Digital Drive

Upsamplers, DACs, jitter, shakes and analogue withdrawals, this is it.

Return to Digital Drive

Minimum Phase Question

99.60.17.41
Posted on September 8, 2009 at 17:59:57
J. Phelan
Audiophile

Posts: 73
Joined: May 12, 2009
 Mainly to Charlie H. - do you think this filter scheme "eliminates errors" at the ADC...or do you think that it simply removes errors previously occuring in playback ?

RE: Minimum Phase Question, posted on September 8, 2009 at 23:40:29
Todd Krieger
Audiophile

Posts: 22458
Location: SW United States
Joined: November 2, 2000
 "Mainly to Charlie H. - do you think this filter scheme 'eliminates errors' at the ADC...or do you think that it simply removes errors previously occuring in playback ?"

The filters do not exactly "remove errors"...... (There are no actual "errors" or "error handling" in the process.) They block out ultrasonic artifacts from reflecting into the audio band for A/D, and block out audio band "aliases" that would reflect into the ultrasonic range for D/A. (There is debate whether the D/A filtering is necessary, but it is definitely necessary for A/D.)

That said, since the "decays" (or "sustains") in music are generally longer in duration than "attacks", I do think minimum phase filtering would preserve "waveform fidelity" of the pre-digitized music signal better than linear phase filtering.


Two different questions, posted on September 8, 2009 at 21:15:55
Charles Hansen
Manufacturer

Posts: 4355
Joined: August 1, 2001
 >> do you think this [minimum phase] filter scheme "eliminates errors" at the ADC? <<

There are two different things involved with the new digital filters in our products, because they offer two different user-selectable options.

Both are minimum-phase. We found in our listening tests that this sounds noticeably better than the typical linear-phase filters. I think that linear-phase filters have been universally used for two reasons:

a) They are simpler (and cheaper) to implement.

b) At the time that digital was being introduced as a new format, "linear phase" loudspeakers were quite the buzzword. It was commonly accepted that this was a commendable goal.

But the problem is that linear-phase digital filters create their own type of time distortions. These are audible when comparing against equivalent minimum-phase filters.

But there is no claim by anybody that minimum-phase filters by themselves can "fix" anything. Instead, Peter Craven wrote a paper published in the AES Journal where he described what he called an "apodizing" filter.

It is this apodizing filter that is *supposed* to filter out problems created by the ADC. But I can't say that the thing really works. It doesn't sound bad, in fact it sounds pretty good. But it doesn't sound as good to me as our "Listen" filter that *doesn't* filter out the alleged ringing that the ADC puts on the disc.

It's all described in the white paper on our website in the "What's New" section. Most people who listen to the filters agree that the "Listen" setting that is *non*-apodizing sounds better than the apodizing "Measure" filter. Below is a link to the Stereophile CES show report that confirms this, although the writer was slightly confused -- the listening comparison was between the two different MP filters. All present preferred the "Listen" position.

So whatever problems may or may not exist during the ADC process, I don't think that the apodizing filter cures many (if any) of them.

RE: Two different questions, posted on September 9, 2009 at 17:43:33
J. Phelan
Audiophile

Posts: 73
Joined: May 12, 2009
 Thanks for the reply. It seems that certain audio reviewers are pushing the ADC problem w/o *any* evidence that this is occuring. Craven, for all his intelligence on digital, opened a small can of worms...

RE: Two different questions, posted on September 9, 2009 at 18:22:28
Tony Lauck
Audiophile

Posts: 3323
Location: Vermont
Joined: November 12, 2007
Contributor
  Since: February 24, 2009
 "It seems that certain audio reviewers are pushing the ADC problem w/o *any* evidence that this is occuring."

There is no need for evidence. It's a fact that filters are used with ADCs and filters are used with DACs. It's also a fact that the transfer function of the record-playback system depends upon both the record and the playback filters. They work together.

Unfortunately, Sony and Philips in the infinite wisdom that brought us "perfect sound, forever" failed to specify these filters, either on the record side or on the playback side. So the situation is much like it was in the early days of LPs before there was general agreement on the RIAA curve. However, as the different ways of doing filtering affect primarily high frequency phase response, differences between approaches are subtle, which is why it has taken so long to get to the present point.

It has been my experience that some recordings are more affected by the choice of playback filters than others. It's not clear whether this depends on the filters used in the recording process or other aspects of the recording production, such as the number and location of microphones, use or non-use of equalizers, etc...

Tony Lauck

"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar

RE: Two different questions, posted on September 9, 2009 at 19:28:20
J. Phelan
Audiophile

Posts: 73
Joined: May 12, 2009
 Exactly my point. No-one can say (for certain) that "ringing" occurs at a properly-filtered ADC, as Charlie H. says. Minimum-phase filtering works by removing artifacts in *playback*, not the ADC....

RE: Two different questions, posted on September 9, 2009 at 21:15:15
Werner
Audiophile

Posts: 1851
Joined: September 30, 1999
 > No-one can say (for certain) that "ringing" occurs at a properly-filtered ADC

For starters, what constitutes 'proper filtering' for an ADC, in the context of a 44.1kHz system. I don't know. Do you?

But given that 99% of ADCs out there uses a linear-phase FIR I can assure you that ringing occurs. That's just math.
And since these ADCs are also half-band, they alias a bit. In it this is not a problem, but it assures that the recorded signal extends right to Fs/2, and this, upon playback, causes the DAC to ring as well.

Mind, we don't know if ADC pre-ringing is an audible problem. When translated to the mid-band, pre-ringing is vile (ask the developers of perceptual coders). But when it sits above 12kHz or so the situation is far less clear. It might be a problem, and a possible solution for that potential problem has been found ('Hurrah go the markets, up goes the stock), and is being applied...


> Minimum-phase filtering works by removing artifacts in *playback*, not the ADC...."

Pray tell us, which playback artefacts exactly?



bring back dynamic range
Proper Filtering for an ADC, posted on September 10, 2009 at 06:41:04
Tony Lauck
Audiophile

Posts: 3323
Location: Vermont
Joined: November 12, 2007
Contributor
  Since: February 24, 2009
 "For starters, what constitutes 'proper filtering' for an ADC, in the context of a 44.1kHz system. I don't know. Do you?"

The Red Book should have specified that the DAC used perfect SINC filtering. (This amounts to defining the analog waveform that corresponds to any particular CD track.) Then the ADC can use whatever filtering the mastering engineer wants. By the sampling theorem he can make the resulting playback come out any way he wants within the Nyquist constraint. The reason for leaving the choice on the recording side is that it is an artistic tradeoff how to fit the characteristics of a particular recording into the format limitations and there are a number of interrelated considerations, e.g. existing high frequency content in the source material, microphone set ups and its effect on imaging, etc... In addition, the chances are that most mastering engineers are more experienced and have better playback equipment than nearly all consumers. Finally, it is more efficient to do the careful engineering once, rather than to leave it up to each consumer.

The Red Book specification didn't include what it should have and many (perhaps most) RBCDs sound better when played back with a linear phase filter. This amounts to a consumer choice of a "tone control" or "EQ" setting. If it works, enjoy it! If a few more RBCDs can be made to sound good this way, great! There is nothing "right" or "wrong" about this.

As to what specifically most ADCs use, I can only speak to the production process that I use, which is to record at 88.2 or 176.4 kHz 24 bits. I assume the ADC uses a linear phase filter, but this is irrelevant, as any ringing is going to be at 44.1 or 88.2 kHz. When it comes time to convert down to 44.1 kHz I use a sample rate converter, and this is where the critical ringing and linear phase/minimum phase issue arises. I am presently using Izotope 64 bit which allows for control over the transition frequency, smoothness of the filtering, and choice of phase (minimum phase, linear phase, or any combination thereof). I generally use minimum phase with the filter parameters set so any aliasing is below -120 dB.


Tony Lauck

"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar

RE: Two different questions, posted on September 9, 2009 at 21:39:35
J. Phelan
Audiophile

Posts: 73
Joined: May 12, 2009
 The playback artifact of pre-ringing.

But I'm glad you admit that there's no proof of *audible* ADC ringing. Meridian Audio says there is and audio reviewers believe them....

RE: Two different questions, posted on September 9, 2009 at 22:24:39
Werner
Audiophile

Posts: 1851
Joined: September 30, 1999
 "The playback artifact of pre-ringing. "

If you read and fully understand Shannon's theorem, then it will become clear that, given a correctly-bandlimited signal, a correctly-implemented playback chain will *not* pre-ring. (This can be formally proven. It can be demonstrated too, provided one can source a correctly-limited input signal.)

If there is pre-ringing upon playback, then either the ADC allowed non-insignificant aliasing to happen (i.e. wasn't down to -120dB or so at Fs/2) and/or the ringing was introduced by the ADC's own anti-aliasing filter.

But I do admit that all of this has a bit of a chicken/egg flavour.




bring back dynamic range
RE: Two different questions, posted on September 9, 2009 at 22:49:23
J. Phelan
Audiophile

Posts: 73
Joined: May 12, 2009
 Pre-ringing *is* happening - ask anyone involved with audio playback design. The use of minimum-phase filters introduces echo *after* the transient, not before - so the theory goes. And what a big difference in sound !!

And since the only parameter we're changing is the playback filter, then it *must* have been the (audible) distortion artifacts of linear-phase filters.......

RE: Two different questions, posted on September 10, 2009 at 08:53:20
Tony Lauck
Audiophile

Posts: 3323
Location: Vermont
Joined: November 12, 2007
Contributor
  Since: February 24, 2009
 It's not easy to understand what is happening. However, it is the case that a linear phase filter followed by a minimum phase filter will produce the same result as a minimum phase filter followed by a linear phase filter. You can see this with test files using commercial sample rate converters, such as the Izotope 64 bit SRC. Or you can do the mathematics, and observe that convolution is commutative. (No need for calculus or infinite series if FIR filters are involved.)

Because of my training as a mathematician I was wary of theoretical arguments that (linear phase) SINC filtering was perfect. Engineers often misuse theory, generally by making assumptions that don't apply in the real world, but which are convenient because they simplify the mathematical calculations. In the case at point, the sampling theorem provides for perfect reconstruction of sampled data, but only where the original signal was bandlimited to below the Nyquist frequency. Unfortunately, it also a theorem of mathematics that there are no band-limited signals that have finite duration, with the exception of the zero signal. In other words, no musically interesting recordings can possibly meet the conditions of the sampling theorem. At best an approximation is possible. So I decided to test existing sample rate converters and see how close an approximation to theory they achieved. I expected agreement to well below -100 dB with 24 bit formats, ideally, below -130 dB or so. Did I see that? No. Not close. Of several that I measured, the errors when converting band limited signals were more like -60 dB. It appeared that both practice and theory were bankrupt. However, it is the existing converters that are imperfect, but I was able to convince myself only after figuring out how to build a sample rate converter that doesn't have these problems, i.e. it implements projection and injection between signal spaces "correctly". So yes, the theory is valid, and it can be realized in practice as closely as desired, but that's not generally how most implementations work. (I am not giving away the secret here so as to leave it as an interesting challenge for those interested.)

If it makes you feel any better, none of this is easy. Claude Shannon didn't even state the sampling theory correctly in his classic paper: He allowed a signal at the Nyquist frequency, rather than restricting Theorem 1 to signals below the Nyquist frequency. This is a small point, perhaps, but it is right at the "edge" where difficulties exist in practice.

How important the differences are between minimum phase and linear phase filters is an entirely separate question. I wouldn't go so far as to call them big. The differences are subtle at best. If they were big, then people wouldn't have the slightest difficulty passing ABX tests on these filters starting with actual recordings. (I say "at best" because with some recordings there won't be transients that will provoke filter differences.)

Tony Lauck

"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar

RE: Two different questions, posted on September 9, 2009 at 23:51:06
Werner
Audiophile

Posts: 1851
Joined: September 30, 1999
 "Pre-ringing *is* happening - ask anyone involved with audio playback design."

Please read carefully what I wrote. I did not tell you that pre-ringing is not happening. I told you, when it happens, what the exact origin of it is. And that may be counter-intuitive. AFAIK Peter Craven was the
first one to get it, or at least to publish about it.


"ask anyone involved with audio playback design"

Yeah ... I have designed many an anti-alias and anti-imaging filter
for audio (oh, and video) in the past years. Reverse-engineered Meridian's too, for fun.



bring back dynamic range

Page processed in 0.034 seconds.

[ Top Top of Page ] [ Contact Us ] [ Support/Wish List ] [ Copyright © 2010, Audio Asylum, all rights reserved ] [ Take5 ]
[ General ] [ Speakers ] [ Tubes ] [ Vinyl ] [ Digital ] [ Hi-Rez ] [ Video Asylum ] [ Cables ] [ Tweaks/DIY ] [ Music ] [ Films ]