Spectrogram block


I am training a model with static data collected from sensor for idle, leftright, wave, and updown motion by adding processing block as spectrogram block , but i am not getting the predicted output.
can we use spectrogram block or no in this case?

Thanks and regards,

Hi @Mahadevi,

The “spectral analysis” block is probably the best to use for this example, but one or more spectrogram blocks could work (they just won’t be as efficient due to the number of FFTs that need to be performed).

From what I’ve seen, the spectrogram block can only take a single axis input. If you’re using the X, Y, and Z components from an accelerometer for input, which of those are you feeding to the spectrogram? Or are you using 3 spectrogram blocks? Also, check your spectrogram settings to make sure that you’re not seeing some underflow (i.e. everything is 0).

Hi @shawn_edgeimpulse

Thank you for the reply,

I am using accX, accY, accZ components from an accelerometer for input and i am using 3 spectrogram blocks as spectrogram1 for x-axis, spectrogram2 for y-axis, spectrogram3 for z-axis but i am getting processed features (some values) only in case of idle. But zero in case for leftright and updown motions. I have attached the screenshots of spectrogram’s parameters and corresponding processed features (Getting 0’s for updown and leftright). Can you please check and suggest why I am getting 0’s (underflow) for updown and leftright motions ?


You are getting zeros in the Processed Features because you are trying to take a Spectrogram of a signal that has no identifiable frequencies. As Shawn suggested I also recommend using the Spectral Analysis Processing Block in the Impulse design.


Hi @MMarcial

Thank you for the solution.

Thanks and Regards