Working with Data Sets

Available Data Sets

The available data sets from Deepwave are shown in the table below.

Data Set	Modulation Types	Source	Training Files	Testing Files	SNR range
Radar	LFM, Nonlinear LFM, Pulsed Doppler, Frank Codes	Synthetic	78,000	7,800	-5 to +20 dB

Using Customer Provided Data

Ultimately AirPack is likely to be used with application specific training data. To use other data sources without code modifications, the following requirements must be met.

1. Data files should be of time interleaved I/Q data of dtype = int16 or float32. Note that I/Q interleaved numpy.float32 is equivalent to numpy.complex64.

2. The numeric label is the name of the bottom-most folder in the tree of data, e.g., for file data/train/x/y/file.bin, the label is y.

Folder Structure for Data Sets

The data should be placed in a folder with a tree structure as defined below:

data/
├── train/
│   ├── 00/ # Label 0
│   │   ├── traindatafile1.bin
│   │   ├── traindatafile2.bin
│   │
│   ├── 01/ # Label 1
│   │   ├── traindatafile3.bin
│   │   ├── traindatafile4.bin
│
├── test/
│   ├── 00/ # Label 0
│   │   ├── testdatafile1.bin
│   │   ├── testdatafile2.bin
│   │
│   ├── 01/ # Label 1
│   │   ├── testdatafile3.bin 
│   │   ├── testdatafile4.bin