Deep Unfolded Extended Conjugate Gradient Method for Massive MIMO Processing with Application to Reciprocity Calibration

In this paper, we consider deep unfolding the standard iterative conjugate gradient (CG) algorithm to solve a linear system of equations. Instead being adjusted with known rules, parameters are learned via backpropagation yield optimal results. However, proposed unfolded CG (UCG) is extended wherein scalar parameter substituted by matrix-parameter augment degrees freedom per layer. Once training completed, UCG has revealed require far smaller number layers than iterations needed using CG. It also shown be very robust noise and outperforms in low signal ratio (SNR) region. A key merit approach fact that no explicit data dedicated learning phase as optimization process relies on residual error which not explicitly expressed function desired data. As an example, applied reciprocity calibration problem encountered massive MIMO (Multiple-Input Multiple-Output) systems.