Beam Alignment in Multipath Environments for Integrated Sensing and Communication using Bandit Learning

Prior works have explored multi-armed bandit (MAB) algorithms for the selection of optimal beams for millimeter-wave (mmW) communications between base station and mobile users. However, when the number of beams is large, the existing MAB algorithms are characterized by long exploration times, resulting in poor overall communication throughput. In this work, we propose augmenting the upper confidence bound (UCB) based MAB with integrated sensing and communication (ISAC) to address this limitation. The premise of the work is that the radar and communication functionalities share the same field-of-view and that communication mobile users are detected by the radar as mobile targets. The radar information is used for significantly reducing the number of candidate beams for the UCB, resulting in an overall reduction in the exploration time. Further, the radar information is used to estimate the realignment time in quasi-stationary scenarios. We have realized the MAB and radar signal processing algorithms on the system on chip (SoC) via hardware-software co-design (HSCD) and fixed-point analysis. We demonstrate the significant gain in execution time using accelerators. The simulations consider complex propagation channels involving direct and multipath, with simple and extended radar targets in the presence of significant static clutter. The resulting experiments show that the proposed ISAC-based MAB achieves a 35% reduction in the overall exploration time and 1.4 factor higher throughput as compared to the conventional MAB that is based only on communications.

💡 Research Summary

The paper addresses the critical problem of rapid beam alignment in millimeter‑wave (mmWave) communication systems, where narrow analog beams must be steered quickly toward mobile users (MUs) in environments rich with multipath and clutter. Conventional approaches that rely solely on communication feedback—typically acknowledgments (ACKs) from the MU—use multi‑armed bandit (MAB) algorithms such as Upper Confidence Bound (UCB) to discover the optimal beam. However, when the codebook contains hundreds of beams, the exploration phase becomes prohibitively long, degrading overall throughput and latency.

To overcome this limitation, the authors propose an Integrated Sensing and Communication (ISAC) framework that couples radar sensing with the UCB‑based MAB. The key idea is that the same antenna array and analog beamforming hardware are shared between radar and communication functions, so the radar can scan the same set of beams as the communication system. By processing the reflected radar signal, the base station (BS) obtains estimates of each MU’s range and Doppler (velocity). These estimates are used in two ways:

1. Candidate‑Beam Reduction – The radar’s location estimate narrows the search space to a small subset of beams that are geometrically aligned with the MU. The UCB algorithm is then applied only to this reduced set, dramatically cutting the number of arm pulls required for exploration.

2. Realignment Prediction – In quasi‑stationary scenarios, the radar continuously monitors the MU’s range‑Doppler trajectory. A lightweight change‑detection module predicts when the MU’s position has drifted enough to warrant a new beam‑search cycle, eliminating unnecessary re‑exploration.

The authors implement the entire pipeline on a system‑on‑chip (SoC) platform that combines a multi‑core CPU with an FPGA. Radar signal processing (RSP) and the bandit learning logic are realized in fixed‑point arithmetic, and the FPGA hosts dedicated accelerators for FFT, matched filtering, and UCB updates. Fixed‑point conversion reduces arithmetic complexity by roughly 30 % compared with floating‑point, while hardware acceleration yields a two‑fold reduction in overall latency.

Simulation experiments model realistic propagation: a direct line‑of‑sight component, multiple reflected paths, static clutter, and extended targets (e.g., vehicles modeled as clusters of scatterers). The performance metrics evaluated include total exploration time, average data throughput, and the accuracy of the predicted realignment instant. Results show that, relative to a baseline communication‑only UCB MAB, the ISAC‑enhanced approach achieves a 35 % reduction in exploration time and a 1.4× increase in throughput. The change‑detection module correctly predicts the need for re‑alignment in over 95 % of cases, further preventing wasted scans.

The paper also surveys related work in three categories: (i) ISAC‑based beam alignment, (ii) bandit‑learning‑based beam selection, and (iii) joint radar‑communication hardware designs. It highlights that prior ISAC studies either required separate radar hardware, full‑duplex operation, or heavy computational loads, while earlier bandit works ignored spatial sensing information. By integrating radar sensing directly into the bandit decision process, the proposed method bridges this gap.

Finally, the authors discuss practical considerations and future directions. The approach assumes reliable radar detection; in low‑SNR conditions the candidate‑beam set may not shrink sufficiently, suggesting adaptive radar waveform design as a remedy. Extending the framework to other ISAC waveforms (e.g., OFDM‑based) and to multi‑user scenarios with simultaneous target tracking are identified as promising research avenues.

In summary, the paper delivers a novel, hardware‑aware ISAC‑enabled bandit learning scheme that substantially accelerates beam alignment in mmWave systems, offering a concrete pathway toward meeting the stringent latency and throughput demands of upcoming 5G/6G networks.