User Tools

Site Tools


Automatic Stroboscope

Next Steps


Demo Parts/Explanations

Source Separation

  • How would you separate human speech from a fly buzzing? (Hear a fly buzzing the background, and determine its pitch, even if it changes!)
  • Similar to HW 3 for MLSP
    • Frequency spectrum of emitted sounds (regardless of pitch) is based on the harmonics it puts out. 2f, 3f, 4f, etc. If f increases by a little, the frequency does not shift, it shifts by multiples. To get around that, Bhiksha recommended taking the log of the frequency response, so the shift is just a shift, and not a linear shift on top of that.
    • Model it as a draw from a shift urn and a draw from the frequency response urn and see how well it lines up with each frame from the frequency response. (build it off of some examples).
    • Learn the probability distributions…you have to do something to normalize for the loudness, so draw multiple times from the frequency response urn with a certain probability.
    • What were the lecture slides that Bhiksha had on restoring a telephone-line recording back to a good original recording by learning the bases of the instruments? Sorta like the lab you did in Advanced DSP with restoring the singers voice of Ave Maria or something.
  • Once you have a good model for the bee/fly, then you “set” that and learn the other bases on-the-fly so that you can get rid of human speech, etc.
  • Okay…back to finals :)

How to detect "pitch"

How it works

I have included a How It Works section because none of the other open
source tuners that I have looked at give any explanation of the
algorithms used. This tuner uses an overlapped Fast Fourier Transform
together with the phase difference between sequential runs of the FFT
to measure the actual frequency accurately. It appears to be about as
accurate as the clock in the sound card.

HMM analysis using Cepstral domain

Calibration / Pulse Type


Radar / Ultrasonic, using Doppler Effect

  • Hard to do really fast on a microcontroller, but not impossible. :-)
    • Uses some trick to get better speed resolution, maybe look at another website too?
  • I like this paper and how it talks about using the CWT and SVM's for classification of birds by how they flap using radar! Nice…good to copy!

Emitted Sound

  • Similar topic, most doppler systems on the road to detect speed use doppler radar, and check the frequency of the returned signal. Similar to this:
    • They also make laser ones, but those seem to be foiled or something, idk. I think the real systems use lasers.
  • There are cheap laser tachometers, but it looks like their signal processing isn't that good. You need reflective tape?!
    • Anyways, maybe I can make something similar, very similar to the laser mosquito thing from Intellectual Ventures (1,2). Sorta similar to how they do barcode scanning (detailed, but not that much new info)
  • Randy suggested a current follower sort of thing using an op-amp. Some of the optical diodes have the circuit already built in, basically it does boosting of the signal for fast response, since the original signal from the diode is not that high current.
    • Just use a photo diode. Should be good enough… over those other things
    • Also need to be able to fluctuate the laser diode
  • Not sure why I want to do this, but it sounds like it would combine all my skills pretty well and push me forward.
  • Slow it way down and you can hear the fan blades! Matlab script that sorta works…
[fan,fs,bits] = wavread('fan.wav');

fft_length = 4000;
start = 20000;
for i = start:fft_length:length(fan)
    fan_s = fan(i:i+fft_length-1);

    fan_f = abs(fft(fan_s));
    fan_f_s = fan_f(1:200);

    %x = xcorr(fan_s);
  • Way cool demo…flapper peening!
  • Also some more good demos (mostly water droplets, but also inspection)
  • Vibration analysis, way cool!

Flashing Things Really Fast

  • Apparently really easy?!

Making Your Code Right

Flash (Xenon) Tube


projects/autostrobe.txt · Last modified: 2017/09/15 17:58 (external edit)