Synthetic Digitally Modulated Signal Datasets for Automatic Modulation Classification
- Citation Author(s):
- Submitted by:
- John Snoap
- Last updated:
- Sat, 10/29/2022 - 18:40
- Data Format:
- Link to Paper:
Synthetic Digitally Modulated Signal Datasets for Automatic Modulation Classification contain CSPB.ML.2018 and CSPB.ML.2022, two high-quality communication signal datasets with eight modulation types: BPSK, QPSK, 8-PSK, pi/4-DQPSK, MSK, 16-QAM, 64-QAM, and 256-QAM. There are 14,000 signals of each modulation type in each dataset for a total of 112,000 signals per dataset. The two datasets are useful for signal processing testing, neural network (NN) training, initial NN testing, and out-of-distribution NN testing as signal generation parameters differ between the two datasets. Extensive testing has been performed on each signal in both datasets to prove their high caliber. The CSPB.ML.2018 and CSPB.ML.2022 datasets were originally posted to the CSP Blog, and are still available there (see the above link "Datasets for the Machine Learning Challenge"). Further detailed descriptions of both datasets are available on the CSP Blog, along with analysis results and various comments from readers.
Unzipping the file will create a folder titled "CSPB.ML.2018 and CSPB.ML.2022" containing datasets CSPB.ML.2018, CSPB.ML.2022, and example MATLAB scripts showing how they can be used for training and testing a neural network to perform modulation classification. MATLAB does not have to be used to access the data, however, as the data are stored in .tim files with parameter labels in .txt files. See attached instructions for more details.
- CSPB.ML.2018 and CSPB.ML.2022.zip (50.61 GB)