Research Topics
Research Topics
Modern wireless systems aim to support many users simultaneously using large antenna arrays operating at millimeter-wave frequencies. However, classical multiuser detection algorithms scale poorly as the number of antennas and users grows. Our work investigates how beamspace signal processing, obtained by applying a spatial FFT across antenna arrays, can dramatically reduce computational complexity while maintaining strong performance.
We develop a geometric and information-theoretic understanding of beamspace dimensionality reduction, showing that the energy of sparse multipath channels concentrates in a small number of spatial frequency bins. This enables scalable multiuser detection using small beamspace windows that do not grow with the array size for both narrowband and wideband communication systems.
To quantify the trade-offs between performance and complexity, we establish information-theoretic benchmarks comparing ideal unconstrained receivers with low-complexity beamspace architectures, using both measured and simulated channel models. Our results demonstrate that beamspace processing can achieve performance close to theoretical limits while drastically reducing computational cost, providing design guidelines for scalable massive MU-MIMO systems.
Next-generation radar systems are moving toward fully digital arrays with hundreds or even thousands of antennas, enabling simultaneous beamforming toward multiple targets while suppressing interference and clutter. However, classical adaptive beamforming algorithms scale poorly with array size, making conventional spatial processing computationally infeasible for massive arrays.
Our work investigates scalable beamforming architectures for wideband massive MIMO radar by exploiting energy concentration in the beam domain. By transforming the received signals into beamspace using spatial FFTs and selecting small beamspace windows that capture the dominant angular energy of each target, adaptive beamforming can be performed in a significantly reduced dimension while maintaining strong detection performance.
To further enable scaling to extremely large arrays, we develop tiled beamspace architectures in which the full antenna array is partitioned into smaller subarrays (tiles) that perform local beamspace processing. Each tile applies spatial transforms and reduced-dimension adaptive filtering, while coordination across tiles enables the system to combine information from the entire array. This distributed architecture allows radar systems with arrays of up to 1024 elements to perform adaptive interference suppression and target detection with manageable computational complexity. By combining beamspace dimension reduction with parallel processing across tiles, this approach provides a practical pathway toward scalable digital beamforming for next-generation large-scale radar platforms.
This research studies hybrid beamforming systems for wideband multiuser MIMO using tiled antenna arrays. In wideband systems, conventional narrow beams suffer from beam squint, where the beam direction varies with frequency and reduces array gain across the signal bandwidth. To address this issue, we investigate broadbeam designs that maintain consistent gain across frequency while relying on phase-only RF beamforming within each tile.
By combining tile-level analog beamforming with digital signal processing across tiles, this architecture enables efficient multiuser communication while keeping hardware complexity manageable. The study also explores different beamforming strategies and tile allocation methods, providing insights into practical design trade-offs for wideband hybrid beamforming systems.
Millimeter-wave phased arrays rely on accurate phase calibration to synthesize reliable beam patterns for communication and sensing. However, manufacturer-provided codebooks are typically designed only for steering beams in a discrete set of directions and do not directly provide the calibration information needed to generate more general beam patterns.
This research investigates a zero-shot calibration framework named EiCal that extracts phase calibration information directly from the advertised beamforming codebook, without requiring any additional measurements. By leveraging eigen-analysis of the codebook structure, the approach enables accurate generation of custom beam patterns, including compressive sensing beams and beams with nulls, while avoiding the cost and difficulty of conventional calibration procedures.