Very-Large-Scale Integration-Friendly Method for Vital Activity Detection with Frequency-Modulated Continuous Wave...

Area of Science:

• Radar Signal Processing
• Embedded Systems
• Machine Learning

Background:

• Frequency-Modulated Continuous-Wave (FMCW) radars offer potential for non-invasive sensing.
• Intelligent sensor development requires energy-efficient data analysis solutions.
• Existing methods for respiratory activity detection may lack hardware implementation feasibility.

Purpose of the Study:

• To present a simple algorithm for respiratory activity detection using FMCW radar data.
• To propose a computational architecture suitable for custom digital-analog VLSI hardware.
• To develop an energy-efficient data analysis solution for life-sign detection and human trafficking prevention.

Main Methods:

• Algorithm development involving data summarization into motion descriptors.
• Classification of descriptors using a recurrent neural network with gated recurrent units.
• Analog VLSI circuit design, manufacturing, and testing to inform neural model constraints.

Main Results:

• A novel training loss component and weight diversification mechanism were introduced to handle VLSI constraints.
• Experimental evaluation using indoor recordings demonstrated high performance.
• The algorithm achieved over 94% human detection accuracy and comparable F1 scores.

Conclusions:

• The proposed algorithm is simple and hardware implementation-friendly.
• The approach offers a unique, energy-efficient solution for intelligent FMCW sensor development.
• The method shows significant promise for applications requiring life-sign detection under limited visibility.