Learning the MMSE Channel Estimator

This webinar will discuss the MMSE channel estimator for a simple SIMO system model, without knowledge of the required channel statistics. Although the derived MMSE estimator is computationally intractable in the general form, its structure can be used to motivate a neural network architecture with lower complexity. The complexity reduction is based on a set of assumptions on the system model that simplify the MMSE estimator. The performance of the simplified MMSE estimator degrades significantly when those assumptions are not met. In contrast, a neural network based on the simplified MMSE estimator can compensate the mismatch by learning from data of the actual channel model. The presenters will also show how to extend the simple SIMO model to other practically relevant scenarios. Finally, they will demonstrate the performance when the neural network is learned based on actual channel measurements as compared to simulated data. patterns and systems for signals that are naturally tensors, e.g., images and video. For a concrete application, they show that functional magnetic resonance imaging (fMRI) acceleration is a tensor sampling problem, where design of practical sampling schemes and an algorithmic framework are used to handle it. Numerical results show that their tensor sampling strategy accelerates the fMRI sampling process significantly without sacrificing reconstruction accuracy.