Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[kraster]

Hi Dan,



I read Richard Black's 1999 AES paper on the effects of this phenomenon (inspired, I believe by Mr. Bob Katz's listening experiments). He maintains if there is(was) insufficient attenuation in Fs/2 at 44.1khz and that frequencies immediately above Fs/2 would be insufficiently filtered and cause Alisaing Intermodulation Distortion in the audible range.

Richard Black's experiments concluded that even small amounts of spurious frequencies could cause Intermodulation distortion:
"(The tweeter) was found to give audible intermodulation when fed with 9kHz (approx.) at -12dBW and 21kHz at -47dBW".

Does the filtering on current chip designs sufficiently attenuate frequencies in the stop band to minimise Alisaing Intermodulation Distortion?

Thanks,

Karl

[Terry Demol]

danlavry wrote on Sun, 26 June 2005 22:50

Terry Demol wrote on Sun, 26 June 2005 13:31 I see your point here, but I'm not sure I agree %100. It would depend on the ADC's inherent IMD at higher frequencies. It should be very low for a well designed unit and as such should not be a significant issue. Possibly some opamp based IP circuits are more susceptible to higher freq IM due to their rising distortion Vs rising freq characteristics but there are others that are very good HF performers. Also IME low gain apps such as ADC front ends are the easiest to keep linear at higher frequencies. Regards, Terry Terry, I believe you were talking about some ultrasonic impact that happens in the air. David Satz. pointed out that such interaction only takes place at very high pressure levels, and to the best of my knowledge he is correct. But I decided to go with the assumption that some degree of what you said may possibly be correct, and the simple logic I used suggests that it is UNDESIRABLE to include the ultrasonic. Now you are talking about AD non linearity at ultrasonic frequencies, which could completely changes the conversation. But "surprisingly", my answer is the same: Any non linearity at high frequency, be it a converter, the air itself, the speakers, is undesirable. The presence of such non linearity will ADD signals that were not present prior to recording. The signals we want are already picked by the mic.

Dan,

My apologies, I glossed through your post too quickly,
too busy these days.

Yes, I totally understand what you were referring to
WRT the "air IMD" happening in two instances. I thought you
were referring to electronic IMD at the ADC.

It would require speakers of sufficient bandwidth  however
there are plenty of tweeters that go out to at least 40k
these days. The popular ring modulator style come to mind.

It is an interesting subject in it's own right.  

Regards,

Terry

[danlavry]

kraster wrote on Mon, 27 June 2005 04:02

Hi Dan, I read Richard Black's 1999 AES paper on the effects of this phenomenon (inspired, I believe by Mr. Bob Katz's listening experiments). He maintains if there is(was) insufficient attenuation in Fs/2 at 44.1khz and that frequencies immediately above Fs/2 would be insufficiently filtered and cause Alisaing Intermodulation Distortion in the audible range. Richard Black's experiments concluded that even small amounts of spurious frequencies could cause Intermodulation distortion: "(The tweeter) was found to give audible intermodulation when fed with 9kHz (approx.) at -12dBW and 21kHz at -47dBW". Does the filtering on current chip designs sufficiently attenuate frequencies in the stop band to minimise Alisaing Intermodulation Distortion? Thanks, Karl

Karl,

You can go to my web site at www.lavryengineering.com and click on support. Look for my article named: "Sampling, Oversampling, Imaging, Aliasing". It is a PDF file.

In the file you will see some plots showing what happens when you have no oversampling, a X2 oversampling, X4 oversampling...
Notice the following:
When one begins with audio data that is limited to say 22KHz (CD format), and oversamples by say X2 to 88.2KHz, the frequency range between 22KHz and about 66KHz (88.2KHz-22KHz) is free of activity.
Say you up sample by X4 from 44.1KHz to 176.4KHz, there is a "dead zone" between 22KHz and 176-22=154KH.
Of course if you up sample higher, the "dead zone" increases.

The analog filter foe a DA needs to remove the high frequency image energy. In the case of X1 oversampling, you need to pass 20KHz and block 22.1KHz and that is a tough job - a 2KHz transition band. But in the case of X4, you need to pass 20KHz and block above 154KHz, Now you can have the ability to pass audio all the way to say even 54KHz, and still have 100KHz filter transition band. Moving the pass band to 54KHz helps gets you out of a big mess - the phase problems when the filter is right at 20KHz. Also, obviously, 100KHz transition band is much simpler filter than a 2KHz transition band!

Most DA's today have oversampling so high that one can set the filter way high above the audio and still reject the image energy by 120dB or more. The up sampling by some factor X pushes the image energy to very high frequencies by digital computation that can be phase linear.

A similar story, ending with a digital computation (that can be linear phase) hold for the AD side.

That is why I am saying: The analog 20KHz filter problem with it's associated phase problems is history and should be put aside. Your AD and DA, and everyone elses (unless it is very old gear) has oversampling and upsampling in it, therefore the problem of 20KHz analog filter is long gone.

Improving gear is not about 20KHz analog filters. We are long passed this bottleneck. Of course there are still issues to deal with, but we are making some progress.  

Regards
Dan Lavry
www.lavryengineering.com

[danlavry]

Richard Black's experiments concluded that even small amounts of spurious frequencies could cause Intermodulation distortion:
"(The tweeter) was found to give audible intermodulation when fed with 9kHz (approx.) at -12dBW and 21kHz at -47dBW".

Does the filtering on current chip designs sufficiently attenuate frequencies in the stop band to minimise Alisaing Intermodulation Distortion?

Thanks,

Karl

Please note that I am not taking issue with Mr. Blacks finding. I did not yet read his paper but see no reason to question that a 9KHz and 21KHz fed into some specific tweeter will generate inter-modulation. And maybe all tweeters generate some of that distortion.

But my comment was about audibility of analog filters at 20KHz, and my answer was - we do not need to worry about such filters that we no longer use.

The tweeter issue you mentioned seems to be a speaker maker problem. The makers of passive speakers have to work with analog filters for the cross over region between drivers (transducers), and the problem is "very serious" because the cross over frequencies are in the audible range, way below 20KHz. But that is a whole other issue - the issue of making good speakers over the audible range.

We are talking about the pro and con of extending the sampling rate, thus the audio bandwidth, assuming that the speakers and mic can cover the true audio hearing range (whatever it may be).

The problem is indeed that while we can not do a near perfect job to 20KHz, some people are looking for answer based on extending the bandwidth capability to 96KHz (192KHz sampling), about 3-4 time of what the best ear can hear.

If you want to clean your house, cleaning the houses of the 3 next door neighbours will not help

Regards
Dan Lavry
www.lavryengineering.com  

[kraster]

Thanks Dan,

Mr. Black's paper can be found here:
http://www.musaeus.co.uk/aespaper.htm


Thanks for your response. I wasn't sure if the technology Richard Black referred to in his paper was outdated (it was written in 1999). He acknowledges that oversampling techniques have improved filter performance but according to the late Julian Dunn it's the frequencies immediately above 20khz that cause the problem even at low levels. This may be a moot point now if filters are more efficient but it is surprising how little it takes to produce intermodulation effects in speakers.

I am not suggesting that increasing the sample rate will rectify the problem. On the contrary, since the majority of speakers top out at about 20khz I believe that higher sample rates might increase ID by allowing ultrasonic frequencies into speakers bandlimited to 20khz. If we lived in a world were speakers were completely linear up to 40khz it would not be a problem.


Karl

[danlavry]

kraster wrote on Tue, 28 June 2005 22:00

Thanks Dan, Mr. Black's paper can be found here: http://www.musaeus.co.uk/aespaper.htm Thanks for your response. I wasn't sure if the technology Richard Black referred to in his paper was outdated (it was written in 1999). He acknowledges that oversampling techniques have improved filter performance but according to the late Julian Dunn it's the frequencies immediately above 20khz that cause the problem even at low levels. This may be a moot point now if filters are more efficient but it is surprising how little it takes to produce intermodulation effects in speakers. I am not suggesting that increasing the sample rate will rectify the problem. On the contrary, since the majority of speakers top out at about 20khz I believe that higher sample rates might increase ID by allowing ultrasonic frequencies into speakers bandlimited to 20khz. If we lived in a world were speakers were completely linear up to 40khz it would not be a problem. Karl

Thank you for the link. I read the paper, and it talks about the inability of a speaker to deal with ultrasonic frequencies, causing intermode. The paper even went as far as to suggest some non real time (enabling a lot of computations) DIGITAL "mastering filter" to make sure there is no energy over 20KHz, to help the speaker problem.

After reading the paper, I do not see the paper as advocating going to higher sampling for the sake of elimination of 20KHz filters. I do not see the paper advocating capture of ultrasonic frequencies. On the contrary! The paper suggests complete elimination of ultrasonics, because it may cause inter-modulation distortions in the speaker. In other words, if I understand it correctly, the author would rather use a sharp 20KHz (or so) decimation filter, even for 96KHz and 192KHz sampling, to protect against inter-modulation based on ultrasonics.

In other words, I read what he says as: a 96KHz sampling system with a sharp 20KHz good decimation filter (like a 44.1KHz system) has the advantage over a 96KHz (or 192KHz) system with a more gradual filter.

I don't disagree. I see some other considerations for some slight increase in sample rate, and after taking all the factors into account we may find the optimum point - the best  sampling compromise, which in my view is around 50-70KHz, mics that do not go much above 20KHz, improved speakers and much more...

Regards
Dan Lavry
www.lavryengineering.com

[kraster]

Quote:

Thank you for the link. I read the paper, and it talks about the inability of a speaker to deal with ultrasonic frequencies, causing intermode. The paper even went as far as to suggest some non real time (enabling a lot of computations) DIGITAL "mastering filter" to make sure there is no energy over 20KHz, to help the speaker problem. After reading the paper, I do not see the paper as advocating going to higher sampling for the sake of elimination of 20KHz filters. I do not see the paper advocating capture of ultrasonic frequencies. On the contrary! The paper suggests complete elimination of ultrasonics, because it may cause inter-modulation distortions in the speaker. In other words, if I understand it correctly, the author would rather use a sharp 20KHz (or so) decimation filter, even for 96KHz and 192KHz sampling, to protect against inter-modulation based on ultrasonics. In other words, I read what he says as: a 96KHz sampling system with a sharp 20KHz good decimation filter (like a 44.1KHz system) has the advantage over a 96KHz (or 192KHz) system with a more gradual filter. I don't disagree. I see some other considerations for some slight increase in sample rate, and after taking all the factors into account we may find the optimum point - the best  sampling compromise, which in my view is around 50-70KHz, mics that do not go much above 20KHz, improved speakers and much more... Regards Dan Lavry www.lavryengineering.com

It shines a rather dubious light on the validity of listening tests. Perceived differences in listening tests could have a lot more to do with ID than "ultrasonic perception". As 96k will let more ultrasonic frequencies through (if the Mic is capable of capturing them from the source) it increases the chances of non linearity in the speaker.

An interesting point I was considering was that a lot of people perceive differences in inharmonic sources, cymbals, percussion etc. at higher sample rates. The frequency content of these sources can extend way up the frequency spectrum and would be a likely candidate to cause ID in speakers if captured. But because the source is inharmonic the resultant ID in speakers would appear as being correlated with the source thus giving the impression of more 'presence' where in fact it's just distortion!


Karl

[danlavry]

“It shines a rather dubious light on the validity of listening tests. Perceived differences in listening tests could have a lot more to do with ID than "ultrasonic perception". As 96k will let more ultrasonic frequencies through (if the Mic is capable of capturing them from the source) it increases the chances of non linearity in the speaker”

I am glad to see that you are including the mic in the equation. Most mics don’t pick up  much above 20KHz! Also, while talking about listening tests, I wish we were talking about some double blind ABX tests. The fact is we are talking about “reports” which is a “different thing” all together.

An interesting point I was considering was that a lot of people perceive differences in inharmonic sources, cymbals, percussion etc. at higher sample rates. The frequency content of these sources can extend way up the frequency spectrum and would be a likely candidate to cause ID in speakers if captured. But because the source is inharmonic the resultant ID in speakers would appear as being correlated with the source thus giving the impression of more 'presence' where in fact it's just distortion!

Again, I really do not think ultrasonics plays much of a role here, because very few mics just work there. But your comment is very interesting:

You said: “The frequency content of these sources can extend way up the frequency spectrum and would be a likely candidate to cause ID in speakers if captured. But because the source is inharmonic the resultant ID in speakers would appear as being correlated with the source thus giving the impression of more 'presence' where in fact it's just distortion!”

This comment may be applied to the audio content without any ultasonics present in the signal. Inter-modulation due to ultrasonic energy is only one mechanism for distortion…

Regards
Dan Lavry
www.lavryengineering.com

[hiendaudio]

Hi:

If all people that claim beyond 20K audio , see a far field impulse response of a concert hall, they

[kraster]

Given the evidence in Dan's 192khz white paper and the acknowledgement by many researchers that somewhere around 60khz is the optimal samping frequency. Whose bright idea was it to jump up to 96 and ,in particular, 192k sampling rates? Is there a collusion between various hardware manufacturers (ie. Hard-disk makers and DSP makers) and AD/DA makers (with the exception of Mr. Lavry) to keep pushing the sample rate up?

When all is said and done the most compelling reason for utilising high sample rates is that it financially benefits a select few hardware manufacturers.

I know the above statement can be taken as a given but without any compelling technical argument in its favour how can they (the disk/dsp makers) still get away with it?

Karl