ABX DAC comparison 100% Accuracy... HELP!

[Blumlein 88]

Topping D90 vs D10B

 

[davmol]

I've had a good listen in Lacinato and to my ear the Sugden is marginally louder than the Topping D10 & D90, which are the same but marginally louder again than the Denefrips and SMSL, which are the same. I couldn't hear anything tonally different between them but the volume, to me, was audibly different in the groups I mentioned and particularly noticeable when you get to the higher frequencies on the Cycles track.

The track length differences are a bit annoying as it makes it hard to do seamless switching between A and B in Lacinato, as there is always a miniscule jump forward or back depending on which DACs are being compared.

 

[HiFiJosh]

OP here...

Wow thank you all for this investigation! The internet can be an amazing place .

To clarify a few things about the recordings, the 'in-room' recordings were used for video and he used the Roland ones for the download. The reason they are V2 is in the pinned comment in the original video, the original recordings were slightly volume mismatched.

The findings of Sokel in post # 21 and Blumlein 88 in post #25 are VERY odd. The maker of the video is quite responsive to comments and is aware of ASR, hopefully I can get him to chime into the discussion.

However, per Blumlein in post #41 and #21, it appears that the two Topping files are comparable, and I was still able to tell which was which within a few seconds of hearing.

Just to clarify, I've only listened to the 'I've Got You Under My Skin' tracks and I'm not doing any funny business like checking the track lengths or anything. Has anyone else tried doing quick A/B comparisons between the tracks? An example of what I'm hearing is the singer's vibrato at around 10 seconds. If you listen to the D90, you can hear a relatively even sustain and decay whereas on the D10, it doesn't pick up any of the nuances of it and abruptly fades out.

 

[SIY]

However, per Blumlein in post #41 and #21, it appears that the two Topping files are comparable, and I was still able to tell which was which within a few seconds of hearing.

Can you post your ABX logs? DeltaWave is a good way to do it so that the timing differences aren't cues.

 

[Bwaslo]

Had some spare time today, so I ran some DiffMaker runs on the files. Hadn't had it up and running in years.
But, I have to say, some things don't make sense, at least in the stats report it gives at the end, so take this with a grain of salt. I wish I still had the source code for this to see what it's doing but all I have anymore is from an earlier version.

First, the "stats". These were run on the first 30 seconds of ..."Under My Skin", comparing to the Denefrips-Pontus-15 (picked arbitrarily).

SMSL/Dene parameters: 106.8usec, 0.000dB (L), -0.001dB (R). rate adj=7.594 ppm. Corr Depth: 22.7 dB (L), 22.8 dB (R)
Sugden/Dene parameters: -121.2nsec, 0.000dB (L), 0.000dB (R). rate adj=-0.0237 ppm. Corr Depth: 24.8 dB (L), 25.1 dB (R)
D10/Dene (INV)parameters: 203.3msec, 0.181dB (L), 0.181dB (R). rate adj=-30.3653 ppm. Corr Depth: 32.3 dB (L), 29.7 dB (R)
D90/Dene (INV)parameters: -60.33msec, -0.025dB (L), -0.087dB (R). rate adj=6.1361 ppm. Corr Depth: 34.1 dB (L), 30.7 dB (R)

The "Corr Depth" doesn't seem to make sense, as the D10 compared against Dene (or really against any of them) sounds not cancelled at all, though it has 2nd best "Corr Depth" of the group. The D10 is reported near 0.2dB higher level (the highest) and the furthest sample rate from the other four DACs (in the vicinity of 30ppm low). I suspect the big sample rate difference is the main issue with cancellation not going well.

The most cancellation (by my ear and by the "Corr Depth") was the SMSL against the SUGDEN. --

SMSL/SUGDEN parameters: -194.2msec, -0.081dB (L), -0.025dB (R). rate adj=0.0102 ppm.Corr Depth: 41.2 dB (L), 45.1 dB (R)
SUGDEN/SMSL Parameters: 194.2msec, 0.082dB (L), 0.027dB (R). rate adj=-0.0322 ppm.Corr Depth: 40.4 dB (L), 43.6 dB (R)

That was also the pair with the least sample rate difference.
The Sugden/Dene pair, which were also close in sample rate, had decent cancellation (though with lousy "corr")!. Difference file was audible but quite a bit down from the uncancelled.

Which leads me to believe that the sample rate converter I was using in DiffMaker isn't so great.

Anyway, in case that adds anything to the discussion....

 

[Blumlein 88]

Had some spare time today, so I ran some DiffMaker runs on the files. Hadn't had it up and running in years.
But, I have to say, some things don't make sense, at least in the stats report it gives at the end, so take this with a grain of salt. I wish I still had the source code for this to see what it's doing but all I have anymore is from an earlier version.

First, the "stats". These were run on the first 30 seconds of ..."Under My Skin", comparing to the Denefrips-Pontus-15 (picked arbitrarily).

SMSL/Dene parameters: 106.8usec, 0.000dB (L), -0.001dB (R). rate adj=7.594 ppm. Corr Depth: 22.7 dB (L), 22.8 dB (R)
Sugden/Dene parameters: -121.2nsec, 0.000dB (L), 0.000dB (R). rate adj=-0.0237 ppm. Corr Depth: 24.8 dB (L), 25.1 dB (R)
D10/Dene (INV)parameters: 203.3msec, 0.181dB (L), 0.181dB (R). rate adj=-30.3653 ppm. Corr Depth: 32.3 dB (L), 29.7 dB (R)
D90/Dene (INV)parameters: -60.33msec, -0.025dB (L), -0.087dB (R). rate adj=6.1361 ppm. Corr Depth: 34.1 dB (L), 30.7 dB (R)

The "Corr Depth" doesn't seem to make sense, as the D10 compared against Dene (or really against any of them) sounds not cancelled at all, though it has 2nd best "Corr Depth" of the group. The D10 is reported near 0.2dB higher level (the highest) and the furthest sample rate from the other four DACs (in the vicinity of 30ppm low). I suspect the big sample rate difference is the main issue with cancellation not going well.

The most cancellation (by my ear and by the "Corr Depth") was the SMSL against the SUGDEN. --

SMSL/SUGDEN parameters: -194.2msec, -0.081dB (L), -0.025dB (R). rate adj=0.0102 ppm.Corr Depth: 41.2 dB (L), 45.1 dB (R)
SUGDEN/SMSL Parameters: 194.2msec, 0.082dB (L), 0.027dB (R). rate adj=-0.0322 ppm.Corr Depth: 40.4 dB (L), 43.6 dB (R)

That was also the pair with the least sample rate difference.
The Sugden/Dene pair, which were also close in sample rate, had decent cancellation (though with lousy "corr")!. Difference file was audible but quite a bit down from the uncancelled.

Which leads me to believe that the sample rate converter I was using in DiffMaker isn't so great.

Anyway, in case that adds anything to the discussion....

Go get your free copy of Deltawave. I guess you would say Diffmaker inspired Paul to make Deltawave. It is Diffmaker on steroids only it doesn't crash like it. Started by user pkane.

DeltaWave Audio Null Comparator | DeltaWave documentation

deltaw.org

He also has created other useful things like Multitone and Distort.

 

[Blumlein 88]

Had some spare time today, so I ran some DiffMaker runs on the files. Hadn't had it up and running in years.
But, I have to say, some things don't make sense, at least in the stats report it gives at the end, so take this with a grain of salt. I wish I still had the source code for this to see what it's doing but all I have anymore is from an earlier version.

First, the "stats". These were run on the first 30 seconds of ..."Under My Skin", comparing to the Denefrips-Pontus-15 (picked arbitrarily).

SMSL/Dene parameters: 106.8usec, 0.000dB (L), -0.001dB (R). rate adj=7.594 ppm. Corr Depth: 22.7 dB (L), 22.8 dB (R)
Sugden/Dene parameters: -121.2nsec, 0.000dB (L), 0.000dB (R). rate adj=-0.0237 ppm. Corr Depth: 24.8 dB (L), 25.1 dB (R)
D10/Dene (INV)parameters: 203.3msec, 0.181dB (L), 0.181dB (R). rate adj=-30.3653 ppm. Corr Depth: 32.3 dB (L), 29.7 dB (R)
D90/Dene (INV)parameters: -60.33msec, -0.025dB (L), -0.087dB (R). rate adj=6.1361 ppm. Corr Depth: 34.1 dB (L), 30.7 dB (R)

The "Corr Depth" doesn't seem to make sense, as the D10 compared against Dene (or really against any of them) sounds not cancelled at all, though it has 2nd best "Corr Depth" of the group. The D10 is reported near 0.2dB higher level (the highest) and the furthest sample rate from the other four DACs (in the vicinity of 30ppm low). I suspect the big sample rate difference is the main issue with cancellation not going well.

The most cancellation (by my ear and by the "Corr Depth") was the SMSL against the SUGDEN. --

SMSL/SUGDEN parameters: -194.2msec, -0.081dB (L), -0.025dB (R). rate adj=0.0102 ppm.Corr Depth: 41.2 dB (L), 45.1 dB (R)
SUGDEN/SMSL Parameters: 194.2msec, 0.082dB (L), 0.027dB (R). rate adj=-0.0322 ppm.Corr Depth: 40.4 dB (L), 43.6 dB (R)

That was also the pair with the least sample rate difference.
The Sugden/Dene pair, which were also close in sample rate, had decent cancellation (though with lousy "corr")!. Difference file was audible but quite a bit down from the uncancelled.

Which leads me to believe that the sample rate converter I was using in DiffMaker isn't so great.

Anyway, in case that adds anything to the discussion....

I didn't notice at first you were Bill Waslo who wrote Diffmaker. Still give a look at what Paul has done with Deltawave.

 

[Blumlein 88]

OP here...

Wow thank you all for this investigation! The internet can be an amazing place .

To clarify a few things about the recordings, the 'in-room' recordings were used for video and he used the Roland ones for the download. The reason they are V2 is in the pinned comment in the original video, the original recordings were slightly volume mismatched.

The findings of Sokel in post # 21 and Blumlein 88 in post #25 are VERY odd. The maker of the video is quite responsive to comments and is aware of ASR, hopefully I can get him to chime into the discussion.

However, per Blumlein in post #41 and #21, it appears that the two Topping files are comparable, and I was still able to tell which was which within a few seconds of hearing.

Just to clarify, I've only listened to the 'I've Got You Under My Skin' tracks and I'm not doing any funny business like checking the track lengths or anything. Has anyone else tried doing quick A/B comparisons between the tracks? An example of what I'm hearing is the singer's vibrato at around 10 seconds. If you listen to the D90, you can hear a relatively even sustain and decay whereas on the D10, it doesn't pick up any of the nuances of it and abruptly fades out.

Okay, So I tried the comparator under Play in Deltawave. The two Topping tracks are offset at the beginning by .26 seconds which by itself would let you pick up on which is which without even realizing why. With Deltawave it aligns all of that and level matches even better than the .11 db match. I keyed on the sound from about 8-12 seconds. I managed 9 of 10 twice using AKG 371 phones out of the headphone jack of my Asus gaming computer. It is as you described it. I would say the D10 sounds hard and lacking in detail or air as the wavering voice fades out. So this is very interesting. I'll look at why a bit more when I get a chance.

Paul could you see if you know why this might be from the results. I btw, trimmed off 10 seconds of both Topping files in the "I've got you" track on the end. They fade out differently from each other if you don't. Sokel maybe you could find something too. @Sokel
@pkane

The null is -60 db and a -75 db PK metric.

The only unusual thing is the phase.

 

[Blumlein 88]

So some further tests. I let Deltawave correct the Phase EQ, not the Level EQ. I could no longer hear a difference. I scored 6 of 10 and 5 of 10. Also prior to this, listening to the Difference track with some 40 db gain it sounds almost exactly like the original with little noise. As if it were a plain slight level difference causing this. When I corrected for phase using 40 db gain all that was left was some noise and some hints of the music though not much music. The null improved to 72.6 db while the Pk metric changed less than a db. If I use Level and Phase EQ the PK metric is up to 83 and the null barely improves to 72.7 so it is a phase thing going on.

I would say whatever is causing the phase oddities is the reason for one sounding different. I have a D10B and it doesn't show any issues with phase vs a Babyface Pro, Antelope Audio, or March audio DAC. So I don't know what happened in the recording or why this would happen. Anyway good catch by the OP @HiFiJosh

 

[kemmler3D]

So some further tests. I let Deltawave correct the Phase EQ, not the Level EQ. I could no longer hear a difference. I scored 6 of 10 and 5 of 10. Also prior to this, listening to the Difference track with some 40 db gain it sounds almost exactly like the original with little noise. As if it were a plain slight level difference causing this. When I corrected for phase using 40 db gain all that was left was some noise and some hints of the music though not much music. The null improved to 72.6 db while the Pk metric changed less than a db. If I use Level and Phase EQ the PK metric is up to 83 and the null barely improves to 72.7 so it is a phase thing going on.

I would say whatever is causing the phase oddities is the reason for one sounding different. I have a D10B and it doesn't show any issues with phase vs a Babyface Pro, Antelope Audio, or March audio DAC. So I don't know what happened in the recording or why this would happen. Anyway good catch by the OP @HiFiJosh

Very interesting, you don't normally come across phase distortion that's easily audible. I wonder where things went wrong - surely the D10 isn't actually normally that distorted, so maybe something happened in the recording or conversion process?

 

[pkane]

Okay, So I tried the comparator under Play in Deltawave. The two Topping tracks are offset at the beginning by .26 seconds which by itself would let you pick up on which is which without even realizing why. With Deltawave it aligns all of that and level matches even better than the .11 db match. I keyed on the sound from about 8-12 seconds. I managed 9 of 10 twice using AKG 371 phones out of the headphone jack of my Asus gaming computer. It is as you described it. I would say the D10 sounds hard and lacking in detail or air as the wavering voice fades out. So this is very interesting. I'll look at why a bit more when I get a chance.

Paul could you see if you know why this might be from the results. I btw, trimmed off 10 seconds of both Topping files in the "I've got you" track on the end. They fade out differently from each other if you don't. Sokel maybe you could find something too. @Sokel
@pkane

The null is -60 db and a -75 db PK metric.

The only unusual thing is the phase.
View attachment 405994

The "I've got you" Topping tracks have about 0.1dB difference in level, which I assume isn't very significant. There's a fairly large clock speed difference of about 40ppm.

What's likely more significant is that there are a few fairly large excursions in the delta waveform:

At each of those peaks, I find a large spike in 20Hz-150Hz area. For example, at 1:09:

And at 1:56:

It would appear those areas have more energy in the lower frequencies. Possibly the effect of the DC filter in one or both of these DACs -- don't know if DC filter can be turned off for comparison. Here's the delta waveform when both waveforms are LP-filtered at 200Hz:

 

[pkane]

The "I've got you" Topping tracks have about 0.1dB difference in level, which I assume isn't very significant. There's a fairly large clock speed difference of about 40ppm.

What's likely more significant is that there are a few fairly large excursions in the delta waveform:

View attachment 406062


At each of those peaks, I find a large spike in 20Hz-150Hz area. For example, at 1:09:

View attachment 406063

And at 1:56:
View attachment 406064

It would appear those areas have more energy in the lower frequencies. Possibly the effect of the DC filter in one or both of these DACs -- don't know if DC filter can be turned off for comparison. Here's the delta waveform when both waveforms are LP-filtered at 200Hz:

View attachment 406067

This difference in lower frequencies is also fairly obvious in the delta waveform frequency analysis, with peak difference landing on around 60-70Hz (this is for the average over the whole recording):

 

[Blumlein 88]

The "I've got you" Topping tracks have about 0.1dB difference in level, which I assume isn't very significant. There's a fairly large clock speed difference of about 40ppm.

What's likely more significant is that there are a few fairly large excursions in the delta waveform:

View attachment 406062


At each of those peaks, I find a large spike in 20Hz-150Hz area. For example, at 1:09:

View attachment 406063

And at 1:56:
View attachment 406064

It would appear those areas have more energy in the lower frequencies. Possibly the effect of the DC filter in one or both of these DACs -- don't know if DC filter can be turned off for comparison. Here's the delta waveform when both waveforms are LP-filtered at 200Hz:

View attachment 406067

If all I do is allow Phase EQ those peaks go away. What would cause something like that? This is without filtering.

 

[Sokel]

If all I do is allow Phase EQ those peaks go away. What would cause something like that? This is without filtering.

View attachment 406071

View attachment 406072
View attachment 406073


View attachment 406074

I would like to know as well,specially if it's the ADC or the DAC doing this (or any interaction)
A DC filter as pkane said seems reasonable as like any filter can introduce phase alteration.

Thing is that it goes way high to blame only that.The repeated excursions is something to look as well.

 

[Blumlein 88]

I would like to know as well,specially if it's the ADC or the DAC doing this (or any interaction)
A DC filter as pkane said seems reasonable as like any filter can introduce phase alteration.

Thing is that it goes way high to blame only that.The repeated excursions is something to look as well.

I suspect it is the recording ADC. All the files compared in all combinations with each other show odd results in the phase plots. Even when two of the files show 0 ppm clock drift vs the SMSL. All clean up quite a lot if you correct for Phase, but not for Level EQ except for the Sugden which has additional issues that would appear to be variable FR with level or variable FR over time. I don't know if the Roland R-88 was malfunctioning or has some built in filter which interacts with other factors. I think the Roland is the source of the problem in these files at least.

 

[pkane]

If all I do is allow Phase EQ those peaks go away. What would cause something like that? This is without filtering.

View attachment 406071

View attachment 406072
View attachment 406073


View attachment 406074

I'd investigate the DC filter and its effect on phase. RME ADI-2 Pro, for exampe, lets one chose the DC filter or turn it off completely. To borrow a subjective descriptor, the effect "is not even subtle" At least on the measured differences. Depending on the transducer being used and its ability to handle lower frequencies such differences can become audible.

 

[Blumlein 88]

According to the Roland Manual it has a Low Cut filter that can be set as OFF, or 60 hz, 120 hz or 240 hz. Each input can be set individually. Perhaps the 60 hz filter was on. It was mentioned in the video each device was connected to its own input set of XLRs.

It also mentioned that an AGC is on the microphone inputs at all times. It is not clear if you are using the inputs at line level if this is the case or not. If so the R-88 is not fit for this purpose.

 

[Blumlein 88]

I think Paul had it right regarding it being the DC filter. The ADC of the Roland probably uses a blocking cap. The differences between DACs may have been some had the 60 hz low cut in and others did not. Or even that the built in DC cap interacted with different output impedances of the various devices to cause a bit of phase difference between them. Plus we know some variance in the upper end due to various anti-imaging filters.

Here is a file original vs itself if you have a 60 hz 1st order low cut filter involved. Not too dissimilar to what we are seeing.

 

[MaxwellsEq]

We have had claims that the RME DC protection filter "robs music of life" (paraphrase). Perhaps some people can on a ABX sense something about the phase changes introduced by DC blocking and impedance interactions. There's also the possibility that the Roland needs a service.

Fix your RME ADI 2 - FS DAC sound.

EDIT: I feel like a should buy some flowers to fellow members of audiosciene forum. Based on their response, whole problem was with Digital DC Protection which was on (filter) by DEFAULT! I dont know why RME though that its good idea to turn on such function to mess with your audio by default...

audiosciencereview.com

 

[Sokel]

I think Paul had it right regarding it being the DC filter. The ADC of the Roland probably uses a blocking cap. The differences between DACs may have been some had the 60 hz low cut in and others did not. Or even that the built in DC cap interacted with different output impedances of the various devices to cause a bit of phase difference between them. Plus we know some variance in the upper end due to various anti-imaging filters.

Here is a file original vs itself if you have a 60 hz 1st order low cut filter involved. Not too dissimilar to what we are seeing.
View attachment 406272

So comes the question if the ADC DC blocking is the reason:
Can we expect the same interaction with an amp with DC blocking (through caps,etc) vs another open one?