm{Hz}, 1.70 \mathrm{bps} / \mathrm{Hz}$, and $1.70 \mathrm{bps} / \mathrm{Hz}$, outperforming federated proximal, centralized training, and average local training methods, respectively.