Active noise-cancelling

For active noise-cancellation to work, what distance does the speaker producing the so-called anti-noise have to be away? in headphones they appear to have everything all encapsulated but I thought you needed to be half a wavelength away or something? So for very low frequencies you would need to be a fair distance. Suppose you wanted to cancel 40Hz for example.

On Sunday, September 8, 2013 2:02:30 AM UTC-4, gyans...@gmail.com wrote:
> For active noise-cancellation to work, what distance does the speaker producing the so-called anti-noise have to be away? in headphones they appear to have everything all encapsulated but I thought you needed to be half a wavelength away or something? So for very low frequencies you would need to be a fair distance. Suppose you wanted to cancel 40Hz for example.



You don't have to be a 1/2 wavelength away.  

The most important guidelines with respect to secondary source placement are:

- Unless you are using a headset or headrest, don't point the speaker directly at the person if you are doing high level tonal cancellation.  Many people can localize high level tonal sounds when the speaker is trained on them. 

- Don't point the speaker in directions that excessively lengthen the transfer path delay.  It will cause tracking delays and affect stability.  An odd direction may also cause you to need to go thermonuclear to create your equal-but-opposite sound level at the head location.  That increases the likelihood of distortion, observer localization [see above], and unwanted hotspots. 

- Don't have the speaker close enough that the observer can sense the air movement.  People dislike their hair moving more than the don't like loud sounds.

- Don't have the speaker so far away that you need to go thermonuclear... [see above]

- Don't use ported speakers in places without a trash can.

That last one isn't so much a guideline as a rule.

Basically, a cancellation speaker is always too far away *and* too close - for some reason or another.

That reminds me of an old joke.  What is the perfect distance for an outhouse?  Answer: Too far in the winter and too close in the summer.  An ANC speaker is the same. 

Keep in mind, also, that the farther that you are away from your error sensor - assuming that you are still able to close your loop with some reasonable level of coherence - the less able you are to control non-tonal noise.  A Nyquist plot will tell you why headsets work on whooshy things and non-contact ANC systems work on hummy things.  It's the law (of physics).

That's about it.

Dan

On Wednesday, October 30, 2013 7:33:58 AM UTC+13, djma...@gmail.com wrote:
> On Sunday, September 8, 2013 2:02:30 AM UTC-4, gyans...@gmail.com wrote:
> 
> > For active noise-cancellation to work, what distance does the speaker producing the so-called anti-noise have to be away? in headphones they appear to have everything all encapsulated but I thought you needed to be half a wavelength away or something? So for very low frequencies you would need to be a fair distance. Suppose you wanted to cancel 40Hz for example.
> 
> 
> 
> 
> 
> 
> 
> You don't have to be a 1/2 wavelength away.  
> 
> 
> 
> The most important guidelines with respect to secondary source placement are:
> 
> 
> 
> - Unless you are using a headset or headrest, don't point the speaker directly at the person if you are doing high level tonal cancellation.  Many people can localize high level tonal sounds when the speaker is trained on them. 
> 
> 
> 
> - Don't point the speaker in directions that excessively lengthen the transfer path delay.  It will cause tracking delays and affect stability.  An odd direction may also cause you to need to go thermonuclear to create your equal-but-opposite sound level at the head location.  That increases the likelihood of distortion, observer localization [see above], and unwanted hotspots. 
> 
> 
> 
> - Don't have the speaker close enough that the observer can sense the air movement.  People dislike their hair moving more than the don't like loud sounds.
> 
> 
> 
> - Don't have the speaker so far away that you need to go thermonuclear... [see above]
> 
> 
> 
> - Don't use ported speakers in places without a trash can.
> 
> 
> 
> That last one isn't so much a guideline as a rule.
> 
> 
> 
> Basically, a cancellation speaker is always too far away *and* too close - for some reason or another.
> 
> 
> 
> That reminds me of an old joke.  What is the perfect distance for an outhouse?  Answer: Too far in the winter and too close in the summer.  An ANC speaker is the same. 
> 
> 
> 
> Keep in mind, also, that the farther that you are away from your error sensor - assuming that you are still able to close your loop with some reasonable level of coherence - the less able you are to control non-tonal noise.  A Nyquist plot will tell you why headsets work on whooshy things and non-contact ANC systems work on hummy things.  It's the law (of physics).
> 
> 
> 
> That's about it.
> 
> 
> 
> Dan

A Nyquist plot won't tell me anything useful, that's why I use Bode-Plots. Thanks for the answer but you seem to be saying that you can cancel anything at all and I cannot see this to be the case. Surely you are limited to low frequencies and this is why my original question was asked ie low frequencies low wavelength and the problem is easier to solve. are you saying that you could cancel (say) a person speaking in the same room?

On Sunday, November 3, 2013 11:52:44 AM UTC-5, gyans...@gmail.com wrote:
> On Wednesday, October 30, 2013 7:33:58 AM UTC+13, djma...@gmail.com wrote:
> 
> [snip]
> 
> 
> A Nyquist plot won't tell me anything useful, that's why I use Bode-Plots.
> Thanks for the answer but you seem to be saying that you can cancel anything
> at all and I cannot see this to be the case. Surely you are limited to low
> frequencies and this is why my original question was asked ie low frequencies
> low wavelength and the problem is easier to solve. are you saying that you
> could cancel (say) a person speaking in the same room?


Where did I say that you can cancel anything at all?!  That's idiotic.

This is what I wrote:
"...the farther that you are away from your error sensor - assuming that you are still able to close your loop with some reasonable level of coherence - the less able you are to control non-tonal noise."

We were talking about control speaker placement.  I am saying that the farther you are away from your error sensor with your control speaker, the less able you are to control non-tonal noises due because of the delay that you are introducing into the control path.

The effects of loop delays in a feedback active noise control systems are brutal.  It does not take much distance at all between the control speaker and error sensor before you are restricted to tonal control only

Regarding "limitations..."

With respect to standard destructive interference noise cancellation using squared pressure control, your ability to do low frequency noise tonal reduction is typically limited by the capability of your speaker and amplifier to produce the equal but opposite destructive interference signal.  Your ability to do high frequency tonal reduction is typically limited by the size of your destructive interference "zone" (see Tokhi, et al.). 

The same limitations exist for non-tonal noise but they are trumped huge by the issue of control loop delay.  

...and just because a Nyquist plot isn't telling you anything useful doesn't mean that it doesn't have anything useful to tell you.

Dang it!

Twice I typed Nyquist when I intended to type Nichols.  I needed to think faster and type slower.

So...  You have my sincere apologies for the typo.  It damaged the discussion.  I am sure that you will agree that a Nichols plot *will* tell you what you need to know.  A Bode plot is fine, also.  Sometimes it is a bit harder to visualize things, though. 

I reread the rest.  I stand by it.  Please let me know if anything else that I said is unclear.  ...and, once again, my apologies.

Dan

You have to be careful about which type of ANC you are talking about. If you have one or more noise sources that you can pick up with a microphone or other sensor then you can use an adaptive feed-forward system where the error-sense mic, which is close to the listener, is only used to adapt the filters but is not actually used as a feedback signal. In that case the only requirement is that the cancellation speaker must be closer to the listener than the noise source (assuming 0- latency processing). In the case where you want to create a zone of silence for all noise sources, and you don't have sensors to pick up the noise, then you must use the error mic signal to close the loop in a classical feedback control sense, and then the cancellation speaker must be located very close to the error mic if you want to get any reasonable cancellation bandwidth (same principle as used in noise- canceling headphones).  These are two distinct and very different systems. 

Bob

On Sunday, November 3, 2013 5:27:29 PM UTC-5, radam...@gmail.com wrote:
> You have to be careful about which type of ANC you are talking about. If you have one or more noise sources that you can pick up with a microphone or other sensor then you can use an adaptive feed-forward system where the error-sense mic, which is close to the listener, is only used to adapt the filters but is not actually used as a feedback signal. In that case the only requirement is that the cancellation speaker must be closer to the listener than the noise source (assuming 0- latency processing). In the case where you want to create a zone of silence for all noise sources, and you don't have sensors to pick up the noise, then you must use the error mic signal to close the loop in a classical feedback control sense, and then the cancellation speaker must be located very close to the error mic if you want to get any reasonable cancellation bandwidth (same principle as used in noise- canceling headphones).  These are two distinct and very different systems. 
> 
> 
> 
> Bob


I agree about the type of control being an important factor, Bob.  My observations were skewed toward feedback ANC in closed cabins using squared pressure error signals.  

I do not agree that the *only* requirement in a feedforward arrangement is that the cancellation speaker be closer to the noise source.  I believe that my original points still apply (with some qualification):

Point #1 was don't point the speaker directly at the person if you are doing high level tonal cancellation because many people can localize high level tonal sounds when the speaker is trained on them. In a cabin quieting arrangement, this is true whether the configuration is feedback or feedforward.  For specialized free field applications like the acoustic shadowing of transformer noise (Wright, et al), this does not apply.

Point #2 was don't point the speaker in directions that excessively lengthen the transfer path delay because it will cause tracking delays and affect stability.  This is true for feedback.  Delays in the feedforward secondary path loop will not affect stability, though.    

Point #3 was don't have the speaker close enough that the observer can sense the air movement because people dislike their hair moving more than the don't like loud sounds. I think that this is true for everything.  I have never had an application where anyone was fine with ANC blowing their hair.

Point #4 was don't have the speaker so far away that you need to go thermonuclear. I am basically saying that you don't want to be farther away than you need to be because no good comes from it.  You need a bigger speaker and amp and risk localization and hotspots in the interference pattern.  That is the same for feedforward or feedback.

Point #5 was don't use ported speakers in places without a trash can.  That was just for humor's sake.  It is true, though.  A company that I used to work for once had someone put their styrofoam coffee cup in upward facing port and then complain that their ANC system "wasn't working so good anymore."  

Dan

Dan

Thanks , I find your points to be very informative and obviously you have a lot of real- world experience. I guess your point about localizing tonal components probably come from the fact that below 1lhz or so , the ear localizes tones by inter- aural phase shifts, and the cancellation signal will not have the same effect at both ears so you may get some strange localization artifacts. In general it's probably better to shoot for moderate cancellation over a wider region rather than deep cancellation at a single location, otherwise small movements of the listeners head will cause large changes in the amount  (and phase) of cancellation which would probably be annoying; do you agree? 

Bob

I agree with you - and for more reasons than just localization.

About the localization, I do think that you are correct.  I am not approaching this technical area from the physio/psychoacoustic side, so I am not the one to ask about ear theory.  ...but ears are darned amazing devices.  

You can be getting 6dBA of low frequency cancellation and the listener will tell you of the sampling noise tone that they hear because you went just a little too low of order on the reconstruction filters in an effort to keep the group delay down.  It's fascinating.  Aggravating, at times.  ...but fascinating.

About the deep cancellation... yes to this.  Going huge on reduction at a single point is problematic.  The shape of the quiet zone wouldn't change necessarily, but the gradient would be much more steep leading to the condition you describe.

A second problem is quiet zone shift.  Let's say that someone is doing stock-and-standard low frequency tonal squared pressure adaptive feedback ANC with a single mic and single speaker in a small mobile equipment cabin.  The mic is - in essence - looking only in one direction according to its field characteristics.  It is also only measuring the stored acoustic energy in the cabin - not the kinetic energy.  This means that the quiet zone shape is going to be highly sensitive to the physical characteristics in the cabin.  So, when little tractor driver #1 gets out of the cab and big tractor driver #2 gets in, your quiet zone shifts.  ...and, chances are, it is not going to shift to a better location.

A third problem is regulation.  ...and this one is complete craziness.  The federal regulations (US, EC, etc.) for quieting in a locomotive - let's say - are very specific as to the measurement location.  ...but it is NOT NECESSARILY AT THE OPERATOR'S HEADS.  

It is not uncommon to have the regulated measurement location to be specified to be "in the center of the two operator head locations at 1.5m above the floor" or some nonsense like that. 

So you and I go in with our top notch ANC system and bag 4dBA at the ear locations of both the fireman and the engineer.  Great right?  Bzzzt.  Nope.  We just made it 3dBA louder at the regulated location in the process.  ...so to meet federal regulations, you better either detune that controller to get less reduction at the ear locations or add a third real or virtual error microphone to also control at a location in the locomotive cabin that nobody will ever likely have their head.  Brilliant.

Technically, ANC is a lot of fun.  It tends to have more than a small amount of psychology, physiology, and politics, also.  That is less fun.

Dan

On Monday, November 4, 2013 9:17:57 AM UTC+13, djma...@gmail.com wrote:
> On Sunday, November 3, 2013 11:52:44 AM UTC-5, gyans...@gmail.com wrote:
> 
> > On Wednesday, October 30, 2013 7:33:58 AM UTC+13, djma...@gmail.com wrote:
> 
> > 
> 
> > [snip]
> 
> > 
> 
> > 
> 
> > A Nyquist plot won't tell me anything useful, that's why I use Bode-Plots.
> 
> > Thanks for the answer but you seem to be saying that you can cancel anything
> 
> > at all and I cannot see this to be the case. Surely you are limited to low
> 
> > frequencies and this is why my original question was asked ie low frequencies
> 
> > low wavelength and the problem is easier to solve. are you saying that you
> 
> > could cancel (say) a person speaking in the same room?
> 
> 
> 
> 
> 
> Where did I say that you can cancel anything at all?!  That's idiotic.
> 
> 
> 
> This is what I wrote:
> 
> "...the farther that you are away from your error sensor - assuming that you are still able to close your loop with some reasonable level of coherence - the less able you are to control non-tonal noise."
> 
> 
> 
> We were talking about control speaker placement.  I am saying that the farther you are away from your error sensor with your control speaker, the less able you are to control non-tonal noises due because of the delay that you are introducing into the control path.
> 
> 
> 
> The effects of loop delays in a feedback active noise control systems are brutal.  It does not take much distance at all between the control speaker and error sensor before you are restricted to tonal control only
> 
> 
> 
> Regarding "limitations..."
> 
> 
> 
> With respect to standard destructive interference noise cancellation using squared pressure control, your ability to do low frequency noise tonal reduction is typically limited by the capability of your speaker and amplifier to produce the equal but opposite destructive interference signal.  Your ability to do high frequency tonal reduction is typically limited by the size of your destructive interference "zone" (see Tokhi, et al.). 
> 
> 
> 
> The same limitations exist for non-tonal noise but they are trumped huge by the issue of control loop delay.  
> 
> 
> 
> ...and just because a Nyquist plot isn't telling you anything useful doesn't mean that it doesn't have anything useful to tell you.

Sincere thanks, but is there a formula somewhere where I can say for a given bandwidth of noise that can be cancelled what is required in the real world in terms of distances etc and if it is possible at all. Forget tonal noise, what about random noise or at least non narrowband type noise. suppose I wish to cancel 50Hz to 5kHz - why would this be not possible or possible as the case may be. Does this mean that the cancellation speaker would have to be so close that it is not practical?