How to process raw signal data from FMCW radar RFbeam v-md3 befor using it in classification by depp learning

I am currently working on a project involving FMCW radar, specifically the RFbeam v-md3 model. My goal is to process the raw signal data from this radar before using it in a deep learning classification person motion task.

I am looking for advice on the optimal methods and techniques to preprocess raw signal data obtained from the RFbeam v-md3 FMCW radar. What are the essential steps to enhance the quality of the data for subsequent deep learning classification and if any one have already code for this.

I hope someone can give me some instructions and pointers such as how to pre-process (remove noise, do fft/stft) the signal. Sorry for the bad english, Thank You.

I attached the .bin files from the original file in google drive link Output data in google drive

and thank you in advance.

Answer 1

[Arka]

Hi @Akkad Alahmad,

Some preprocessing steps to refine the raw signal data are mentioned below:

1. Noise filtering - You can use lowpass, bandpass, or any other filtering option
2. Clutter removal - You can try Moving Target Indication filters or background subtraction to remove stationary objects
3. Range processing - Convert time-domain signal to frequency-domain and analyze range using Fast Fourier Transform, Hamming, or Hann
4. Doppler processing - Use FFT along the pulse repetition interval to get Doppler frequency shift
5. Normalization - Scale the features to have mean and standard deviation of zero and one respectively
6. Data Augmentation - Generate more training data by introducing minor variations
7. Dimensionality reduction - Principal Component Analysis
8. Training/Validation split
9. etc.

Hope this helps!

[Akkad Alahmad]

Thank you so much for your response! Your guidance, for sure was helpful, I appreciate it .

Thanks again

[Arka]

I am glad to know that it helped you!