Abstract
This paper introduces a novel approach that enables a number of cognitive radio devices that are observing the availability pattern of a number of primary users (PUs), to cooperate and use Bayesian nonparametric techniques to estimate the distributions of the PUs' activity pattern. To address this problem, a coalitional game is formulated between the cognitive devices and an algorithm for cooperative coalition formation is proposed. It is shown that the proposed coalition formation algorithm allows the cognitive nodes that are experiencing a similar behavior from some PUs to self-organize into disjoint, independent coalitions. Inside each coalition, the cooperative cognitive nodes use Bayesian nonparametric techniques so as to improve the accuracy of the estimated PUs' activity distributions. Simulation results show that the proposed algorithm significantly improves the estimates of the PUs' activity patterns.
Introduction
In cognitive radio (CR) networks, unlicensed Secondary Users (SUs) must opportunistically access the spectrum of licensed Primary Users (PUs) without causing harmful interference. A critical requirement for this is the accurate detection of PU activity. While traditional cooperative spectrum sensing focuses on real-time detection, it often assumes the statistical distribution of PU activity is known beforehand. In practice, these distributions are unknown and can vary significantly across the network due to spatial factors like distance and signal fading.
This research addresses the challenge of how distributed cognitive nodes—specifically Cognitive Pilot Channel (CPC) nodes—can cooperatively learn and estimate these unknown, heterogeneous probability distributions using limited local observations.
Methodology
The proposed framework utilizes a three-phase distributed algorithm combining game theory with advanced statistical methods.
1. Hedonic Coalitional Game Theory
Nodes autonomously self-organize into disjoint groups, or coalitions. This process is modeled as a hedonic game where a node joins a coalition only if it improves its own utility (estimation accuracy) without negatively impacting other members, accounting for a linear communication overhead cost.
2. Priors Validity Check
To ensure data compatibility, nodes perform a two-sample Kolmogorov-Smirnov (KS) goodness-of-fit test. This prevents nodes from incorporating data from peers experiencing different channel conditions, ensuring that combined observations represent the same underlying distribution.
3. Bayesian Nonparametric Estimation
Once validated, estimates are merged using a Dirichlet process. This Bayesian nonparametric approach avoids assuming a fixed distribution shape (like a Gaussian). Instead, it allows nodes to dynamically combine local kernel density estimates with validated "priors" from coalition partners, weighted by the volume of observations.
Key Insights
- Selective Cooperation: Global data sharing is often inefficient due to spatial fading. Localized, disjoint coalitions optimize the balance between accuracy and overhead.
- Distribution-Agnostic Learning: The use of Dirichlet processes and kernel density estimation allows the system to adapt to any true PU behavior without prior assumptions.
- Self-Evaluating Utility: By using Kullback-Leibler (KL) distance to measure estimation stability, nodes can quantify their learning progress without knowing the ground truth.
Results and Discussion
Simulation results in a 2 km × 2 km network demonstrate the effectiveness of the proposed framework:
- Accuracy Gains: The cooperative approach reduced the average KL distance between estimated and actual PU distributions by up to 36.5% compared to non-cooperative methods in a 30-node network.
- Efficiency and Convergence: The distributed game converged rapidly to a Nash-stable partition, typically within 7 to 12 iterations for 30 nodes.
- Scalability: The algorithm naturally favors small groups (average size ~3.5 nodes), keeping communication overhead manageable.
- Mobility: The framework remained robust even with PUs moving at speeds up to 100 km/h, dynamically updating coalitions to reflect the changing environment.
Conclusion
The proposed cooperative Bayesian nonparametric framework provides a robust and efficient method for monitoring PU activity in CR networks. By combining coalitional game theory with distribution-agnostic estimation, cognitive nodes can significantly improve their understanding of the spectral environment. Future work may extend this to estimate activity durations and apply the framework to other decentralized monitoring domains like weather or multimedia data reconstruction.
Summary updated