Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Learning based Three-stage Solution for ISAC Beamforming Optimization

Qian Gao, Ruikang Zhong, Yuanwei Liu

TL;DR

The paper tackles joint downlink communication and radar sensing in a full-duplex ISAC system by reframing beamforming optimization as a beampattern design problem. It introduces a three-stage deep learning framework: (i) unsupervised CSI feature extraction via an Autoencoder, (ii) reinforcement learning (A2C) to optimize beampatterns in the latent space, and (iii) supervised learning to reconstruct beamforming vectors from the optimized beampatterns. The approach improves generalization, training stability, and interpretability compared to direct RL on beamformers, and experiments show higher sum-rate performance under sensing-rate constraints with ablation confirming the value of each module. This framework provides a scalable, implementable solution for ISAC beamforming in complex, high-dimensional MIMO systems. The results suggest practical benefits for real-time ISAC deployments by leveraging structured beampattern optimization and modular reconstruction.

Abstract

In this paper, a general ISAC system where the base station (BS) communicates with multiple users and performs target detection is considered. Then, a sum communication rate maximization problem is formulated, subjected to the constraints of transmit power and the minimum sensing rates of users. To solve this problem, we develop a framework that leverages deep learning algorithms to provide a three-stage solution for ISAC beamforming. The three-stage beamforming optimization solution includes three modules: 1) an unsupervised learning based feature extraction algorithm is proposed to extract fixed-size latent features while keeping its essential information from the variable channel state information (CSI); 2) a reinforcement learning (RL) based beampattern optimization algorithm is proposed to search the desired beampattern according to the extracted features; 3) a supervised learning based beamforming reconstruction algorithm is proposed to reconstruct the beamforming vector from beampattern given by the RL agent. Simulation results demonstrate that the proposed three-stage solution outperforms the baseline RL algorithm by optimizing the intuitional beampattern rather than beamforming.

Deep Learning based Three-stage Solution for ISAC Beamforming Optimization

TL;DR

The paper tackles joint downlink communication and radar sensing in a full-duplex ISAC system by reframing beamforming optimization as a beampattern design problem. It introduces a three-stage deep learning framework: (i) unsupervised CSI feature extraction via an Autoencoder, (ii) reinforcement learning (A2C) to optimize beampatterns in the latent space, and (iii) supervised learning to reconstruct beamforming vectors from the optimized beampatterns. The approach improves generalization, training stability, and interpretability compared to direct RL on beamformers, and experiments show higher sum-rate performance under sensing-rate constraints with ablation confirming the value of each module. This framework provides a scalable, implementable solution for ISAC beamforming in complex, high-dimensional MIMO systems. The results suggest practical benefits for real-time ISAC deployments by leveraging structured beampattern optimization and modular reconstruction.

Abstract

In this paper, a general ISAC system where the base station (BS) communicates with multiple users and performs target detection is considered. Then, a sum communication rate maximization problem is formulated, subjected to the constraints of transmit power and the minimum sensing rates of users. To solve this problem, we develop a framework that leverages deep learning algorithms to provide a three-stage solution for ISAC beamforming. The three-stage beamforming optimization solution includes three modules: 1) an unsupervised learning based feature extraction algorithm is proposed to extract fixed-size latent features while keeping its essential information from the variable channel state information (CSI); 2) a reinforcement learning (RL) based beampattern optimization algorithm is proposed to search the desired beampattern according to the extracted features; 3) a supervised learning based beamforming reconstruction algorithm is proposed to reconstruct the beamforming vector from beampattern given by the RL agent. Simulation results demonstrate that the proposed three-stage solution outperforms the baseline RL algorithm by optimizing the intuitional beampattern rather than beamforming.
Paper Structure (21 sections, 25 equations, 3 figures)

This paper contains 21 sections, 25 equations, 3 figures.

Table of Contents

  1. Introduction
  2. System Model and Problem Formulation
  3. Signal Model
  4. Communication and Sensing Rates
  5. Problem Formulation
  6. Proposed Solution
  7. Unsupervised learning based CSI Feature Extraction
  8. CSI Dataset Generation
  9. Data Preprocessing
  10. Network Structure
  11. RL based Beampattern Optimization
  12. State
  13. Action
  14. Reward Function
  15. Network Structure
  16. ...and 6 more sections

Figures (3)

  • Figure 1: Illustration of the FD ISAC system.
  • Figure 2: Workflow of the proposed ICT framework.
  • Figure 3: Hyperparameter selection and performance comparison.