Brain-Inspired Cognitive Architecture for Non-Centralized Spectrum Intelligence
TL;DR
The proposed architecture demonstrates the feasibility of persistent, context-aware, non-centralized spectrum cognition and can host neuro-fuzzy, swarm, evolutionary, reinforcement-learning, or hybrid mechanisms.
Abstract
Objective: To develop and evaluate a brain-inspired cognitive architecture that enables non-centralized spectrum intelligence while preserving primary-user (PU) protection and local radio autonomy. Methods: Each cognitive radio is modeled as an autonomous agent with perception and attention, working and long-term memory, contextual reasoning, adaptive goal management, learning, and decision execution. These functions are organized into a twelve-stage perception-memory-reasoning-action lifecycle with optional bounded peer exchange, a PU-safety gate, convergence monitoring, low-overhead steady-state operation, and event-triggered reactivation. The architecture is algorithm-agnostic and can host neuro-fuzzy, swarm, evolutionary, reinforcement-learning, or hybrid mechanisms. Results: A proof-of-concept 15-agent network achieved 0.7340 accuracy, 0.8670 specificity, 0.1330 false-positive rate, and 0.7068 ROC AUC. Recall was 0.2569 and F1 score was 0.2963, indicating a conservative operating point that reliably rejects unsafe channels but misses many usable opportunities. Conclusion: The proposed architecture demonstrates the feasibility of persistent, context-aware, non-centralized spectrum cognition. The main improvement priorities are confidence-threshold calibration, opportunity detection, larger-scale repeated experiments, and module-level ablation studies.