Radio Frequency Database Construction and Modulation Recognition in Wireless Sensor Networks.

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Current modulation recognition methods in wireless sensor networks (WSNs) heavily depend on simulation data, limiting practical application effectiveness.
• The reliance on simulated datasets hinders the ability of WSNs to perform accurately in real-world, complex electromagnetic environments.

Purpose of the Study:

• To address the limitations of simulation-based methods, this research aims to develop and validate techniques for modulation recognition using real-world wireless signal data.
• The study focuses on improving the practical performance of signal identification within WSNs operating in challenging RF conditions.

Main Methods:

• A real radio frequency (RF) database architecture and preprocessing operators were developed using data collected from a USRP 2920 software radio device.
• A multidimensional feature hybrid network (MFHN) was proposed for identifying unknown signals by analyzing diverse signal features.
• An improved MFHN, termed the multifeatured joint migration network (MJMN), was designed specifically for recognizing small-sample targets.

Main Results:

• The MFHN achieved an 82.7% recognition rate for unknown target signals.
• The improved MJMN demonstrated a higher recognition rate of 93.2% for small-sample targets.
• Both proposed methods significantly enhance recognition performance for single-node WSNs in complex electromagnetic environments.

Conclusions:

• The developed real RF database and advanced neural network architectures (MFHN and MJMN) offer a robust solution for modulation recognition in WSNs.
• The findings provide a valuable reference for subsequent research in decision fusion and enhancing WSN capabilities in realistic operational settings.