Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Agentic Verifier-in-the-Loop Solver Orchestration for Cell-Free Massive MIMO Downlink Power Control

Zhichao Gao

Abstract

Cell-free massive multiple-input multiple-output (MIMO) systems can provide uniformly strong service through distributed access points, but performance still depends critically on downlink power control. Existing methods are typically selected offline and then applied uniformly across channel and load regimes, even though no single solver is uniformly best. We therefore propose VISO-PC, a verifier-in-the-loop solver-orchestration framework in which an agent routes among trusted solvers rather than generating power coefficients directly. Given a structured instance descriptor, the router selects an initial solver and fallback order, and an independent verifier accepts only candidates that satisfy the constraints and produce a valid verified common rate. For fairness-oriented downlink cell-free power control under per-AP constraints, verification-aware orchestration improves accepted rate over all fixed single-solver baselines on a reproducible prototype benchmark. Moreover, a lightweight memory-based router matches the accepted rate of a strong rule-based router while reducing average runtime and fallback rate. These results show that solver orchestration is a practical agentic layer for cell-free massive MIMO downlink power control.

Agentic Verifier-in-the-Loop Solver Orchestration for Cell-Free Massive MIMO Downlink Power Control

Abstract

Cell-free massive multiple-input multiple-output (MIMO) systems can provide uniformly strong service through distributed access points, but performance still depends critically on downlink power control. Existing methods are typically selected offline and then applied uniformly across channel and load regimes, even though no single solver is uniformly best. We therefore propose VISO-PC, a verifier-in-the-loop solver-orchestration framework in which an agent routes among trusted solvers rather than generating power coefficients directly. Given a structured instance descriptor, the router selects an initial solver and fallback order, and an independent verifier accepts only candidates that satisfy the constraints and produce a valid verified common rate. For fairness-oriented downlink cell-free power control under per-AP constraints, verification-aware orchestration improves accepted rate over all fixed single-solver baselines on a reproducible prototype benchmark. Moreover, a lightweight memory-based router matches the accepted rate of a strong rule-based router while reducing average runtime and fallback rate. These results show that solver orchestration is a practical agentic layer for cell-free massive MIMO downlink power control.
Paper Structure (9 sections, 10 equations, 4 figures, 4 tables, 1 algorithm)

This paper contains 9 sections, 10 equations, 4 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model and Problem Formulation
  3. Proposed Method
  4. Router Instantiation and Complexity
  5. Simulation Results and Analysis
  6. Simulation Setup
  7. Overall Results
  8. Split-wise and Failure Analysis
  9. Conclusion

Figures (4)

  • Figure 1: Overview of VISO-PC. A raw power-control instance is summarized into a structured descriptor, routed to a solver portfolio, and checked by an independent verifier. If the current solver fails verification, the fallback plan is invoked until a verified solution is found or the budget is exhausted.
  • Figure 2: Accepted-rate versus average-runtime trade-off on the prototype benchmark. Router-based orchestration improves accepted rate over all fixed single-solver policies. Agent-Router matches the accepted rate of Rule-Router while using lower average runtime, indicating a favorable efficiency trade-off after early stopping is enabled.
  • Figure 3: Split-wise accepted-rate comparison for the two routing methods. The accepted rates match on all four splits, so the main differences between the routers appear in runtime and regret rather than in acceptance itself. This pattern is consistent with stress being solver-limited and shifted cases exposing retrieval fragility.
  • Figure 4: Per-instance solver profiling in average-runtime--verified-common-rate space. Exact-solver instances occupy the strongest quality region, fast-solver instances cover many easier cases at lower runtime, and the distributed solver remains a low-cost but weak fallback. The spread across solvers motivates orchestration over a solver portfolio.