Integrating Phase Measurements in DAC reviews

[melowman]

Some recent posts at gearspace made me write this post.

I would like you to refer to this post from Paul Frindle (if you don't know him, look him up! He is – amongst many things – the engineer behind many iconic SSL analog mixing consoles and the legendary Oxford OXF-R3 digital mixing console). Please read the full post. Paul Frindle claims this:

My conclusion from doing loads of double blind A,B,X listening tests of input/output loop testing was that the preamble of the phase linear filter was entirely inaudible on anything I could listen to, including concocted signals to try and make it as bad as possible.
However as little as a 15deg relative phase lag at 15KHz was audible all the time.. :-(
So no - for a system that was supposed to be transparent under all conditions, we simply could not get away with an apodizing filter, or anything simpler than a full blown phase linear design.

Ok - to go back a full cycle to the original discussion:
Would I prefer and apodizing filter in a converter - over a linear phase one?
Most definitely not - because as a professional I want to hear what really comes out of my whole process when working :-(

In other terms, it would be a useful addition to DAC reviews to integrate phase measurements.

Paul Frindle is an absolute expert, well renewed and highly respected engineer, with decades of experience. I very much trust his experiments, and in the name of "perfect" audio fidelity, I think phase has to be taken into account.
I personally can make the difference between minimum phase and linear phase DA filters (although in AB testing and not ABX testing, but I'm pretty confident about what I'm hearing (I can hear above 20kHz, btw)).

Would this request be integrated for future DAC measurements?

 

[Blumlein 88]

I know who Frindle is, but most of his blind testing was single blind. There is ample investigation indicating people generally don't hear or care about phase shift in the upper frequencies. Now, if you can still hear to 20 khz or beyond, most filters like the apodizing filters also dip the response up there enough to probably be audible. So more than just phase you are hearing slightly reduced response above 10 khz. Amir already shows response of various filters. If something is rolling off slowly or early you know it is also altering phase.

J_J has indicated there are some instances where envelopes of sound at higher frequency can be heard if phase is altered. So it might be a minor concern. I imagine the main difference is most adults don't hear to 20 khz by the time they are 30. You are lucky indeed if you do so take care of that hearing.

A side note: the reason many recording interfaces and ADCs use apodizing filters is because you can have reduced latency which is important in recording musicians who want to hear things without latency that interferes with their timing among themselves.

 

[Blumlein 88]

Notice in this recent review of the SMSL DAC by Amir how the apodizing filter would be audible above 14 khz to someone with your hearing. Whereas unfortunately someone my age my hearing does not go above that 14 khz point. I'd likely not notice.

 

[GXAlan]

If you can only do it in AB testing, but not ABX testing then it becomes very unreliable. A proper blind A-B test is OK, but the ABX is how you can achieve this solo.

One thing that I encourage is to try to measure your own gear at the extremes of sound and then trying to see if phase is a difference. I haven’t found that to be an issue with my gear, but maybe my gear handles phase properly at the expense of noise or distortion?

 

[theREALdotnet]

I’m impressed that Paul Frindle can hear 15kHz well enough to detect a 15 degree phase lag, given that he must be in his 60s?

 

[DVDdoug]

The key phrase is relative phase shift.

You might hear the effects if the left & right channels are out-of-phase. You WILL hear a difference if they are 180 degrees out-of-phase, especially at bass frequencies. (You can do that by reversing the connections to one speaker, or you can do it with Audacity.) But if you invert both channels everything sounds normal again. (i.e. If you reverse the connections to both speakers it will sound normal again.)

At higher frequencies the wavelengths are short and the left & right channels, as well as reflected soundwaves go in-and-out of phase as you move your head or move around the room and the effect is very audible with high frequency test tones. I guess we are used to it and we don't usually notice it with regular program material. In fact, the distance between your ears means that your ears aren't equally distant from both speakers so the sound can be in-phase at one ear and out-of-phase at the other ear.

Most passive crossovers introduce a 90 degree phase shift at the crossover point (a lag in the low-pass and a lead in the high-pass). That puts the two drivers 180 degrees out-of-phase at the crossover frequency. For that reason the connections to the tweeter (or midrange in a 3-way) are usually reversed to put them back in-phase (relative to each other) at the crossover point where the two soundwaves are are combined and "relative" to each other. Beyond the crossover frequency it's out-of-phase compared to the original signal, but since the left & right speakers are presumably identical, the left & right speakers are in phase relative to each other.

 

[sam_adams]

As far as DACs are concerned, any properly designed DAC—with no filter engaged—should have little to no phase shift across the spectrum of audibility (Old Radio Shack DAC, cat no. 1500093, at 96 kHz):

 

[Blumlein 88]

I’m impressed that Paul Frindle can hear 15kHz well enough to detect a 15 degree phase lag, given that he must be in his 60s?

In fairness he was describing work done 30 years ago. He also mentioned in his post if you didn't read it that a 60 khz sample rate would have made it all a non-issue. So apodizing filters were you to use them at 88 khz or 96 khz would pose no issue.

 

[theREALdotnet]

In fairness he was describing work done 30 years ago. He also mentioned in his post if you didn't read it that a 60 khz sample rate would have made it all a non-issue. So apodizing filters were you to use them at 88 khz or 96 khz would pose no issue.

That would explain it.

And yes, to me the greatest benefit of higher sampling rates is getting the filtering artifacts away from the audible range.

 

[melowman]

I know who Frindle is, but most of his blind testing was single blind. There is ample investigation indicating people generally don't hear or care about phase shift in the upper frequencies. Now, if you can still hear to 20 khz or beyond, most filters like the apodizing filters also dip the response up there enough to probably be audible. So more than just phase you are hearing slightly reduced response above 10 khz. Amir already shows response of various filters. If something is rolling off slowly or early you know it is also altering phase.

J_J has indicated there are some instances where envelopes of sound at higher frequency can be heard if phase is altered. So it might be a minor concern. I imagine the main difference is most adults don't hear to 20 khz by the time they are 30. You are lucky indeed if you do so take care of that hearing.

A side note: the reason many recording interfaces and ADCs use apodizing filters is because you can have reduced latency which is important in recording musicians who want to hear things without latency that interferes with their timing among themselves.

Thank you. Actually here I was concerned by phase, not by roll-off.

I can hear the difference between minimum phase and linear phase DA filters, while those filters have the exact same frequency response (I’ve got a RME DAC); so the roll-off doesn’t count in the difference.

If something is rolling off slowly or early you know it is also altering phase.

I’m not sure this is the case. Linear-phase slow roll-off are a thing — if we are talking about roll-offs towards the end of the human hearing region.

Also, I don’t know if you kept on reading the gearspace thread, but Paul mentions that his DAC design in the OXF-R3 console accounted for the eventual analog filter phase shifts, by applying HF phase advance to the digital filter, so that the whole thing stayed absolutely phase linear.

All in all, phase linearity is, in Paul’s words, a criteria for high fidelity. As much as SINAD, etc. Hence my request here.

 

[GXAlan]

All in all, phase linearity is, in Paul’s words, a criteria for high fidelity. As much as SINAD, etc. Hence my request here.

There is likely some truth to phase, but my experience is that requesting more tests will be hard until we know how to make sense of the information. This is where your own experiences comparing good vs bad DACs can help or better stated, transparent vs not.

Indirectly, I have ABX’d DSD vs CD layers of the same disc (see my signature). I had @Serge Smirnoff look at my source files and he was able to compare downsampling the DSD himself versus the CD layer that was on the disc and he ultimate determined that the CD master had significantly higher phase difference than his own mathematical downsampling (the mastering was done differently).

Since I was able to ABX the two, it does show that something was different even when level matched.

Serge was quick to point out that different doesn’t tell us which if the two is the more “correct” or “preferred” one.

 

[melowman]

@GXAlan This is absolutely correct.
With the RME ADI-2 DAC + Focal Clear headphones, my preference is linear-phase, and slow roll-off. Minimum phase and slow roll-off doesn’t work for me, despite having the same frequency response as the former.

But linear-phase sharp roll-off is the least preferred! So it really is a matter of phase and slope.

But beyond subjective preference, I think the philosophy of “Hi-Fi” (in the DAC context) is to put out an analog signal faithful to the digital one. If phase shifts are audible, then having a linear phase outcome is a concern.

 

[Curvature]

There is likely some truth to phase, but my experience is that requesting more tests will be hard until we know how to make sense of the information. This is where your own experiences comparing good vs bad DACs can help or better stated, transparent vs not.

Indirectly, I have ABX’d DSD vs CD layers of the same disc (see my signature). I had @Serge Smirnoff look at my source files and he was able to compare downsampling the DSD himself versus the CD layer that was on the disc and he ultimate determined that the CD master had significantly higher phase difference than his own mathematical downsampling (the mastering was done differently).

Since I was able to ABX the two, it does show that something was different even when level matched.

Serge was quick to point out that different doesn’t tell us which if the two is the more “correct” or “preferred” one.

In that thread about the Japanese flutes you didn't conclusively prove that the files were level matched.