Group Relative Policy Optimization for Robust Blind Interference Alignment with Fluid Antennas

TL;DR

This work integrates fluid antennas with blind interference alignment to maximize sum-rate in a K-user MISO downlink without CSIT under imperfect CSI. It introduces Group Relative Policy Optimization (GRPO), a critic-free DRL method that leverages group-based relative advantages to efficiently optimize fluid antenna positions, achieving large reductions in model size and FLOPs while improving performance over PPO. The approach decouples the multiuser problem into parallel subproblems and formulates an MDP with states as estimated CSI and actions as antenna positions, guided by a sum-rate reward. Simulation results show GRPO surpasses baselines, including a pre-trained PPO, and dramatically outperforms heuristic position selection strategies, highlighting its robustness to CSI errors and practical viability for fluid-antenna networks.

Abstract

Fluid antenna system (FAS) leverages dynamic reconfigurability to unlock spatial degrees of freedom and reshape wireless channels. This paper proposes, for the first time, a robust fluid antenna-driven blind interference alignment (BIA) framework for a K-user MISO downlink under imperfect channel state information (CSI). We formulate a robust sum-rate maximization problem through optimizing fluid antenna positions. To solve this challenging non-convex problem, we employ group relative policy optimization (GRPO), a novel deep reinforcement learning algorithm that eliminates the critic network. This robust design reduces model size and floating point operations (FLOPs) by nearly half compared to proximal policy optimization (PPO) while significantly enhancing performance through group-based exploration that escapes bad local optima. Simulation results demonstrate that GRPO outperforms PPO by 4.17%, and a 100K-step pre-trained PPO by 30.29%. Due to error distribution learning, GRPO exceeds heuristic MaximumGain and RandomGain by 200.78% and 465.38%, respectively.

Figures (5)

• Figure 1: Illustration of the considered $K$-User MISO downlink BIA system model
• Figure 2: Illustrating the DRL algorithms (PPO-Init, PPO, GRPO) training for our simulations.
• Figure 3: PPO baseline.
• Figure 4: Proposed GRPO solution.
• Figure 5: Performance comparison.