Other Parts Discussed in Thread: DCA1000EVM
hi, engineers
for detecting object and parse out its range, velocity, and angle information, I collected the raw adc data with awr1843boost and DCA1000EVM，and I transfer the two lane adc BIN data (from DCA1000)to raw complex data(such as 12+52i, 53-27i,....), here the Decimal numbers "12","52","53","-27" are transformed from BIN data( collected from DCA1000EVM). I want to use DL or ML method to train the data and obtain a detection model.
I was previously offered this project by TI's engineer, the guide as below:
Gesture With Machine Learning Users Guide (ti.com), and I noticed The ML model used here does not use raw data and is trained on various extracted features from the radar device - more details can be found in the gesture with Machine Learning users guide. it means the output data(features output data) was transmitted by UART to PC or other user's application.
but, actually, I want to train my raw adc data offline, that means no training on DSP and MCU, just transmit raw data from DCA1000, and I train the raw data in PC.
my question is: how to use these complex data(they are IQ data) in a proper Deep Learning model? Surely,in this offline way, the task of object detection will lose some real-time performance, but that's not my focus. maybe I need more robust target detection performance. all in all, whether I can use PC's Deep Learning to train raw data offline so that I can get much more features information, what advice could you give me? and how to apply appropriate data format into DL model?
If it is possible, give me a feasible demo or method ,so that I can train these raw data, thank you.
Regards,
Wembanyama
