Machine learning based CDMA power control

This paper presents binary and multiclass machine learning techniques for CDMA power control. The power control commands are based on estimates of the signal and noise subspace eigenvalues and the signal subspace dimension. Results of two different sets of machine learning algorithms are presented. Binary machine learning algorithms generate fixed-step power control (FSPC) commands based on estimated eigenvalues and SIRs. A fixed-set of power control commands are generated with multiclass machine learning algorithms. The results show the limitations of a fixed-set power control system, but also show that a fixed-set system achieves comparable performance to high complexity closed-loop power control systems.