Design and Real-Time Implementation of an Intelligent Wireless Receiver System

Key Points

• The aim is to design and implement a real-time intelligent wireless receiver system using machine learning techniques in its physical layer.
• Utilized a transformer-based neural network model for channel and phase noise estimation.
• Connected components via a high-speed interface among FPGA, CPU, and GPU.
• Optimized the neural network receiver with PyTorch and TensorRT for efficient GPU processing.
• Achieved parallel processing capabilities for real-time slot handling.
• Achieved over 3dB improvement in signal-to-noise ratio compared to existing hardware receivers.
• Successfully processed up to 7 slots in real-time using a parallel structure with multiple neural network instances.
• Demonstrated practical applicability in indoor testing environments.

Abstract