Stability-Aware Configuration Maintenance for Cooperative Wi-Fi Sensing in Controller-Managed WLANs

Technical Summary: Stability-Aware Configuration Maintenance for Cooperative Wi-Fi Sensing

Problem Statement
In controller-managed Wireless Local Area Networks (WLANs), cooperative Wi-Fi sensing relies on assigning Channel State Information (CSI) measurement-rate tiers to active sensing nodes. While upstream controllers can propose feasible replacement configurations to optimize sensing quality, frequent changes to active node sets and rate tiers (churn) can generate avoidable control traffic and service disturbances. Existing literature addresses sensing quality, CSI compression, and handover stability but lacks a specific post-selection maintenance policy. The core problem is determining whether a controller should accept a proposed feasible replacement configuration or retain the currently accepted one, balancing the potential utility gain against the transition costs of reconfiguration (node changes, rate-tier shifts, and control-byte overhead).

Methodology: Stability-Aware Configuration Maintenance (SACM)
The paper proposes SACM, a transition-cost method that acts as a maintenance layer between the upstream controller and the network state. SACM does not generate new configurations or train classifiers; it evaluates proposed replacements against the current state.

• State Representation: A configuration $c_t$ is defined by an active-node mask ($m_t$) and a vector of measurement-rate tiers ($r_t$).
• Feasibility Check: A proposed replacement $c^*_t$ is only considered if it satisfies airtime budgets, protected-service penalties, and node availability constraints. If the current configuration becomes infeasible, SACM forces a switch or enters a fallback state (no active sensing).
• Utility and Cost Formulation:
  • Utility ($J_t$): A normalized score combining sensing quality, airtime usage, and service penalty.
  • Transition Cost ($I$): A cost function penalizing changes in the active node set (Hamming distance), rate-tier changes, and estimated control-byte exchange.
  • Decision Rule: The system accepts a replacement $c^*_t$ only if it is feasible AND the utility gain ($\Delta J_t$) exceeds the sum of the transition cost ($I$) and a decision margin ($\epsilon$). Otherwise, the current configuration is held.
• Baselines: The method is evaluated against two baselines on the same reconstructed sequence:
  1. Immediate Acceptance: Accepts every feasible replacement.
  2. Adapted Adaptive Hysteresis: Uses a scalar hysteresis margin that adapts based on the incumbent's utility, inspired by WLAN handover control.

Experimental Setup
The evaluation utilizes three distinct data sources:

1. Deterministic Corner-Case Simulations: Stress tests covering scenarios like stable high utility, alternating noisy replacements, budget shocks, node failures, and high-cost deadlocks to verify logical boundaries and failure modes (e.g., regret, deadlock runs).
2. Measured Controller Traces: Three laboratory traces (moderate-load, high-load, and a long-horizon stability trace) collected from an ESP32-based testbed. The long-horizon trace involves three sensing nodes over an 89.65-minute span.
3. Same-Horizon Replay: An offline reconstruction of the long-horizon trace's feature sequences. This replay applies the three maintenance policies to the same sequence of proposed configurations to ensure a fair comparison of maintenance accounting without horizon mismatches.

Key Results

• Maintenance Density: In the same-horizon replay of the long-horizon trace, SACM reduced the accepted configuration-update density to 0.000 events/min, compared to 0.216 events/min for immediate acceptance and 0.011 events/min for the adapted adaptive-hysteresis baseline.
• Control Overhead: Correspondingly, SACM reduced the estimated accepted-update bytes to 0 bytes (from 3072 bytes for immediate acceptance and 160 bytes for the hysteresis baseline).
• Stability vs. Regret: In deterministic simulations, SACM successfully suppressed unnecessary switching in noisy and oscillatory scenarios. However, the study explicitly reports "regret" (foregone utility margin) and "deadlock runs" (periods where a better feasible configuration was rejected due to high transition costs) to demonstrate the trade-off. In the replay, the total foregone positive normalized margin was 0.0227.
• Service Impact: The long-horizon trace maintained protected traffic throughput (mean 8.49 Mbps) and latency (95th percentile 24.1 ms) while the maintenance policy was active, though the paper notes these are descriptive metrics rather than causal proofs of improvement.

Significance and Claims
The paper claims a bounded, measurable contribution to the field of controller-managed WLANs. Its primary significance is the introduction of a post-selection maintenance accounting axis.

• Scope: The method is strictly a configuration maintenance policy, not a sensing accuracy improvement or a new classifier.
• Novelty: It isolates the decision of "switch or hold" as a distinct layer, reporting switch density, rate-tier churn, and estimated byte costs alongside utility trade-offs.
• Limitations: The authors explicitly state that the results are based on offline replay and estimated byte counts (not direct control-plane measurements). The claims are limited to the reconstructed tested-configuration sequence and do not assert universal superiority across all hardware, room geometries, or traffic patterns. Future work requires live, matched-policy experiments with direct control-byte instrumentation to validate causal claims regarding traffic reduction.