Linear Discriminant Analysis-Based Motion Classification Using Distributed Micro-Doppler Radars with Limited Backhaul

Area of Science:

• Radar Systems Engineering
• Signal Processing
• Machine Learning for Sensor Networks

Background:

• Distributed radar systems face challenges with limited backhaul capacity for transmitting high-dimensional data.
• Micro-Doppler (MD) signatures contain rich information for motion classification but are data-intensive.
• Effective data fusion in networked radar requires efficient dimensionality reduction and communication strategies.

Purpose of the Study:

• To develop a cooperative motion classification algorithm for distributed MD radars with limited backhaul.
• To reduce the data transmission burden from individual radars to a central fusion center.
• To enhance motion classification accuracy in networked radar systems.

Main Methods:

• Utilizing Linear Discriminant Analysis (LDA) for dimensionality reduction of MD signatures at each radar.
• Implementing a softmax processing method with pyramid vector quantization to compress and transmit feature information.
• Employing adaptive channel resource allocation based on classification separability to optimize backhaul usage.

Main Results:

• Demonstrated successful dimensionality reduction of multi-aspect angle MD signatures.
• Achieved efficient data compression and transmission through quantized softmax outputs.
• Validated the cooperative LDA-based algorithm's effectiveness using both simulated and experimental USRP-based MD radar data.

Conclusions:

• The proposed cooperative LDA-based algorithm effectively classifies motion in distributed MD radar networks with limited backhaul.
• The combination of LDA dimensionality reduction, softmax processing, and adaptive resource allocation significantly reduces communication overhead.
• The method shows promise for enhancing the performance and scalability of networked radar systems for various applications.