This work explores an energy-efficient integrated sensing, communication, and computing (ISCC) system for privacy-preserving wireless federated learning (FL) at edge devices in 6G and beyond wireless networks. We propose an efficient user selection method based on local computing time concerning highpower processing to enhance user participation and employing reconfigurable intelligent surface (RIS) to improve channel reliability in the wireless FL system. To maximize the energy efficiency (EE), we leverage a nonlinear optimization problem as a mixedinteger nonlinear program (MINLP) with constraints on power, computing frequency, bandwidth, and target sensing metric based on beamforming design matrices. The MINLP is transformed into a tractable nonlinear program, approximated via linear problems using Taylor expansion and solved with an iterative algorithm based on successive linear programming (SLP). Simulation results demonstrate that the proposed wireless FL scheme outperforms the benchmarks. Moreover, the proposed non-orthogonal multiple access (NOMA)-enabled system significantly increases energy efficiency.
