ASSENT: Learning-Based Association Optimization for Distributed Cell-Free ISAC

This repository contains the official implementation of ASSENT (ASSociation and ENTity selection) — a learning-based framework for efficient association optimization in distributed cell-free Integrated Sensing and Communication (ISAC) systems.
The framework is developed as part of the research paper:

"ASSENT: Learning-Based Association Optimization for Distributed Cell-Free ISAC"
Mehdi Zafari and A. Lee Swindlehurst
Accepted for publication at IEEE International Conference on Communications (ICC) 2026.

---

Overview

ASSENT addresses the association problem in distributed cell-free ISAC networks, where a central server determines:

• Which access points (APs) serve which users and sensing targets,
• Each AP’s operating mode (transmit or receive),
• And the scheduling of sensing entities —
  all under fronthaul capacity, RF-chain, and interference constraints.

The system first formulates the problem as a Mixed-Integer Linear Program (MILP) to obtain optimal associations. These optimal solutions are then used to supervise a graph neural network (GNN) that learns to predict association decisions efficiently for new environments.

---

Repository Structure

distributed-cellfree-isac/
├── src/
│   ├── assent/                 # ASSENT code
│   │       ├── checkpoints     # Model parameters saved for different runs
│   │       └── figures/        # Figures for learning evaluation
│   ├── utils/                  # Utility functions and simulation classes
│   ├── optimization/           # MILP formulation and dataset generation
│   │       ├── exp1_baseline/  # Baseline MILP solution
│   │       ├── exp1_pareto/    # Pareto evaluation for MILP solutions
│   │       └── exp2_data_gen/  # MILP-based dataset generation
│   ├── learning_v1/            # GNN-based learning framework (v1)
│   ├── learning_v2/            # GNN-based learning framework (v2)
│   ├── learning_v3/            # GNN-based learning framework (v3)
│   └── learning_v4/            # GNN-based learning framework (v4) --> Current version
│           ├── cache_graphs/   # Cached graphs (precomputed)
│           ├── checkpoints/    # Model parameters saved for different runs
│           └── figures/        # Figures generated for learning evaluation
├── notebooks/                  # Jupyter notebooks for analysis and visualization
├── LICENSE                     # License file
├── requirements.txt            # Python dependencies
└── README.md                   # Repository overview

---

Optimization Module

The MILP problem is implemented in Python using Gurobi.
It optimizes a weighted sum objective of normalized communication and sensing utilities, with a tunable trade-off parameter α.

Key optimization outputs include:

• τ — AP mode selection (TX/RX)
• x — AP-user association
• s — target scheduling
• y_tx, y_rx — AP-target transmit/receive associations

The MILP generates optimal solutions used as training data for the learning framework.
All optimization scripts and dataset generation tools are located in:

src/optimization/

---

Learning Framework

All the modules for the GNN-based learning framework are implemented in src/learning_v4/. Ready-to-run scripts for ASSENT are in the src/assent/ directory.

Features:

• Heterogeneous Graph Representation of the ISAC network
  (APs, Users, and Targets as node types)
• Multiple Message-Passing Architectures:
  • NNConv
  • GATv2Conv
  • TransformerConv
• Learnable Task Uncertainty for multi-task loss balancing
• Penalized Loss Function with RF-chain and coupling regularization
• Comprehensive Evaluation using F1, Precision, Recall, and Brier score metrics

---

Quickstart

Setup

Clone the repository and install dependencies:

git clone https://github.com/LS-Wireless/ASSENT-CellFree-ISAC.git
cd ASSENT-CellFree-ISAC
pip install -r requirements.txt

Import Utilities

Make sure to import the necessary utilities from the utils module:

import src.utils.visualization_utils as viz
import src.utils.optimization_utils as opt
import src.utils.network_utils as net

Running Simulations

Below are simple examples of how to write scripts to generate network layout, run the MILP optimization, and visualize the results.

netparams = net.NetworkParams(N_ap=8, M_a=16, N_RF=4, N_cu=10, N_tg=4)
network = net.NetworkEnvironment(netparams)
network.generate_topology()
network.plot_topology()

G_comm, S_comm = network.generate_commLink_features()
G_sens = network.generate_sensLink_features()

optparams = opt.ProblemParams(G_comm=G_comm, G_sens=G_sens, S_mat=S_comm, alpha=0.5)
optparams.change(lambda_cu=1.0, lambda_tg=1.0)
solution = opt.solve_problem(optparams)

viz.print_solution_summary(solution)
# System Snapshot
viz.plot_system_snapshot(ap_pos=network.ap_positions, user_pos=network.user_positions, target_pos=network.target_positions, sol=solution)

---

Citation

If you use this repository in your research, please cite:

Paper

@inproceedings{zafari2026assent,
  title     = {ASSENT: Learning-Based Association Optimization for Distributed Cell-Free ISAC},
  author    = {Zafari, Mehdi and Swindlehurst, A. Lee},
  month     = {January},
  year      = {2026},
  note      = {accepted at IEEE ICC 2026}
}

Code/Repository

@misc{zafari2025assent,
  title        = {ASSENT-CellFree-ISAC: Simulation Framework},
  author       = {Zafari, Mehdi},
  year         = {2025},
  howpublished = {GitHub repository},
  url          = {https://github.com/LS-Wireless/ASSENT-CellFree-ISAC},
  note         = {Version 1.0.0}
}

---

Contact

For questions or collaborations, please contact:
🌐 https://mehdizd97.github.io/

Note: This repository and associated algorithms are part of ongoing research. Final results and additional updates will be provided upon publication acceptance.

---

© 2025 Mehdi Zafari. All rights reserved.