This projects makes an attempt to get insights from nearly-raw radar data, just lightly processed with FFT. Early experiments show that it's trivial to achieve 90+% accuracy for simple classifiers. The project goal is to establish boundaries of what's possible with this approach.

1. Get TI mmWave SDK and install it locally.

2. Check out this repository into ${MMWAVE_SDK_INSTALL_PATH}/packages/ti/demo/xwr14xx/DeepRadar.

3. Copy ${MMWAVE_SDK_INSTALL_PATH}/packages/ti/demo/xwr14xx/mmw/mmw.cfg into ${MMWAVE_SDK_INSTALL_PATH}/packages/ti/demo/xwr14xx/DeepRadar/mmw.cfg. This file is not released under open source license by TI and can't be included into this repository.

4. Set environment variables via ${MMWAVE_SDK_INSTALL_PATH}/packages/scripts/unix/setenv.sh as described in TI mmWave SDK User Guide.

5. Build the mss binary:

DeepRadar$ make all
Configuring RTSC packages...
<...>
******************************************************************************
Built the Millimeter Wave OUT and BIN Formats
******************************************************************************

6. Flash rss and mss binaries into your TI IWR1443BOOST board. See "How to flash an image onto xWR14xx/xWR16xx EVM" in the TI mmWave SDK User Guide.

Note: flashing might not properly work on Linux. Use a Windows machine for this operation as a workaround.