Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Massive MIMO Adaptive Modulation and Coding Using Online Deep Learning Algorithm

Evgeny Bobrov, Dmitry Kropotov, Hao Lu, Danila Zaev

TL;DR

The paper tackles adaptive modulation and coding (AMC) in massive MIMO by introducing an online neural-network-based AMC (ODL) that predicts the ACK probability for each MCS from SINR and CQI information and selects the MCS maximizing the expected throughput. Unlike Q-learning, the method frames AMC as a binary classification problem with online updates using a sample buffer and the Adam optimizer, achieving competitive throughput improvements while remaining simple and compliant with 5G NR. System-level simulations show ODL outperforms both OLLA and Q-learning across varying channel types and UE speeds, delivering about 10–20% higher throughput in full-buffer traffic. The approach operates entirely at the base station, with feasible computational and storage requirements, offering a practical, robust solution for dynamic beamforming in massive MIMO deployments.

Abstract

The paper describes an online deep learning algorithm (ODL) for adaptive modulation and coding in massive MIMO. The algorithm is based on a fully connected neural network, which is initially trained on the output of the traditional algorithm and then incrementally retrained by the service feedback of its output. We show the advantage of our solution over the state-of-the-art Q-learning approach. We provide system-level simulation results to support this conclusion in various scenarios with different channel characteristics and different user speeds. Compared with traditional OLLA, the algorithm shows a 10\% to 20\% improvement in user throughput in the full-buffer case.

Massive MIMO Adaptive Modulation and Coding Using Online Deep Learning Algorithm

TL;DR

The paper tackles adaptive modulation and coding (AMC) in massive MIMO by introducing an online neural-network-based AMC (ODL) that predicts the ACK probability for each MCS from SINR and CQI information and selects the MCS maximizing the expected throughput. Unlike Q-learning, the method frames AMC as a binary classification problem with online updates using a sample buffer and the Adam optimizer, achieving competitive throughput improvements while remaining simple and compliant with 5G NR. System-level simulations show ODL outperforms both OLLA and Q-learning across varying channel types and UE speeds, delivering about 10–20% higher throughput in full-buffer traffic. The approach operates entirely at the base station, with feasible computational and storage requirements, offering a practical, robust solution for dynamic beamforming in massive MIMO deployments.

Abstract

The paper describes an online deep learning algorithm (ODL) for adaptive modulation and coding in massive MIMO. The algorithm is based on a fully connected neural network, which is initially trained on the output of the traditional algorithm and then incrementally retrained by the service feedback of its output. We show the advantage of our solution over the state-of-the-art Q-learning approach. We provide system-level simulation results to support this conclusion in various scenarios with different channel characteristics and different user speeds. Compared with traditional OLLA, the algorithm shows a 10\% to 20\% improvement in user throughput in the full-buffer case.

Paper Structure

This paper contains 10 sections, 6 equations, 8 figures, 1 table.

Table of Contents

  1. Introduction
  2. System Model
  3. Structure of the proposed algorithm
  4. Neural Network Model
  5. Proposed algorithm complexity
  6. The structure of the neural network
  7. System-level simulation results
  8. Quality improvement with machine learning
  9. ODL and Q-learning performance comparison
  10. Conclusion

Figures (8)

  • Figure 1: Online Deep Learning algorithm block scheme.
  • Figure 2: Working Algorithm Time Axis.
  • Figure 3: Algorithm Sample Buffer.
  • Figure 4: The scheme of both Online Deep Learning (proposed) and Q-learning with a sample buffer. The key difference between the algorithms is in the different activation and loss functions.
  • Figure 5: Block diagram of the neural network used.
  • ...and 3 more figures