The research disciplines whose use I advocate include inference, learning, and optimization. The aim is to model, design, and optimize different types of systems. There are numerous thrust areas of my research group.

The following samples a few key thrusts of my research, with selected publications in each area.

Machine Learning for Physical Layer

Traditionally, physical layer algorithm design was based on accurate linear system models. Information theory has played a central role in motivating the use of adaptation to improve spectrum efficiency drastically. Recently, AI/ML-based physical layer design has received substantial attention due to its exploitation of existing data and simple addon solutions. However, instead of making analytical contributions to understanding AI/ML-based physical layer techniques, many existing AI/ML-based approaches throw away the theory while relying heavily on training.

Of particular interest are optimization methodologies that engage performance insights to understand the fundamental limits of simple AI/ML-based physical layer techniques. We engage information theory, coding theory, and optimization theory to quantify data-driven enhancement to theory-based physical layer techniques. We have been looked at possible avenues for approaching “previously unsolvable problems” in signal processing and information theory such as IRS control, denoising, channel output feedback capacity, and approximate message passing problems to be addressed by AI/ML techniques.

Related Publications

Feature Learning and Unsupervied Domain Adaptation

We propose a dual-module network architecture that employs a domain discriminative feature module to encourage the domain invariant feature module to learn more domain invariant features. This technique can be applied to any model that utilizes domain invariant features for unsupervised domain adaptation to improve its ability to extract domain invariant features.

Kolmogorov model (KM) learning has recently received great attention from source separation community due to its interpretability (in terms of extracting additional information or insights that are hidden inside the data) and predictive power. we for the first time enabled the application of KM learning in “big data” regimes, something that was previously impractical. The contributions are also applicable to a large family of data sets with missing entries.

Related Publications

Millimeter Wave, Tera Hertz Hybrid Precoding and Channel Estimation

The so-called “beam alignment” is a procedure for finding the best transmit-and-receive beam pair to establish a link without estimating the channel state information (CSI). However, the use of directional narrow beams for searching the entire beam space is an extremely time-consuming operation.

In this thrust area, we have been focusing on developing theories and algorithms for the purpose of achieving fast beam alignment and tracking (refinement) performance by leveraging advanced channel representation on virtual beamspace, unit probability simplex, Krylov subspace, and restricted isometry. Recently, the emphasis has been placed on the learnability and interpretability of the beam refinement process and its time-varying causality with CSI estimation.

Related Publications

Distributed Interference Management, Resource Allocation, and Edge Computing

In this thurst, we have been developing distributed interference mitigation and resource allocation algorithms for multicell multiuser (MCMU) networks. The focus is providing performance guarantees with the minimum possible resource utilization. This thrust has been providng sufficient conditions for the feasibility of the proposed distributed algorithms by developing precoding, combining, power control, subchannel allocation, and relay selection frameworks. The underlying principle is decomposing the origianl sum rate maximization or performance-guranteed power minimization problems into forward and backward (FB) subproblems to ensure desired properites of the original problems such as strong duality, convexity, and convergence.

Related Publications

High Frequency Network Planning

Although mmWave cellular systems can carry a larger volume of traffic, the recent increase in the cost of deploying and maintaining mmWave small-cell BSs is a practical concern that wireless service providers are constantly facing.

My group has been studying methodologies to the problem of mmWave BS deployment in urban environments by minimizing BS deployment cost subject to BS association and user equipment (UE) outage constraints by integrating physical blockage, UE access-limited blockage, and signal-to-interference-plus-noise-ratio (SINR) outage into its quantification. Interestingly, the proposed methods produce a unique distribution of the macro-diversity orders over the network that is distinct from other benchmark deployments.

Related Publications

Algorithms on Manifolds

My group has been focusing on developing theories and methodologies for representing large-dimensional multiple-input multiple-output (MIMO) channels on lower-dimensional subspaces by engaging manifold structure. Many algorithms on manifold bring expressiveness and accountability benefits, which can be used for developing optimized methods with considerably reduced complexity. My research has successfully engaged various manifold structures (Grassmannian, Riemannian, unit-simplex, etc.) for MIMO precoding, beamforming, channel estimation, and detection.

Related Publications