[jaddie]

Quote:

BTW, is there any way to bargain on that 30db noise floor? Can we find out the max recommended level for a recording studio and go with that number? Granted nobody lives in an anechoic chamber, but 25db seems easily attainable, and 30db is more than 2x as loud as that.  

Studios are built to meet NC curves (Noise Criteria) which are based on noise audibility vs frequency.  When octave band noise is measured the SPL reading is plotted.  The NC specification is the curve below which all octave band readings fit.  One high band can bump the entire reading up.
F-6_graph1.gif
 
NC-20 is good, there are many studios at NC-15, and a few at NC-10.  The NC figure is almost always impacted the hardest by air handling, but outside impinging noise can also be a factor, such as trying to build a studio next to a 4-lane highway. 
 
Home living rooms easily do NC-30, NC-25 isn't all that unusual, and there are quite a few at NC-20 between air-handler cycles.
 
To make this meaningful in a headphone system the noise presented at the ear would be what's important, and would be calculated as the result of system noise, total system gain and headphone/IEM sensitivity.  Plotting this against NC-type curves might seem to be relevant, but in reality electronic systems have greater noise at higher frequencies, so higher tolerance of LF noise would be unnecessary.  Some sort of noise weighting would make sense though. A weighting might be reasonable since it seems to persist, but wouldn't it be cool to use ITU-R 468?  So, scale the noise spec to SPL with ITU-R 468 weighting.  Could spec "ears" with minimum noise at 0dB SPL with the same weighting, which is technically wrong, but easy to do and frankly close enough.
 
Footnote: It's a small point, but a 10dB change is perceived as either doubling or halving the volume even though a little more than 6dB is a double/halving of pressure/voltage and 3dB doubling/halving in power. 

 

[ultrabike]

 

[jaddie]

Quote:

10% second-order seems very high, easy to detect.  I just generated 220 Hz + 440 Hz (2nd harmonic at 10% amplitude) vs. pure 220 Hz, replaygain, and it was obvious, at least for that pure tone.

I had to dig and think to recall where these figures came from.  I thought it was an AES paper, but no.  Finally got it.  David L. Clark used to take a demo around to AES chapters that included his "chamber of horrors" box, capable of generating fixed amounts of distortion applied to music program material.  He would audition the box vs no box in his ABX system.  10% of second-order harmonic distortion in music material was actually much more difficult to detect than you'd ever expect.  Odd-order amounts of 3% were detectable reliably, 1% levels were not, I think we all scored about 50% (guessing) on that one.  He also threw in a ton of group delay (don't recall the specific amount, but it was many times that of a typical 8-pole anti-aliasing filter), which was also not reliably detectable. The take-away from the demo was how much distortion of certain kinds was not detectable. 
 
However, for this discussion, the typical vanishingly low distortion figures of amps are way below what can be detected.  To simplify things, if we were to say that if all forms of harmonic distortion generated by typical nonlinearities were kept below .1% we'd be below what's detectable by normal hearing.
 
It's probably not reasonable to test for distortion audibility by synthesizing a contrived pure-tone test with only distantly removed harmonics, like first and fifth.  That structure never happens as an actual result of nonlinearity.  All pure tone tests show distortion audibility at lower levels than program material.  

 

[anetode]

Quote:

I had to dig and think to recall where these figures came from.  I thought it was an AES paper, but no.  Finally got it.  David L. Clark used to take a demo around to AES chapters that included his "chamber of horrors" box, capable of generating fixed amounts of distortion applied to music program material.  He would audition the box vs no box in his ABX system.  10% of second-order harmonic distortion in music material was actually much more difficult to detect than you'd ever expect.  Odd-order amounts of 3% were detectable reliably, 1% levels were not, I think we all scored about 50% (guessing) on that one.  He also threw in a ton of group delay (don't recall the specific amount, but it was many times that of a typical 8-pole anti-aliasing filter), which was also not reliably detectable. The take-away from the demo was how much distortion of certain kinds was not detectable. 
 
However, for this discussion, the typical vanishingly low distortion figures of amps are way below what can be detected.  To simplify things, if we were to say that if all forms of harmonic distortion generated by typical nonlinearities were kept below .1% we'd be below what's detectable by normal hearing.
 
It's probably not reasonable to test for distortion audibility by synthesizing a contrived pure-tone test with only distantly removed harmonics, like first and fifth.  That structure never happens as an actual result of nonlinearity.  All pure tone tests show distortion audibility at lower levels than program material.  

 
Testing down the ladder I would agree that it became much tougher after the harmonics were 30db below the fundamental. With samples of music it's much more difficult to ABX a difference and there are circumstances where up to 25% THD has been judged as imperceptible by testers.
 
As for the simplification, I'd still like to advise 1% for the simple reason that it is very hard to find a headphone/speaker that keeps to below .1% over a wide bandwidth at any sort of decent volume, while there are plenty of headphones which will keep to below 1% 20-20 even on peaks.

 

[jaddie]

As for the simplification, I'd still like to advise 1% for the simple reason that it is very hard to find a headphone/speaker that keeps to below .1% over a wide bandwidth at any sort of decent volume, while there are plenty of headphones which will keep to below 1% 20-20 even on peaks.

The problem with 1% is that there are 1% distortions that are audible and others that are not. And we're back to not using a single number. Yes .1% won't be realistic for transducers. If we're trying to come up with specs for systems that relate to hearing, perhaps there still needs to be a set for different types of devices that are achievable and realistic rather than idealized.

 

[Strangelove424]

Studios are built to meet NC curves (Noise Criteria) which are based on noise audibility vs frequency.  When octave band noise is measured the SPL reading is plotted.  The NC specification is the curve below which all octave band readings fit.  One high band can bump the entire reading up.
F-6_graph1.gif
 
NC-20 is good, there are many studios at NC-15, and a few at NC-10.  The NC figure is almost always impacted the hardest by air handling, but outside impinging noise can also be a factor, such as trying to build a studio next to a 4-lane highway. 
 
Home living rooms easily do NC-30, NC-25 isn't all that unusual, and there are quite a few at NC-20 between air-handler cycles.
 
To make this meaningful in a headphone system the noise presented at the ear would be what's important, and would be calculated as the result of system noise, total system gain and headphone/IEM sensitivity.  Plotting this against NC-type curves might seem to be relevant, but in reality electronic systems have greater noise at higher frequencies, so higher tolerance of LF noise would be unnecessary.  Some sort of noise weighting would make sense though. A weighting might be reasonable since it seems to persist, but wouldn't it be cool to use ITU-R 468?  So, scale the noise spec to SPL with ITU-R 468 weighting.  Could spec "ears" with minimum noise at 0dB SPL with the same weighting, which is technically wrong, but easy to do and frankly close enough.
 
Footnote: It's a small point, but a 10dB change is perceived as either doubling or halving the volume even though a little more than 6dB is a double/halving of pressure/voltage and 3dB doubling/halving in power. 

 
 
Thanks for the info, that answers a lot of questions for me. Some of the info is surprising though, and I have more questions now... 50db at 53Hz for NC-20 shocked me, is this just a function of structural sound waves, wind, etc. blowing on the building being unavoidable? And how do these become expressed on this kind of chart as 15-30dB, is that level an average*? Also, how realistic is ITU-R468? I thought that was anechoic chamber territory. I can imagine corporate-run studios fighting it and amateur ones ignoring it.
 
The footnote was also helpful. I thought 3db=2x. I'm assuming the pressure resistance for air and impedance in circuits change the other numbers? I kinda of pelted you with questions, but any further info you could give me would be appreciated. 
 
* This is the chart I'm talking about in the spoiler, from http://www.engineeringtoolbox.com/nc-noise-criterion-d_725.html
 

Spoiler: Warning%3A%20Spoiler!

 
[size=14.399999618530273px]

Type of Room - Space Type	Recommended NC Level NC Curve	Equivalent Sound Level dBA
Residences
Apartment Houses	25-35	35-45
Assembly Halls	25-30	35-40
Churches	30-35	40-45
Courtrooms	30-40	40-50
Factories	40-65	50-75
Private Homes, rural and suburban	20-30	30-38
Private Homes, urban	25-30	34-42
Hotels/Motels
- Individual rooms or suites	25-35	35-45
- Meeting or banquet rooms	25-35	35-45
- Service and Support Areas	40-45	45-50
- Halls, corridors, lobbies	35-40	50-55
Offices
- Conference rooms	25-30	35-40
- Private	30-35	40-45
- Open-plan areas	35-40	45-50
- Business machines/computers	40-45	50-55
Hospitals and Clinics
- Private rooms	25-30	35-40
- Operating rooms	25-30	35-40
- Wards	30-35	40-45
- Laboratories	35-40	45-50
- Corridors	30-35	40-45
- Public areas	35-40	45-50
Schools
- Lecture and classrooms	25-30	35-40
- Open-plan classrooms	35-40	45-50
Movie motion picture theaters	30-35	40-45
Libraries	35-40	40-50
Legitimate theaters	20-25	30-65
Private Residences	25-35	35-45
Restaurants	40-45	50-55
TV Broadcast studies	15-25	25-35
Recording Studios	15-20	25-30
Concert and recital halls	15-20	25-30
Sport Coliseums	45-55	55-65
Sound broadcasting	15-20	25-30

​

[/size]  

 


Quote:

 
Testing down the ladder I would agree that it became much tougher after the harmonics were 30db below the fundamental. With samples of music it's much more difficult to ABX a difference and there are circumstances where up to 25% THD has been judged as imperceptible by testers.
 
As for the simplification, I'd still like to advise 1% for the simple reason that it is very hard to find a headphone/speaker that keeps to below .1% over a wide bandwidth at any sort of decent volume, while there are plenty of headphones which will keep to below 1% 20-20 even on peaks.

 
 
I think THDs to a certain extent represent engineering quality, and value/dollar. Whether it can be heard or not is debatable, and the bottleneck is in the transducer unless you find a real stinker of an DAC or amp. But if you're aim is all out neutrality, as a measurement of value during purchase, they can be helpful.  

 

[jaddie]

Quote:

 
 
Thanks for the info, that answers a lot of questions for me. Some of the info is surprising though, and I have more questions now... 50db at 53Hz for NC-20 shocked me, is this just a function of structural sound waves, wind, etc. blowing on the building being unavoidable? And how do these become expressed on this kind of chart as 15-30dB, is that level an average*? 

The NC curves represent how noise is heard.  Hearing is much less sensitive at low frequencies, and even more so at lower specific SPL.  So the chart reflects that you can have a 50dB SPL noise in the octave band centered at 53Hz, and it will contribute no more to the perceived noise level than 23dB SPL in the 1KHz octave band.  The chart helps assign a single figure, such as NC-20, to a set of octave band SPL measurements for which no one band exceeds the NC-20 curve.  
 
 
Quote:

 Also, how realistic is ITU-R468? I thought that was anechoic chamber territory. I can imagine corporate-run studios fighting it and amateur ones ignoring it.

 ITU-R468 is a specified weighting curve applied to noise measurements in electronics, not acoustics.  It's similar in purpose to A weighting (which is also used in acoustics) but reflects more recent research into the type of noise that we can hear, and weights that area of spectrum more appropriately.  The in-between weighting curve was CCIR, preferred by Dolby when they introduced their B-type noise reduction for cassettes and consumer tape.
 
The purpose of using a weighting curve or network for noise measurements is to make them more meaningful.  Weighting curves are trying to put more emphasis on noise we can easily hear and less emphasis on noise we can't.  An example: a very wide spectrum noise measurement of an amp shows a -70dB noise floor...not very impressive.  But the bandwidth of that test might be 1MHz. We can't hear above 20KHz, and in electronic circuits, noise levels rise with frequency, typically at 6dB/Octave.  So cutting off the noise measurement above 20KHz cuts of around 6 or 7 octaves worth of noise, and the result might be a significantly lower figure.  But, at low levels, we can't hear LF noise well at all, so a weighting filter is added to de-emphasize LF noise, and the figure drops further.  It's all legal because we're not measuring noise we can't hear.  The NC curves, the A-weighting curve, and the  ITU-R468 curve all do this basic job, but for different purposes and with varying accuracy.
 
The purpose of an anechoic chamber is to have an acoustic measurement space with no reverb or reflections that could reduce measurement resolution.  It's also useful to have high noise isolation too, but depending on the type of testing, having some residual noise in the chamber may not make it unusable.  They typically are quiet spaces, though.  
 
Quote:

The footnote was also helpful. I thought 3db=2x. I'm assuming the pressure resistance for air and impedance in circuits change the other numbers? I kinda of pelted you with questions, but any further info you could give me would be appreciated. 

Approximately...
6dB=2X or .5X voltage or pressure (SPL)
3dB=2X or .5X power
 
But...what we perceive is different.  If you ask someone to select a volume change that seems to be twice the volume, they pick +10dB change.  Same thing for half the volume, -10dB change.  If you ask someone to press twice as hard on your arm, the approximate difference in pressure corresponds to the same ratio as a 10dB change.  It also applies to how we see light.  It's human perception. 
 
 

 
 
I think THDs to a certain extent represent engineering quality, and value/dollar. Whether it can be heard or not is debatable, and the bottleneck is in the transducer unless you find a real stinker of an DAC or amp. But if you're aim is all out neutrality, as a measurement of value during purchase, they can be helpful.  

I'd be very surprised at any DAC or amp that would have THD as high as some transducers.  They are usually at least an order of magnitude better.  The problem this causes is consumer confusion.  Why should I pay for an amp or DAC that does .005% THD vs one that does .01% THD? You can't make that decision without more info, like the conditions of the test at very least.  Of course, that difference is inaudible anyway, but you get the idea.  

 

[Strangelove424]

Ok, some of the charts I was looking at confused me, but I think I'm understanding it a bit better now. I originally thought ITU-R468 was an acoustic SPL standard, not a circuit or equipment noise standard, so that explains a few things. I'm a little foggy on the idea of weighting still, but it seems to me that it's basically there, at differing degrees of accuracy, to create a little extra wiggle room for less sensitive bands in human perception. Thanks for clarifying.  

 

[jaddie]

I'm a little foggy on the idea of weighting still, but it seems to me that it's basically there, at differing degrees of accuracy, to create a little extra wiggle room for less sensitive bands in human perception. Thanks for clarifying.  

I wouldn't say it quite that way. Weighting is used so that measured noise figures corellate well with how we would hear that particular noise. The intent is to present a more meaningful noise number.

All weighting curves have at their core the sensitivity of human hearing at low levels, most prominently the rather significant reduction in sensitivity as frequency goes down, but also including smaller changes in Hf sensitivity. The problem with fixed weighting curves is that they are only really correct at a certain SPL (hence the NC curve family). Since when measuring electronics we probably have no idea of total system gain, we also have no idea of the specific SPL for each octave band, so a generalized fixed single weighting curve is used. It's not really a problem, though, as noise measurements are presented as comparisons to other devices measured the same way.

 

[bigshot]

I think I'd rather have a little slack in the figure than use a different measuring system than the one they will be seeing on equipment specs. I'll err on the side of being too conservative and use the quietest library figure, even though most living rooms would be much higher. Are there published specs on the degree of external noise that creeps through IEMs? That would probably be considered the noise floor for them. 15-20 dB?
 
1% distortion seems to me to be pretty safe for a non linear distortion threshold.
 
Can someone point to a web page that gives specs on jitter audibility?
 
What amount of dB is audible? 1 dB? .5 dB? Web cite?
 
Length of time for auditory memory?
 
What other specs would be useful?

 

[stv014]

Quote:

As for the simplification, I'd still like to advise 1% for the simple reason that it is very hard to find a headphone/speaker that keeps to below .1% over a wide bandwidth at any sort of decent volume, while there are plenty of headphones which will keep to below 1% 20-20 even on peaks.

 
<0.1% distortion at 20 Hz to 20 kHz is hard to find in dynamic transducers, but at mid to high frequencies (e.g. 200 Hz to 20 kHz at 90 dB SPL at 1 kHz) it is not that rare. Basically, the distortion usually increases with excursion, so it is highest at low frequencies and at high SPL. So, high frequency distortion in badly designed electronics can actually exceed that of the headphone or loudspeaker (which is normally worst at bass distortion and low+high frequency IMD).
 
Here are a few distortion graphs I made using my DT880 250 Ω:
 
THD vs. frequency (the highest frequencies are not included, because the measurement has very inaccurate frequency response there, and the distortion is also dominated by the microphone, in fact not unlikely already in the kHz range). Green is frequency response, yellow is THD (2nd and 3rd order included, anything higher is basically under the noise floor) at -10 dBv, red is THD (up to 4th order) at 0 dBv. 0 dBv is about 100 dB SPL with this headphone, based on InnerFidelity and doctorhead.ru measurements. Note that the THD is not a percentage, but is referenced to the same 0 dB level as the frequency response, so the THD % can be calculated from the difference from the frequency response.
 

 
THD spectrum at 500 Hz (yellow: -10 dBv, red: 0 dBv, this applies to all the graphs). At 3000 Hz, the peak is interference or ambient noise, it is not distortion.
 

 
55 Hz + 2000 Hz 4:1 IMD (-2 and -14 dB levels relative to the single sine waves used in the previous tests). This is where headphones perform the most badly, and the distortion increases with the level.
 

 
15000 Hz + 16000 Hz 1:1 IMD (-6 dB relative levels). The reason why the level seen on the graph is not -6 dB is that the tones are attenuated partly by the headphone, and especially by the measurement setup (not by the microphone itself, but because of its placement). However, the most important 1 kHz IMD product from the headphone is still measured accurately enough. Only at 1, 14, and 17 kHz at the higher level can real distortion be seen, the rest is noise from various sources. At ~-80 dB vs. the -6 dB test tones it is not bad, and this is where a good headphone can easily be better than poorly designed electronics.
 

 

[stv014]

Quote:

Originally Posted by bigshot /img/forum/go_quote.gif
 
What amount of dB is audible? 1 dB? .5 dB? Web cite?

 
Do you mean loudness or frequency response ? For loudness with fast switching, >0.1 dB is commonly assumed to be the threshold (someone at the HydrogenAudio forums actually posted a successful ABX of 0.1 dB level difference, but hearing that small difference with music is not typical). For frequency response, I was able to ABX a 0.5 dB bass boost without major difficulty, but I was not specifically looking for a threshold, so smaller differences could very well be audible. A variation of at most 0.1 dB is again a reasonably safe threshold for a transparent, flat response, although at some frequencies it is overkill. Of course, all the above is with almost instantaneous switching (< 200 ms), see also auditory memory below. Without fast switching, it is much harder to hear the difference. The "1% magnitude rule" could also be extended to crosstalk, i.e. better than -40 dB at 20 Hz to 20 kHz might not be much of an issue.
 
Quote:

Originally Posted by bigshot /img/forum/go_quote.gif
 
Length of time for auditory memory?

 
200 ms is a figure I have seen at more than one place for accurate detection of small level differences (down to 0.1 dB). But the loss of information occurs in multiple stages, this is explained at this page (I cannot say for sure if the information there is correct).
 
Quote:

What other specs would be useful?

 
I see group delay is listed, but did anyone mention ITD (delay between stereo channels, rather than the same delay on both) ? I think the minimum that was found to be audible is 10 us, but I need to find a link.

 

[anetode]

Quote:

 
<0.1% distortion at 20 Hz to 20 kHz is hard to find in dynamic transducers, but at mid to high frequencies (e.g. 200 Hz to 20 kHz at 90 dB SPL at 1 kHz) it is not that rare. Basically, the distortion usually increases with excursion, so it is highest at low frequencies and at high SPL. So, high frequency distortion in badly designed electronics can actually exceed that of the headphone or loudspeaker (which is normally worst at bass distortion and low+high frequency IMD).

 
Yup, also the detection threshold goes down (edit: that is, less sensitivity) in the bass frequencies. The IMD distortion introduced by electronics is the most interesting part for me as it might explain some of the "flavor" people attribute to amps/dacs.
 
Quote:

 
I see group delay is listed, but did anyone mention ITD (delay between stereo channels, rather than the same delay on both) ? I think the minimum that was found to be audible is 10 us, but I need to find a link.

 
jnjn brought this up in a past thread, the citation is
Binaural time discrimination, Jan O. Nordmark, Journal  Acoustic Society of America, Vol 60, no. 4, October 1976
 
I haven't read the above, but about the only difference I could imagine this having would be the slightest shift in terms of lateral localization.

 

[xnor]

Dunno if this has been posted before:
ABX Amplitude vs. Frequency Matching Criteria

 

[obobskivich]

I think I'd rather have a little slack in the figure than use a different measuring system than the one they will be seeing on equipment specs. I'll err on the side of being too conservative and use the quietest library figure, even though most living rooms would be much higher. Are there published specs on the degree of external noise that creeps through IEMs? That would probably be considered the noise floor for them. 15-20 dB?

1% distortion seems to me to be pretty safe for a non linear distortion threshold.

Can someone point to a web page that gives specs on jitter audibility?

What amount of dB is audible? 1 dB? .5 dB? Web cite?

Length of time for auditory memory?

What other specs would be useful?

I've seen demonstrations on dB steps - I agree with stv014 on either .1 or .5 but it's not something you're likely to notice in conventional listening, especially with background noise (as was mentioned earlier). 1 dB is probably a safer overall bet, but I think .1 dB is JND range; I can dig up a cite if you want. Regarding frequency response variations, Russell's site claims ~2 dB for JND in FR variations - but I don't know if there's a frequency component or not (I'm guessing there might be, like Ethan said regarding sensitivity to different frequency ranges). I think this plays more into channel matching than anything else (so you have manufacturers talk about .000001 dB versus .0001 dB and all that when it comes to channel matching - my understanding is that both are under audible threshold, and this is usually more of a problem with really cheap pots (that won't track evenly) and speakers).

On auditory memory, 200ms-250ms is what I've heard in the past (I've seen 250ms stated as overall average response time in haptic systems (can be coached/trained lower, but not much lower afaik), I don't know if that's a limiting factor - the action potential is certainly moving faster)) - I've seen higher for SM/echoic memory though - up to around 5s; seems to correlate to age *and* cognitive development - try this one out: http://www.kjp.med.uni-muenchen.de/download/Glass-JNT2008.pdf (I'm too lazy to go dig in my basement for books, that was on Wikipedia). I'm sure this could be coached to some extent as well, but my understanding is that echoic memory doesn't work in the same way as visual memory because the listener only gets the impulse once, while the viewer gets the impulse for as long as the condition allows.

Group delay would be another good one - I remember Danley saying years ago that group delay, like THD, is frequency dependent (it can go higher inversely related to frequency), but I forget what the general rule of thumb was - something like 1/6th of frequency iirc. But I don't remember that being a hard limit (it isn't all or nothing, there are some systems that will be *slightly* over (usually due to a port)) and aren't doomed like systems that are *massively* over. I'm unaware of anything conclusive on audibility. And group delay tends to be ignored with headphones.

Here's some quick searching:
http://www.adireaudio.com/Files/GroupDelay.pdf
http://www.speakerdesignworks.com/group_delay.html

Gets back into discussion of Q and damping.