Unified Edge AI Processing System with PCIe104Jet and PCIe104-RFSOC

Introduction

In the rapidly evolving landscape of Edge AI, the demand for high-performance, low-power processing solutions is greater than ever. Applications in aerospace, medical, and industrial automation require a system that can handle diverse and complex data types, from radio frequency (RF) signals to high-resolution imagery, all in a compact and rugged form factor. This application note details how the combination of the PCIe104Jet and PCIe104-RFSOC from Sundance DSP creates a powerful and versatile processing system. By leveraging the NVIDIA Jetson as the main host processor, this architecture delivers a comprehensive solution where the RFSoC acts as a high-performance, intelligent peripheral for advanced RF signal processing.

The Components of the System

This integrated system is built upon two core PCIe104 form-factor boards:

  • PCIe104Jet: This carrier board hosts an NVIDIA Jetson module (e.g., Orin NX, Orin Nano). This board serves as the main processing hub for the entire system.
    • CPU (ARM in NVIDIA Jetson): The ARM cores within the Jetson module function as the main application processor. They run the primary Linux operating system, manage the high-level application logic, and act as the root complex for the PCIe104 stack. Their role is to orchestrate all system activities, from managing data flow to coordinating tasks between the GPU and the RFSoC peripheral.
    • GPU (NVIDIA Jetson): The NVIDIA GPU is specialized for massive parallel computing. This is the engine for a wide range of AI and machine learning tasks, particularly for image and video processing. It excels at tasks like object detection, image classification, and segmentation, making it perfect for interpreting data from the MIPI camera interface.
  • PCIe104-RFSOC: This board is based on the AMD (formerly Xilinx) UltraScale+ RFSoC, which acts as a powerful, high-throughput peripheral to the Jetson. The RFSoC handles dedicated RF and DSP tasks, offloading these compute-intensive functions from the main CPU.
    • CPU (ARM in UltraScale+ RFSoC): The ARM cores within the RFSoC’s processing system are tasked with managing the low-level, real-time operations of the RF front-end. They handle the configuration of the ADCs/DACs, manage data transfer to the FPGA, and communicate with the host Jetson over the PCIe bus. Their role is dedicated and specialized, supporting the primary host processor.
    • FPGA (Programmable Logic in UltraScale+ RFSoC): The RFSoC’s programmable logic is a massively parallel, reconfigurable fabric. This is where high-speed, real-time RF signal processing takes place. The FPGA is ideal for tasks like digital signal processing (DSP), waveform generation, and implementing custom, high-throughput data pipelines that are too demanding for a CPU or GPU.

The Synergy: Leveraging an Advanced RF Peripheral

The true power of this system lies in the seamless integration of these two boards, with the PCIe104Jet acting as the intelligent host and the RFSoC as the advanced peripheral. The Jetson’s powerful ARM and GPU can now focus on complex AI tasks and system management, while the RFSoC handles the specialized, high-bandwidth RF signal processing. This architecture allows for true sensor fusion, where the host Jetson can simultaneously receive and analyze both high-resolution visual data from its MIPI interface and sophisticated RF data streamed over the PCIe bus from the RFSOC.

For example, a drone application could use the RFSoC peripheral to process radar or electronic warfare signals to detect and track targets in real-time. This processed data is then sent to the Jetson’s main CPU, which fuses it with high-resolution video from a camera (via the MIPI interface). The Jetson’s GPU can then run powerful object detection algorithms on the video stream, providing visual confirmation of the targets identified by the RFSoC. This cohesive system offers unparalleled situational awareness for critical applications.

Suggested Applications

The unique capabilities of this host-peripheral system open up a wide range of possibilities across various industries:

  • Aerospace & Defense:
    • UAVs (Unmanned Aerial Vehicles): Implementing sensor fusion for autonomous navigation, where radar data (processed by the RFSoC) is combined with visual data (processed by the Jetson) for enhanced situational awareness and target tracking.
    • Electronic Warfare: Developing systems for real-time signal intelligence, jamming, and spectrum monitoring, with the Jetson providing a powerful interface for displaying and analyzing the results, driven by its AI capabilities.
    • Satellite Communications: Creating high-throughput ground terminals that can process RF signals and use the Jetson’s AI to optimize data transmission and network management.
  • Medical:
    • Portable Ultrasound: Building compact, high-performance ultrasound devices. The RFSoC handles the complex beamforming and signal processing, while the Jetson’s GPU accelerates the image reconstruction and analysis, potentially using AI to detect anomalies in real-time.
    • Surgical Robotics: Developing robotic systems that use both RF (e.g., for haptic feedback or sensor control) and vision (e.g., for visual guidance) to perform intricate procedures.
    • Medical Imaging: Creating advanced diagnostic equipment that processes signals from various sensors (e.g., MRI, CT) and uses the Jetson’s AI for automated image interpretation and diagnosis.
  • Related Applications:
    • Industrial Automation: Deploying systems for quality control in manufacturing. The RFSOC could handle signals from sensors like LiDAR or radar for distance and material analysis, while the Jetson’s camera interface processes visual data for defect detection.
    • 5G/6G Base Stations: Creating compact and intelligent base station solutions where the RFSoC processes the radio signals and the Jetson manages the network traffic and implements AI-driven optimization for efficiency and performance.

By integrating the PCIe104Jet as a powerful host with the PCIe104-RFSOC as a specialized RF peripheral, developers can create a robust, modular, and highly adaptable platform that meets the stringent demands of Edge AI applications, pushing the boundaries of what is possible in a compact, embedded system.