Complexity Reduction of the Super-Resolution Iterative Reweighted Algorithm

Future wireless communication require accurate channel estimation (CE) to guarantee the system performance. Compressive sensing is one of the approaches to CE. It allows to accurately reconstruct the channel and provide significant pilot overhead reduction compared to other methods. To apply the conventional compressed sensing technique the continuous parameter space has to be discretized to a finite set of grid points, based on which a sensing dictionary is constructed for signal recovery. However, true parameters do not necessarily lie on the discretized grid, which leads to grid mismatch and deteriorated recovery performance. To solve the problem, the sensing dictionary is parametrized and the parameters are learned during estimation. This approach is called super-resolution and it noticeably improves CE. However, popular super resolution algorithms have high computational complexity compared to conventional compressed sensing algorithms. In this paper, by simplifying the analytical expression for the gradient and avoiding direct calculation of inverse matrices, we greatly reduce the complexity of the original SURE-IR algorithm. Simulation results show that the proposed optimization yields 50–650 times complexity reduction and even provides 5–8 dB accuracy improvement. This optimization brings super resolution algorithms closer to widespread use in consumer devices.