This is an implementation of the SoundWave paper on the web. It enables you to detect motion using only the microphone and speakers!
Just run it like this
doppler.init(function(bandwidth) {
console.log(bandwidth.left - bandwidth.right);
});See more in example.html. (Note that doppler uses navigator.getUserMedia, which can't be run on the local filesystem, so you'll have to start a server to run this.) Read more about the theory of how this works on the github-pages site.
What to contribute? Here's what is most needed:
Add support for using multiple sinusoids, and combining the data (could be as simple as taking the average), to improve robustness.
Up for a challenge? It'd be great to implement the various tricks described on HN on improving the robustness/accuracy for this (using tricks from radar tech).
In the SoundWave paper they talk about a phenomenon that occurs when you move your hand too quickly. (See Figure 2d.) A new bulge is formed. I didn't implement the method they described for reducing this, but it should be pretty easy. What I'm doing at the moment to calculate the bandwidth (see getBandwidth), is just iteratively step to the right and left until I've hit a frequency with amplitude 0.001 (see maxVolumeRatio) of the doppler tone (see the global variable freq). What should be done instead (as suggested by the paper) is
perform a second scan, looking beyond the stopping point of the first scan. If a second peak with at least 30% of the primary tone’s energy is found, the first scan is repeated to find amplitude drops calculated from the second peak.
This improvement can/should all occur in the getBandwidth function.
Unfortunately this doesn't work on Firefox since it doesn't seem to support the echoCancellation: false parameter to navigator.getUserMedia. This means there's no way to turn off it filtering out the sounds which are coming from the computer itself (which is precisely what we want to measure).
The awesome Jasper Lu implemented a version of this to android. Go check it out!