RISE: Real-time Image Processing for Spectral Energy Detection and Localization

The code base for the DySPAN'26 accepted paper

@inproceedings{tung2026rise,
  title={{RISE}: Real-time Image Processing for Spectral Energy Detection and Localization},
  author={Tung, Chung-Hsuan and Qi, Zhenzhou and Chen, Tingjun},
  booktitle={2026 IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN)},
  year={2026},
  organization={IEEE}
}

Overview

The RISE release contains:

1. A local copy of the RISE-modified Savannah C++ real-time streaming and sensing pipeline.
2. Two Python reference scripts for offline reshaping and visualization: reference/reshape_dumpped_iq.py and reference/visualize.py.
3. RFSynth integration configs under external/rfsynth/config/.

RFSynth is an external dependency and is not redistributed in this repository. See external/rfsynth/README.md for project references, the tested RFSynth commit, and config usage.

Prerequisites

• An Intel CPU that supports AVX-512 for the current Savannah build configuration.
• FlexRAN SDK or ACC100 accelerator support may be required by Savannah's LDPC paths, although RISE does not use LDPC decoding directly.
• Python and Pipenv for the reference scripts.
• MATLAB R2021b or later plus the RFSynth-required toolboxes if generating synthetic RF inputs with RFSynth.

See the included Savannah README for detailed Savannah dependencies.

Build

Run the following command to build the real-time processing system.

cd savannah
mkdir build
cd build
cmake .. -DTIME_EXCLUSIVE=True -DMAT_OP_TYPE=AVX512

Confirm that CMake reports TIME_EXCLUSIVE: True and MAT_OP_TYPE: AVX512. The AVX-512 and LDPC-related dependencies are inherited from Savannah and are not fundamental to the RISE sensing algorithm.

Go to savannah/src/agora/sensing/dosensing_freq.cc and choose the mode to be streamer or processor. Note that two options cannot be enabled at the same time due to the timing constraints.

DumpToFile(tag); // savannah dumps the sensed data to files
// imag_proc_partition(tag); // savannah performs image processing with task partition and dumps bounding boxes in csv files

make

Install the Python reference-script dependencies:

cd reference
pipenv install # install the python packages

Enter the virtual environment before running reference scripts:

cd reference
pipenv shell # enter the virtual environment

Usage

1. (Optional) Generate input signals with RFSynth.

   Use a separately obtained RFSynth checkout and one of the RISE configs in external/rfsynth/config/. See external/rfsynth/README.md.

2. Copy the generated signals to the streaming/processing system.

   Some examples are included in the target folder.

   cp /path/to/generated/<file_id>.{32cf,json} savannah/files/experiment/

3. Update the config file for savannah.

   In savannah/files/config/ci/tddconfig-sim-sensing.json, update the following field:

    /* The id used to generate data from RFSynth */
    "sensing_file_id": "<file_id>",

4. Ensure the output folder exists and is empty: savannah/files/sensing.

   mkdir savannah/files/sensing
   rm -rf savannah/files/sensing/*

5. Open two terminals and run the sender and sensor separately but simultaneously.

   cd savannah
   ./savannah.sh -x -s # run the sensor first

   cd savannah
   ./savannah.sh -u -s # run the sender

6. The output files containing bounding boxes and the region of each subband processor in .csv will reside in savannah/files/sensing. An example file content can be:

   frame_start,frame_end,symbol_start,symbol_end,subcarrier_start,subcarrier_end,num_box
   0,19,0,70,0,511,26
   left,top,right,bottom
   259,39,266,47
   ...

   Note that to generate the bounding boxes, imag_proc_partition(tag) in dosensing_freq.cc needs to be enabled.

Offline Visualization

For offline visualization, enable DumpToFile(tag); in dosensing_freq.cc and run the reference scripts under reference/. Savannah writes raw IQ samples as .bin files under savannah/files/sensing.

1. Reshape the dumped IQ samples.

   cd reference
   pipenv shell
   python reshape_dumpped_iq.py

2. Visualize the sensing results.

   cd reference
   pipenv shell
   python visualize.py

The intermediate image-processing results are plotted under reference/figs/.

Author and Contact

For questions or collaboration inquiries, please contact:

• Chung-Hsuan Tung (chunghsuan.tung@duke.edu)

License

This project is released under the Apache 2.0 License.

Acknowledgment

This project builds upon the Savannah SDR framework developed by the RENEW project.