Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Machine Learning (ML)-assisted Beam Management in millimeter (mm)Wave Distributed Multiple Input Multiple Output (D-MIMO) systems

Karthik R M, Dhiraj Nagaraja Hegde, Muris Sarajlic, Abhishek Sarkar

TL;DR

Beam management in mmWave D-MIMO faces prohibitive overhead if every AP/beam is sounded. The paper evaluates ML-based beam inference using RF, MissForest, and c-GAN on data generated from a 3D map and ray tracing, aiming to predict unmeasured $L1$-RSRP values and select top-$K$ beams with partial measurements; the total number of beams is $A=MN$. Results show that RF and MF provide strong performance under high missingness, closely matching the ideal best-beam distribution, while a c-GAN approach is more sensitive to data size but is expected to scale with more data. The work demonstrates substantial overhead reduction for BM in mmWave D-MIMO and points to future directions such as dynamic environment adaptation, priors from AP geometry, and explainable/causal inference for beam decisions.

Abstract

Beam management (BM) protocols are critical for establishing and maintaining connectivity between network radio nodes and User Equipments (UEs). In Distributed Multiple Input Multiple Output systems (D-MIMO), a number of access points (APs), coordinated by a central processing unit (CPU), serves a number of UEs. At mmWave frequencies, the problem of finding the best AP and beam to serve the UEs is challenging due to a large number of beams that need to be sounded with Downlink (DL) reference signals. The objective of this paper is to investigate whether the best AP/beam can be reliably inferred from sounding only a small subset of beams and leveraging AI/ML for inference of best beam/AP. We use Random Forest (RF), MissForest (MF) and conditional Generative Adversarial Networks (c-GAN) for demonstrating the performance benefits of inference.

Machine Learning (ML)-assisted Beam Management in millimeter (mm)Wave Distributed Multiple Input Multiple Output (D-MIMO) systems

TL;DR

Beam management in mmWave D-MIMO faces prohibitive overhead if every AP/beam is sounded. The paper evaluates ML-based beam inference using RF, MissForest, and c-GAN on data generated from a 3D map and ray tracing, aiming to predict unmeasured -RSRP values and select top- beams with partial measurements; the total number of beams is . Results show that RF and MF provide strong performance under high missingness, closely matching the ideal best-beam distribution, while a c-GAN approach is more sensitive to data size but is expected to scale with more data. The work demonstrates substantial overhead reduction for BM in mmWave D-MIMO and points to future directions such as dynamic environment adaptation, priors from AP geometry, and explainable/causal inference for beam decisions.

Abstract

Beam management (BM) protocols are critical for establishing and maintaining connectivity between network radio nodes and User Equipments (UEs). In Distributed Multiple Input Multiple Output systems (D-MIMO), a number of access points (APs), coordinated by a central processing unit (CPU), serves a number of UEs. At mmWave frequencies, the problem of finding the best AP and beam to serve the UEs is challenging due to a large number of beams that need to be sounded with Downlink (DL) reference signals. The objective of this paper is to investigate whether the best AP/beam can be reliably inferred from sounding only a small subset of beams and leveraging AI/ML for inference of best beam/AP. We use Random Forest (RF), MissForest (MF) and conditional Generative Adversarial Networks (c-GAN) for demonstrating the performance benefits of inference.
Paper Structure (10 sections, 15 figures, 1 table)

This paper contains 10 sections, 15 figures, 1 table.

Table of Contents

  1. Introduction and System Model
  2. Beam Inference and simple ML algorithms
  3. Explored Methodologies
  4. Random Forest (RF)
  5. Miss Forest (MF)
  6. Generative Adversarial Networks (GANs)
  7. Performance Evaluation
  8. Performance Evaluation measures
  9. Model performance
  10. Conclusion

Figures (15)

  • Figure 1: Analog beamforming in mmWave D-MIMO
  • Figure 2: Beam candidate inference
  • Figure 3: Stages of Directed Search
  • Figure 4: Architecture of the proposed c-GAN.
  • Figure 5: Comparison of CDF plots of L1-RSRP between top-$k$ predictions and ideal ones
  • ...and 10 more figures