3
$\begingroup$

Background

We live in a multi-unit building where each unit's HVAC system/closet is located directly across a party-wall from the neighboring unit's bedroom, and adjacent to a shared external curtain wall. This is a bad recipe that leads to significant low-frequency structure-borne noise transmission through the curtain wall as well as airborne noise transmission across the party wall. The result is noise complaints and sleepless nights. Numerous HVAC servicing efforts and traditional soundproofing techniques have proven ineffective to address the low frequency noise (more background details are in this post). One option that we have not yet tested however is active noise cancellation (ANC), and that is the basis of this question.

Why Active Noise Cancellation May Have Potential

My understanding is that ANC is not something to consider casually. Outside of very controlled situations (e.g. headphones) it seems like the number of variables that must be controlled are overwhelming. Despite this, I think the concept may still have potential as some variables are mitigated a bit in this case:

  • The troubling noise is consistent and predictable. It's all stems from cyclical mechanical noise, and much of it approximates harmonics of the electrical line voltage (e.g. tonal noise from motors).
  • The majority of problematic sound energy is very low frequency (120Hz or lower) as soundproofing efforts already attenuate the mid-to-high ranges effectively. This should also mean that the wavelengths to be cancelled are generally equal-to or greater-than the dimensions of the room where ANC would be applied (in these conditions 120Hz should have a wavelength of about 10' or 3m).
  • It is possible to implement sampling and/or cancellation very near to the source, even from a neighboring unit that is isolated from the actual HVAC equipment itself. This is because the HVAC blower and compressor motors are only approx 2' (0.6m) away from the party wall.

For further clarification on the quality of the noise, see the spectrum graph below. This helps demonstrate the concentration of problematic sound energy around 120Hz.

Spectrum graph

Possible Proof-of-Concept

To do ANC right is seems like complex digital signal processing and predictive phase-control is needed (e.g. microphone arrays and super-low-latency sampling/processing on very expensive hardware). As we do not currently have the time or resources to go this route, at least without some basic confirmation on the concept, I am wondering if it's worth experimenting with a purely analog proof-of-concept.

My thought was to start with a balanced unidirectional mic with low frequency response (like this one), send that signal to a preamp (it seems like a cheap mixer is the best option, like this one), and then run the line-level out to a self-powered sub which includes an integrated low-pass-filter (I have one already). With that I suppose I could experiment with mic and sub placement (e.g. place the sub 1/2 wavelength away) or even flip pins 2 and 3 on the mic to invert the input and place the mic and sub close together with some attenuation material between them.

The big assumptions I am making include:

  • There would be almost no latency in this whole system. I assume that would be the case, as the whole pathway is analog, but I may be mistaken.
  • Physical space and/or attenuation material between the mic and sub would be sufficient to avoid a negative feedback loop that would negate the whole ANC effect.
  • Mid-to-high frequencies could be reduced reasonably via the preamp equalizer plus low pass filter on sub. Without that I suppose the added noise and distortion would be worse than the original noise.
  • Room harmonics will not drastically interfere with the ANC effect, at least not in some target areas (like the head of the bed). This is probably the most risky assumption.

So what I'm wondering is if there is any chance ANC could be feasible in a case like this. I have a background is systems engineering, and am not afraid to experiment, but I'd also like to be told if I'm suffering from the Dunning–Kruger effect by even considering this.

If there is any thread of possibly benefit here, is my proposed proof-of-concept a solid way to begin experimentation?

$\endgroup$

migrated from sound.stackexchange.com Jun 26 '18 at 22:10

This question came from our site for sound engineers, producers, editors, and enthusiasts.

  • $\begingroup$ @Tetsujin - thanks, that seems correct. I've flagged the question and requested it to be moved by a moderator. $\endgroup$ – rjacobs Jun 26 '18 at 17:29
  • 1
    $\begingroup$ Wish you luck :) $\endgroup$ – Tetsujin Jun 26 '18 at 17:55
0
$\begingroup$

I went ahead with the proof of concept. $60 is a minimal investment to make and implementing high-school-level science experiments at home is never a bad idea.

What I discovered is that yes, low frequency ANC is possible with a simple analog setup for predictable read-world noise, but no, it's not practical in any useful way in an open-space environment.

Setup

Keeping the mic close to the speaker with a inverted signal (flipping pins 2 and 3) was not workable. I assume this was because at the low frequencies I was targeting a negative feedback loop cancelled out any reduction effects. There was just no way to attenuate the signal between the mic and speaker with only inches to work with.

I then tried moving the mic and speaker apart a distance of ~1/2 the wavelength of the strongest noise frequency (about 5' for 114Hz), as shown below. The noise source is at the wall near the window:

enter image description here

In this case pins 2 and 3 were not flipped.

Results

After a lot of fiddling with gain/eq values and exact mic/speaker placement, some positive results appeared.

The measurement at the noise source (near the mic) without the system active showed -38dB at 114Hz:

enter image description here

And at the target cancellation area (bed) without the system active I measured -54dB at 114Hz:

enter image description here

Then, after activating the test setup (turning the volume up) the total energy at 114Hz did in fact go down about 10dB. In fact, it was no longer the dominant tone in the spectrum as it was reduced below the unchanged 60Hz component of the noise.

enter image description here

This was great, and there was something incredibly gratifying about seeing and hearing that 114Hz tone go down while turning a volume knob up.

Shortcomings

Despite what I would call a "successful" proof-of-concept, a setup of this nature still lacks strong practical value. Based on additional observations, reasons for this include:

  1. A single target noise frequency could be reduced slightly, but not enough to eliminate the problem. 114Hz tonal noise could still be heard from the direction of the source (uncancelled) and from other directions (reflections with phase shifts).
  2. Targeting other problematic frequencies would require additional setups with different measurements. This could become impractical very quickly.
  3. The setup was extremely finicky. Very minor alterations to mic placement, or gain values on the preamp, could make-or-break the whole effect. There are surely ways to improve on this, but it was clearly not a robust setup.
  4. The area in which the cancellation effect could be measured was very small (maybe 1-2 cubic feet). There are probably ways to improve on this as well, especially given the relatively long wavelengths involved, but it's still a notable limitation.

I am of course beginning to wonder if a DSP solution (adaptive LMS or something like that) would address any of these points, or if these are all unavoidable issues in an open-space environment. That of course is a separate topic.

$\endgroup$
  • $\begingroup$ Have you considered sound masking instead? $\endgroup$ – Ben May 1 at 19:12
  • 1
    $\begingroup$ @Ben, yep, that's exactly what I ended up doing. After lots of experimentation with cancellation techniques (including adaptive LMS) I ended up building a custom brown noise machine instead. I just re-purposed a DSP board to produce some filtered brown noise, targeted at a specific frequency range, and with an amplitude that activity changed based on ambient sound measurements from a microphone. $\endgroup$ – rjacobs May 2 at 14:46
  • $\begingroup$ A friend of mine works in sound-masking and active noise cancellation. He told me basically that active noise cancellation works well when there is one noise source, when the sound waves travel only in one direction (no reflections). $\endgroup$ – Ben May 2 at 15:10
  • $\begingroup$ "active noise cancellation works well when there is one noise source, when the sound waves travel only in one direction (no reflections)" - that's a good summary. The DSP experiments I tried, based on the advice from dsp.stackexchange.com/q/50711/36462, also supported that. Even though the sound source was fairly isolated, there were simply too many reflections and other uncontrollable ambient factors that made the direct cancellation idea impractical. $\endgroup$ – rjacobs May 2 at 15:25

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.