Exploring Feature Selection Scenarios for Deep Learning-based Side-Channel Analysis

This repository contains the source code used to generate the results from the paper: "Exploring Feature Scenarios in Deep Learning-based Profiled Side-Channel Analysis".

Setting up

1. Configure dataset paths

Go to experimets/paths.py file and set all dataset path folders for all feature selection scenarios (RPOI, OPOI, NOPOI, NOPOI_DESYNC) and for all datasets (ASCADf, ASCADr, DPAV42, CHESCTF).

2. Configure results folder paths

Go to experimets/paths.py file and set all result path folders for all feature selection scenarios (RPOI, OPOI, NOPOI, NOPOI_DESYNC) and for all datasets (ASCADf, ASCADr, DPAV42, CHESCTF).

3. Download raw datasets

1. Download ASCADf raw traces: https://github.com/ANSSI-FR/ASCAD/tree/master/ATMEGA_AES_v1/ATM_AES_v1_fixed_key
2. Download ASCADr raw traces: https://github.com/ANSSI-FR/ASCAD/tree/master/ATMEGA_AES_v1/ATM_AES_v1_variable_key
3. Download DPAV42 raw traces: https://www.dpacontest.org/v4/42_traces.php
4. Download CHESCTF raw traces: https://zenodo.org/record/3733418#.Yc2iq1ko9Pa

4. Set sys.path.append in scripts (to avoid import issues)

Set the following line:

sys.path.append('/project_root_folder')

in the following files:

ASCADf:

experiments/ASCADf/generate_dataset.py

experiments/ASCADf/grid_search_gta_mlp_cnn.py

experiments/ASCADf/grid_search_mlp_cnn_no_lda.py

experiments/ASCADf/random_search.py

experiments/ASCADf/random_search_da.py

experiments/ASCADf/test_best_models.py

ASCADr:

experiments/ASCADr/generate_dataset.py

experiments/ASCADr/grid_search_gta_mlp_cnn.py

experiments/ASCADr/grid_search_mlp_cnn_no_lda.py

experiments/ASCADr/random_search.py

experiments/ASCADr/random_search_da.py

experiments/ASCADr/test_best_models.py

DPAV42:

experiments/DPAV42/generate_dataset.py

experiments/DPAV42/grid_search_gta_mlp_cnn.py

experiments/DPAV42/grid_search_mlp_cnn_no_lda.py

experiments/DPAV42/random_search.py

experiments/DPAV42/random_search_da.py

experiments/DPAV42/test_best_models.py

CHESCTF:

experiments/CHESCTF/generate_dataset.py

experiments/CHESCTF/random_search.py

experiments/CHESCTF/random_search_da.py

experiments/CHESCTF/test_best_models.py

5. Generate RPOI, OPOI, NOPOI and NOPOI_DESYNC datasets

To prepare datasets, run following python scripts:

ASCADf:

experiments/ASCADf/generate_dataset.py

ASCADr:

experiments/ASCADr/generate_dataset.py

CHESCTF:

experiments/CHESCTF/generate_dataset.py

DPAV42:

experiments/DPAV42/convert_to_h5.py experiments/DPAV42/generate_dataset.py

Executing random search

For random search, you have to run random_search.py (or random_search_da.py, for data augmentation) files for each dataset. There are seven parameters to pass with the python file call:

1. Leakage Model: HW or ID
2. Model type: mlp or cnn
3. Feature selection type: RPOI, OPOI, NOPOI or NOPOI_DESYNC
4. Number of POIS: (e.g. 700 for ASCADf and OPOI)
5. Target key byte: 0 to 15
6. Regularization: True or False
7. Resampling Window: 10, 20, 40 or 80 (for NOPOI and NOPOI_DESYNC. For RPOI and OPOI the value is ignored, but it has to be provided.)

Testing best models

To test best models, you have to run test_best_models.py files for each dataset. There are six parameters to pass with the python file call:

1. Leakage Model: HW or ID
2. Model type: mlp or cnn
3. Feature selection type: RPOI, OPOI, NOPOI or NOPOI_DESYNC
4. Number of POIS: (e.g. 700 for ASCADf and OPOI)
5. Target key byte: 0 to 15
6. Resampling Window: 10, 20, 40 or 80 (for NOPOI and NOPOI_DESYNC. For RPOI and OPOI the value is ignored, but it has to be provided.)

Examples:

Testing best OPOI models for ASCADf

python experiments/ASCADf/test_best_models.py HW mlp OPOI 700 2 20

python experiments/ASCADf/test_best_models.py ID mlp OPOI 700 2 20

python experiments/ASCADf/test_best_models.py HW cnn OPOI 700 2 20

python experiments/ASCADf/test_best_models.py ID cnn OPOI 700 2 20

Testing best NOPOI models for ASCADf

To run best found models for ASCADf, key byte 0, 10000 POIs, NOPOI with resampling window of 20:

python experiments/ASCADf/test_best_models.py HW mlp 10000 NOPOI 2 20

python experiments/ASCADf/test_best_models.py ID mlp 10000 NOPOI 2 20

python experiments/ASCADf/test_best_models.py HW cnn 10000 NOPOI 2 20

python experiments/ASCADf/test_best_models.py ID cnn 10000 NOPOI 2 20

Testing best NOPOI models for ASCADr

To run best found models for ASCADr, key byte 0, 10000 POIs, NOPOI with resampling window of 20 :

python experiments/ASCADr/test_best_models.py HW mlp 25000 NOPOI 2 20

python experiments/ASCADr/test_best_models.py ID mlp 25000 NOPOI 2 20

python experiments/ASCADr/test_best_models.py HW cnn 25000 NOPOI 2 20

python experiments/ASCADr/test_best_models.py ID cnn 25000 NOPOI 2 20