Machine Learning-Based Energy-Spectrum Two-Dimensional Cognition in Energy Harvesting CRNs

Yongjian Fan, Wenjun Xu, Chia-Han Lee, Silei Wu, Fan Yang, Ping ZhangIEEE Access 2020 · DOI

Introduction

In modern wireless communications, spectrum scarcity and rising energy consumption present a dual challenge. Cognitive Radio Networks (CRNs) address spectrum limitations by allowing Secondary Users (SUs) to opportunistically access idle channels owned by Primary Users (PUs). However, when these networks incorporate energy harvesting (EH) technologies, traditional one-dimensional sensing—which only monitors spectrum occupancy—becomes insufficient. This paper proposes a two-dimensional (2D) cognition framework that jointly considers spectrum availability and energy levels to optimize network performance.

The Problem: Energy-Spectrum Coupling

Traditional sensing models often ignore the fact that in EH-CRNs, a PU's ability to transmit is fundamentally limited by its harvested energy. If a PU has data but lacks power, the spectrum remains idle. Conversely, an SU must manage its own energy buffer to decide when transmission is most efficient. Ignoring this coupling leads to inaccurate sensing and missed opportunities. Existing methods fail to exploit the spatial and temporal correlations inherent in energy harvesting environments.

Proposed 2D Cognition Framework

The researchers introduce a three-stage pipeline designed to handle the complexities of EH-CRNs:

  1. 2D Sensing: Utilizing a Dynamic Bayesian Network (DBN), the system models the hidden states of the PU (energy and transmission status) based on the SU's observations. This approach leverages the spatial and temporal correlation of energy harvesting; for instance, neighboring sites often experience similar solar or wind conditions.
  2. 2D Prediction: The SU performs multi-slot look-ahead predictions using DBN parameters to forecast future spectrum and energy states, allowing for proactive rather than reactive transmission.
  3. 2D Decision: A Deep Neural Network (DNN) processes these predictions to determine optimal transmission energy. Unlike greedy algorithms, the DNN balances immediate throughput with future energy needs, accounting for the logarithmic nature of Shannon’s formula.
The 2D cognition framework integrates energy and spectrum sensing to improve SU access decisions.

Key Insights and Results

The study identifies a unique U-shaped relationship between false alarm rates and PU loading, suggesting an optimal performance "sweet spot" for 2D cognition. Furthermore, the DNN-based decision maker avoids the pitfalls of greedy allocation by saving energy for high-quality channel conditions.

Simulation results demonstrate significant improvements:

  • Sensing Accuracy: The 2D DBN-based approach significantly reduces miss detection rates compared to 1D sensing by accounting for the PU's energy state.
  • Throughput Gains: The system achieved throughput increases of up to 30% over 1D sensing with greedy allocation and a 132% gain over systems lacking an energy buffer.
  • Robustness: High detection probabilities were maintained even for predictions several slots into the future, proving the effectiveness of the DBN-based forecasting.

Conclusion and Future Work

By treating energy and spectrum as a joint problem, this 2D cognition framework offers a robust solution for EH-CRNs. Future research will explore multi-dimensional models incorporating physical location and multiple frequency bands, as well as strategies for handling missing historical data in real-world deployments where SUs may not always have enough energy to perform sensing.

Summary updated

BibTeX
@article{fan2020machine,
  author = {Yongjian Fan and Wenjun Xu and Chia-Han Lee and Silei Wu and Fan Yang and Ping Zhang},
  title = {{Machine Learning-Based Energy-Spectrum Two-Dimensional Cognition in Energy Harvesting CRNs}},
  journal = {IEEE Access},
  year = {2020},
  doi = {10.1109/access.2020.3019310},
}
Figure