Scenario-Robust Coordinated Beamforming for Cell-Free ISAC under Imperfect CSI: Outage Control and Sensing-Information Preservation

Scenario-Robust Coordinated Beamforming for Cell-Free ISAC under Imperfect CSI This repository contains the full simulator, notebooks, configuration files, and result checkpoints for the paper “Scenario-Robust Coordinated Beamforming for Cell-Free ISAC under Imperfect CSI: Outage Control and Sensing-Information Preservation.” The method, SR-CBSA, is a scenario-robust coordinated block successive convex approximation scheme for joint communication and sensing beamforming in distributed cell-free massive MIMO ISAC networks under pilot-based CSI uncertainty. It enforces direct global received-signal SINR constraints over sampled channel-error scenarios, preserves target-directed sensing power and a transmit-side azimuth-information surrogate, and stabilizes sequential per-access-point updates through trust regions, damping, proximal regularization, and penalized feasibility slack. Training, held-out validation, the physical true channel, and unseen-error test scenarios are kept strictly separate, so the reported gains reflect genuine robustness rather than overfitting to the sampled errors. The code reproduces every result in the paper from seed-controlled experiments: the paired Monte Carlo study across 5, 10, 15, and 20 dB pilot SNR, the same-realization method comparison, the QoS-target feasibility sweep, the responsibility-rule ablation, the scenario-count and synchronization-error sensitivity studies, the full-rank and locally low-rank scalability cases, and the baseline comparisons including the adapted JointOpt framework and a feasible max-min robust SOCP relaxation. All optimization uses CVXPY with CLARABEL, runs on CPU, and writes CSV checkpoints so long runs can be resumed and verified independently. Set the experiment flags in the configuration cell, run the notebook top to bottom, and the module and figures are regenerated automatically. See the README for environment setup, the flag-to-experiment mapping, and the fixed seeds used for each reported table and figure.