1
$\begingroup$

Today, a news blurb appeared that claimed that attackers could identify the location in which an audio record was taken by matching the mains hum to known data. Sadly, I fear that this might be feasible. The mains hum has quite a tight tolerance, but any aberrations (that cannot be done away with) could be spotted easily. In fact, power suppliers have to observe the hum and adjust their stations accordingly.

How could you manipulate a record such that this information is destroyed completely, so that it is not reconstructible by any means? Filtering alone might leave enough remains to still perform that identification. After all, all filters have a finite number of poles.

$\endgroup$
2
  • $\begingroup$ Easiest? Filter and add bandpass noise. $\endgroup$ – jojek Jun 30 '14 at 15:43
  • $\begingroup$ In which order? And please give this comment as an attempt to answer. $\endgroup$ – user7358 Jun 30 '14 at 15:46
9
$\begingroup$

I believe it's possible to get location information from 50/60 Hz hum, particularly if the recording is long-duration. See this paper by Catalin Grigoras from 2009.

One way to defeat this would be to add enough sampling rate jitter to your recording so as to swamp the small variations in mains hum. That paper showed information from variations of around 1 part in 1000, with sampling every 10 sec or so. If you let the sampling rate of your signal drift around randomly on that kind of scale (which would have no perceptible effect), it would obscure the variations in the hum.

So, for instance, if you let your sampling rate drift by +/- 0.1% (e.g., 44,050 to 44,150), with a maximum rate of change of, say, 1 Hz in 10 sec, that would impose a random component on top of the underlying power-grid-derived variations, making them hard to separate. This modification would be completely inaudible.

$\endgroup$
3
$\begingroup$
  1. This seems like a highly unlikely scenario. The signal properties of the power grid are tightly controlled so thinks like frequency and/or harmonic content vary very little over 1000s of miles and millions of locations
  2. A lot of recordings these days are done with battery powered devices: portable recorders, cell phones, "bugs" etc. Since they are not connected to the AC line there is virtual no hum in the recordings
  3. If you really care enough, a set of sharp notch filters at the fundamental and a few harmonics (primarily 60Hz and 180 Hz in the US) will do the trick. This may degrade the original audio to a small extent but this may be tolerable in this specific application
$\endgroup$
1
  • $\begingroup$ i really agree with Hilmar on point #1. and the tolerance on the A/D clock can slip enough that you wouldn't even know for sure what time of day and what day the recording was made. if you can be certain that the sample rate was a precise value when the recording was made and you can hear a little hum in it, you might be able to tell when the recording was made in the U.S., but not where. the amount the AC frequency deviates from the precise 60 Hz is the same throughout the grid. otherwise the various generators could not be hooked up to it simultaneously. $\endgroup$ – robert bristow-johnson Jul 1 '14 at 8:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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