Skip to content

Radar Signal Processing over Network#

Authors: Dylan Eustice (NVIDIA)
Supported platforms: x86_64, aarch64
Language: C++
Last modified: June 10, 2026
Latest version: 1.2
Minimum Holoscan SDK version: 0.6.0
Tested Holoscan SDK versions: 0.6.0
Contribution metric: Level 2 - Trusted

[!NOTE] DAQIRI migration (June 26, 2026): HoloHub networking examples, which previously utilized the Basic or Advanced Networking Operators, have been migrated to the standalone networking library, DAQIRI.

The Network Radar application demonstrates signal processing on data streamed via packets over a network. It uses DAQIRI to send or receive data, combined with the signal processing operators implemented in the Simple Radar Pipeline application.

Using DAQIRI's GPUDirect-capable raw packet path, this pipeline has been tested up to 100 Gbps (Tx/Rx) using a ConnectX-7 NIC and A30 GPU.

The motivation for building this application is to demonstrate how data arrays can be assembled from packet data in real-time for low-latency, high-throughput sensor processing applications. The main components of this work are defining a message format and writing code connecting DAQIRI packet bursts to the signal processing operators.

This application uses DAQIRI raw packet transport.

Prerequisites#

See DAQIRI documentation for requirements and system tuning needed to enable high-throughput GPUDirect capabilities.

Environment#

Note: Dockerfile should be cross-compatible, but has only been tested on x86. Needs to be edited if different versions / architectures are required.

Build#

Please refer to the top level Holohub README.md file for information on how to build this application: ./holohub build network_radar_pipeline.

Run#

Note: must properly configure YAML files before running. To run with DAQIRI raw transport:

  • On Tx machine: ./build/applications/network_radar_pipeline/cpp/network_radar_pipeline source.yaml
  • On Rx machine: ./build/applications/network_radar_pipeline/cpp/network_radar_pipeline process.yaml

DAQIRI Connectors#

Implementation is in daqiri_connectors. RX is configured to run with GPUDirect enabled, in header-data split (HDS) mode. TX supports both GPUDirect/HDS or CPU-only packet payloads.

Testing RX on generic packet data#

The application supports testing the radar processing component in a "spoof packets" mode. This functionality allows for easier benchmarking of the application by ingesting generic packet data and writing in header fields such that the full radar pipeline will still be exercised. When SPOOF_PACKET_DATA (daqiri_connectors/adv_networking_rx.h) is set to true, the index of the packet will be used to set fields appropriately.

Message format#

The message format is defined by RFPacket. It is a byte array, represented by RFPacket::payload, where the first 16 bytes are reserved for metadata and the rest are used for representing complex I/Q samples. The metadata is:

  • Sample index: The starting index for a single pulse/channel of the transmitted samples (2 bytes)
  • Waveform ID: Index of the transmitted waveform (2 bytes)
  • Channel index: Index of the channel (2 bytes)
  • Pulse index: Index of the pulse (2 bytes)
  • Number samples: Number of I/Q samples transmitted (2 bytes)
  • End of array: Boolean - true if this is the last message for the waveform (2 bytes)