IARPA-IQ-CSP-Docker

This Docker container (6.1 GB) was built to run the "IARPA-IQ-API" and "IARPA-IQ-CSP-Fusion" repositories for anomaly detection in LTE signals without the need to manually install specific package versions. For more instructions on how to run the "IARPA-IQ-API" Docker container or interface, visit here.

Please send all questions to either gu.je or d.roy {@northeastern.edu}, thanks!

Contents

• Pre-requisites
• Instructions for Users
  • Running the CSP API
  • Acceptable Input File Format
• Appendix

Pre-requisites

Install Docker engine/app for your specific operating system here.
Install the IARPA-IQ-CSP-Fusion container here.

Instructions for Users

To load the Docker container, go to the directory where the container is and use the following command:

sudo cat iarpa-iq-csp-api.tar | sudo docker import - iarpa-iq-csp-api

To verify that the container was loaded successfully, do:

sudo docker images

which should display something like the following, with the iarpa-iq-csp-api image:

REPOSITORY         TAG         IMAGE ID       CREATED         SIZE
iarpa-iq-csp-api   latest      2ef151bf5f13   5 hours ago     6.87GB

Running the CSP API

This image comes with folders and folder paths hard-coded for convenience. To run the image with the desired dataset, run

sudo docker run -v \
<dataset_absolute_local_path>:\
/home/IARPA-CSP-main/test/CSP iarpa-iq-csp-api \
/home/IARPA-CSP-main/./run_ML_code.sh

where /home/IARPA-CSP-main/test/CSP is the <dataset_absolute_image_path>. An example of this command is
sudo docker run -v /home/jgu1/Downloads/CSP:/home/IARPA-CSP-main/test/CSP iarpa-iq-csp-api /home/IARPA-CSP-main/./run_ML_code.sh.

Acceptable Input File Format

The code predicts anomalies from the non-conjugate cycle frequency features. The model is trained on the 4th column of the non-conjugate CSP features; it skips the conjugate features; hence, the input files must have .NC extension. Any file with other extension, e.g., .C, will be skipped. The folder CSP_samples, containing a variety of combined LTE+DSSS and only LTE signal files in .NC and .C format, has been provided for your container-testing convenience.

For the input dimensions of each .NC file:

• The number of rows can be variable (an empty file with zero rows is accepted as well, but the prediction could be wrong).
• The number of columns must be 4, following the sequence of F, A, C, S.

Example Output

The prediction from CSP features in OnlyLTE_frame_120_131072_3.NC is: OnlyLTE
Total time of execution for OnlyLTE_frame_120_131072_3.NC is : 0.002008676528930664 seconds.

The prediction from CSP features in Combined_LTE_DSSS_frame_127_262144_4.NC is: Combined_LTE_DSSS
Total time of execution for Combined_LTE_DSSS_frame_127_262144_4.NC is : 0.002009153366088867 seconds.

Back to Contents

Appendix

The packages and their respective versions in this container are:

Python 3.8.10
pip3 20.0.3
TensorFlow/Keras 2.8
torch 1.11.0
torchvision 0.12.0
torchaudio 0.11.0
Opencv-python (cv2) 4.4.5
Pillow (PIL) 9.0.1
tqdm 4.62.3
glob2 0.7
Setuptools 60.9.3
pandas 1.4.1

Back to Contents