DPM-Solver-2M: A Fast Multistep DPM-Solver-Based Scheme for Real-Time MIMO Channel Estimation

Abstract

Real-time multiple-input multiple-output (MIMO) channel estimation poses a major challenge due to stringent low-latency constraints. We propose a multistep fast ordinary differential equation (ODE)-based diffusion probabilistic model (DPM), DPM-Solver-2M, for MIMO channel estimation, significantly reducing the number of inference steps while maintaining high accuracy. Unlike conventional discrete-time DPMs, our approach reformulates a lightweight discrete noise prediction model into a continuous-time framework, enabling ODE-based fast multistep solvers with efficient numerical methods. This scheme retains the advantages of discrete models, such as low complexity, while achieving a $\sim 4\times $ speedup in terms of inference steps or number of function evaluations (NFE) during the reverse process over Markovian or ancestral sampling-based DPM estimators, with only a marginal performance trade-off. Theoretical analysis of solver convergence corroborates our simulation results, demonstrating rapid convergence in just 10–15 solver steps, making our approach highly suitable for real-time wireless systems. In addition to achieving strong performance under ideal conditions, our simulation results reveal that the proposed model is robust to changes in the signal-to-noise ratio (SNR) of the observed (received) signal and generalizes well across different wireless channel models for the small number of steps.