Hybrid Precoding Design in MmWave MIMO Systems Using Manifold Learning-Based Extreme Learning Machine Framework

To facilitate real-time processing of high-dimensional signals in massive MIMO systems, a novel algorithm called Manifold Learning-based Extreme Learning Machine (ML-ELM) has been introduced. This algorithm enables hybrid precoding design without the need for channel estimation. Specifically, our approach leverages uplink received signals to capture the underlying low-dimensional manifold structure of the channel, resulting in a significant reduction in data volume. Using the learned representation, we employ an ELM network for nonlinear fitting, with the output label representing Zero-forcing precoding. Traditional optimization techniques are then employed to calculate the hybrid analog and digital precoding given the Zero-forcing precoding. Importantly, the overall process exhibits low complexity. The superior performance of our proposed algorithm in terms of real-time processing capability and high spectral efficiency has been demonstrated through extensive simulations.