FMCW Radar Sensors with Improved Range Precision by Reusing the Neural Network.

Area of Science:

• Radar Systems Engineering
• Machine Learning Applications
• Signal Processing

Background:

• Supervised learning enhances radar range precision but faces challenges with increased labels and learning costs when precision exceeds resolution.
• Background noise removal is critical for indoor radar sensing applications.

Purpose of the Study:

• To propose a methodology for increasing radar range precision while mitigating the escalating number of labels in supervised learning.
• To reduce learning costs and maximize computational efficiency in radar systems.

Main Methods:

• Utilizing neural networks trained for specific sections and reusing them for other applications.
• Applying fractional multiple patterns in the frequency domain for pattern analysis across different FFT bin positions.

Main Results:

• Demonstrated that identical fractional multiple patterns in the frequency domain can analyze patterns at various FFT bin positions.
• Confirmed that neural networks trained on the same data can be repurposed effectively.

Conclusions:

• The proposed methodology successfully increases range precision while managing label proliferation in supervised learning.
• Repurposing trained neural networks enables efficient hardware implementation for radar systems, reducing computational overhead.