1
$\begingroup$

I have a speaker emitting low-frequency sinusoidal wave and I have physically marked a curved edge on the speaker driver as shown in red in the image below, which I would like to track:

I detect the movement of the speaker driver using a Dynamic vision sensor(DVS) based event-based camera which can capture changes in intensity at every pixel with very high resolution(in the order of microseconds).

Can anyone point me towards how to model and predict the movement of this red edge on the speaker driver, moving back and forth in the image? If I were to plot the column displacement (with respect to the image) of this edge versus time, I am guessing the plot would resemble a sinusoid matching the frequency of the sound wave the speaker is emitting. Based on this, I think I can use a Kalman filter to predict the sinusoidal movement but I am not sure how to set up the equations.

I would appreciate any tips, pointers or resources on this. Thank you!

EDIT:

The speaker is emitting a sinusoidal wave of single, constant frequency and amplitude.

This is the setup of the entire experiment:

$\endgroup$
5
  • $\begingroup$ is the audio playing only a single frequency or a fixed set of frequencies? $\endgroup$ Commented Apr 21, 2020 at 9:09
  • $\begingroup$ @Dspguysam it is playing a single frequency $\endgroup$ Commented Apr 21, 2020 at 9:16
  • $\begingroup$ What does the acquisition setup looks like? Where is the camera with respect to what it is looking at and what do the primary data look like? This will determine the type of predictor that would be better in this case. If you had a "distance" estimate, the Kalman would be straightforward, but now, you could perhaps predict the "wave" of events that corresponds to the movement of the curve you are tracking. $\endgroup$
    – A_A
    Commented Apr 21, 2020 at 9:18
  • $\begingroup$ @Abhijith then a single frequency coming out of speaker is very much a stationary process, then why go for a Kalman filter? use a Weiner filter instead, much more easier to setup and makes sense for your problem $\endgroup$ Commented Apr 21, 2020 at 9:20
  • $\begingroup$ @A_A, I have edited the question and added an image of the setup and what the data looks like. Basically, as the driver moves forward, ON events are generated at the location of the curve I marked on the driver and as the driver moves back, OFF events are generated. Ideally, yes I would like to predict this wave of events corresponding to the curve but I am not sure how to go about it $\endgroup$ Commented Apr 21, 2020 at 9:49

1 Answer 1

1
$\begingroup$

A single frequency coming out of speaker is very much a stationary process, then why go for a Kalman filter? Use a Weiner filter instead, much more easier to setup and makes sense for your problem.

Just go for a standard weiner filter, you could go for recursive least squares as well to model this.

$\endgroup$
1
  • $\begingroup$ Thank you. I will try implementing a weiner filter $\endgroup$ Commented Apr 21, 2020 at 9:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.