A Real Valued Neural Network Based Autoregressive Energy Detector for Cognitive Radio Application

Area of Science:

• Electrical Engineering
• Signal Processing
• Machine Learning

Background:

• Cognitive radio (CR) requires efficient energy detection (ED) for spectrum sensing.
• Traditional ED techniques face limitations in performance under varying conditions.
• Autoregressive (AR) system modeling is crucial for signal characterization.

Purpose of the Study:

• To propose and analyze a real-valued neural network (RVNN) based energy detector (ED) for cognitive radio (CR) applications.
• To evaluate the performance of the RVNN-based ED in terms of detection accuracy and false alarm rate.
• To compare the proposed RVNN-based ED against established energy detection methods.

Main Methods:

• A two-layered RVNN model was employed to estimate AR system coefficients.
• Power spectral density (PSD) estimation was performed using the developed RVNN model.
• Receiver operating characteristic (ROC) curves were generated and analyzed to assess detector performance.
• The RVNN-based ED was benchmarked against Simple Periodogram (SP), Welch Periodogram (WP), Multitaper (MT), Yule-Walker (YW), Burg (BG), and Covariance (CV) methods.

Main Results:

• The RVNN-based ED demonstrated high detection performance across various signal-to-noise ratios (SNR), sample numbers, and model orders.
• A low false alarm rate was consistently achieved by the proposed detector.
• The RVNN-based ED outperformed SP, WP, and MT methods in detection.
• Compared to YW, BG, and CV methods, the RVNN-based ED exhibited superior false alarm performance.

Conclusions:

• The proposed RVNN-based ED is effective for cognitive radio spectrum sensing.
• RVNNs offer a promising approach for enhancing energy detection in dynamic radio environments.
• The developed detector provides a robust and accurate solution for CR applications.