• Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Audio by Van Alstine ABX Comparator Review

My reading is there are several issues with the ABX box, some/any/all of which might affect the result, all of which are undesirable, and some of which required definite action by Josh to prevent it affecting his results.
If I'm reading the APS thread correctly, Josh claims to have taken the appropriate actions for each of these problems, even though he didn't mention some/all of them prior to them being raised...
Could someone summarise where we actually are? I'm getting lost with all the different possibilities...

Amir did this yesterday, generally agreed to be a pretty good summary:

 
Honestly, this thread was IMHO "done" with the first evidence of the "comparator" being inferior to devices it was meant to compare.
So, we are essentially in an endless loop...
Yes,, I moved out of this discussion around no new information. This ABX box is simply not good enough for testing DACs. So discussion about relay clicks makes no sense for me.
 
Honestly, this thread was IMHO "done" with the first evidence of the "comparator" being inferior to devices it was meant to compare.
So, we are essentially in an endless loop...
I don't see it that way. What Chris at AS calls a fishing expedition is catching quite a lot. Some which can possibly explain how Josh got those ABX results, and some that just show problems with the box in general and why it isn't very good for ABX. I appreciate the time and the work, @amirm
 
I don't see it that way. What Chris at AS calls a fishing expedition is catching quite a lot. Some which can possibly explain how Josh got those ABX results, and some that just show problems with the box in general and why it isn't very good for ABX. I appreciate the time and the work, @amirm
I also admire the patience of @amirm - I would not have 10% of it.
 
Effect of impedance in series with opamp buffer input on frequency response. Opamp buffer has 100k//330pF input impedance. The ripple comes from Topping D10s DAC as a signal source.

AudioBuffer_Rs_effect_onFR.png

Plots 1, 2 and 5 do overlap.


Effect on distortion
AudioBuffer_Rs_effect.png


Effect on distortion can be neglected in this case.
 
Effect of impedance in series with opamp buffer input on frequency response. Opamp buffer has 100k//330pF input impedance. The ripple comes from Topping D10s DAC as a signal source.

View attachment 563415
Plots 1, 2 and 5 do overlap.


Effect on distortion
View attachment 563416

Effect on distortion can be neglected in this case.
Do we have any idea on modeling HPA4 input characteristics? @John_Siau ?
 
Last edited:
The RME DAC's delay probably also comes from its DC filter if enabled apart from the usual route.
 
So this is indeed the situation. I wonder if @Josh83 would try one of the ABX DAC comparisons that @pma made in the other thread, one that does not have passband ripple differences (echoes). That might give some indication if this is the issue, although it wouldn't be perfectly conclusive either way.

Although with something like just two sample difference the echo and main signal are extremely close together, and the passband ripple will have very large frequency period. And I think the ripple does not even exist if there is not both pre echo and post echo happening. (maybe time to call for help from @J.J. !)
So I think it would be more a rolloff issue or a wide dip or bump in passband flatness if there is ripple, not so much an echo in that case (mathematically the same as a very short delay echo). But if relative delay is longer then it could be more like an echo.

another edit: here is the relationship between passbnad ripple and echoes for those interested. You see that pre echo and post echo are created by passband ripple. But if one signal has pre echo and the other has post echo, it seems they both would have the same amplitude spectrum so then this issue is different than passband ripple caused by a filter.

H(f)=1+2αcos⁡(2πfT) [this is the ripple in frequency domain]

which corresponds in the time domain to

h(t)=δ(t)+αδ(t−T)+αδ(t+T)
The numbers for group delay I listed only account for the delay from the reconstruction filter. The total DAC latency encompasses more than that. So the differences between the DAC pairs could end up smaller or larger for each pair. The additional delays added by other components would mean that the highest potential time differences are larger than the numbers from my table, though.

Also: While each individual channel of the comparator only sees one echo delayed by of a couple of samples (60 µs is about three samples at 44.1 kHz), the channels both end up with an echo in the reversed direction. So the comparator output would be
Code:
Channel A: A-input + B-input at -40 dB (+60 µs time shifted)
vs
Code:
Channel B: B-input + A-input at -40 dB (-60 µs time shifted)
The comparison of channel A to channel B would end up including both echos (pre- and post), but only one is active in each channel. I simulated this with a song and when you compare the results, the +3 samples vs -3 samples delay looks like this (mind the log frequency axis scaling):

Delta of Spectra_+3S vs. -3S.jpg

PK Metric_+3S vs. -3S.jpg


Clearly, some ripple is there and the PK metric is not great. In contrast, the original vs. -3 samples delay would look like this - but this is not what you would listen to at the comparator output:

Delta of Spectra_ORIG vs. -3S.jpg

PK Metric_ORIG vs. -3S.jpg


Instead of one ripple, this combination only shows a dip around 10 kHz. The PK Metric is actually slightly worse.


EDIT: "+3 samples vs -3 samples" means the graphs compare the [original song + a 3 samples advanced copy at -40 dB] with the [original song + a 3 samples delayed copy at -40 dB]. "original vs. -3 samples" means they compare the [original song] with the [original song + a 3 samples delayed copy at -40 dB]. My example used 44.1 kHz material, so 3 samples equals approx. 68 µs. For 96 kHz material like that used by Josh, 60 µs would be closer to 6 samples.

EDIT 2: Found a problem with one file & plot and regenerated them all.
 
Last edited:
The numbers for group delay I listed only account for the delay from the reconstruction filter. The total DAC latency encompasses more than that. So the differences between the DAC pairs could end up smaller or larger for each pair. The additional delays added by other components would mean that the highest potential time differences are larger than the numbers from my table, though.

Also: While each individual channel of the comparator only sees one echo delayed by of a couple of samples (60 µs is about three samples at 44.1 kHz), the channels both end up with an echo in the reversed direction. So the comparator output would be
Code:
Channel A: A-input + B-input at -40 dB (+60 µs time shifted)
vs
Code:
Channel B: B-input + A-input at -40 dB (-60 µs time shifted)
The comparison of channel A to channel B would end up including both echos (pre- and post), but only one is active in each channel. I simulated this with a song and when you compare the results, the +3 samples vs -3 samples delay looks like this (mind the log frequency axis scaling):

View attachment 563423
View attachment 563424

Clearly, some ripple is there and the PK metric is not great. In contrast, the original vs. -3 samples delay would look like this - but this is not what you would listen to at the comparator output:

View attachment 563425
View attachment 563426

Ironically, the delta spectrum looks "better" disregarding the sub-bass region while the PK metric is quite bad now :eek:


EDIT: "+3 samples vs -3 samples" means the graphs compare the [original song + a 3 samples advanced copy at -40 dB] with the [original song + a 3 samples delayed copy at -40 dB]. "original vs. -3 samples" means they compare the [original song] with the [original song + a 3 samples delayed copy at -40 dB].
Is your delta of spectra plot A vs B or one of these vs original? You will definitely get ripple vs original but spectra should be the same A vs B (I did similar with DW and that is what I saw). Think about it: in each case with the crosstalk the total spectrum is the original spectrum plus a -40dB copy of the original spectrum.

Here is what I get. This is a 2 sample delay with a 96K file.
Vs original:
1791551861071.png


A vs B
1791551943210.png
 
Last edited:
Is your delta of spectra plot A vs B or one of these vs original? You will definitely get ripple vs original but spectra should be the same A vs B (I did similar with DW and that is what I saw). Think about it: in each case with the crosstalk the total spectrum is the original spectrum plus a -40dB copy of the original spectrum.
Both are shown, as explained. And the -40 dB copy will be delayed in one and advanced in the other channel (relative to the main signal). That's the important difference.

EDIT: I found a problem with the original vs -3 samples files and regenerated the audio and plots. See updated post above :)
 
Last edited:
I don't see it that way. What Chris at AS calls a fishing expedition is catching quite a lot. Some which can possibly explain how Josh got those ABX results, and some that just show problems with the box in general and why it isn't very good for ABX. I appreciate the time and the work, @amirm

I hope those who are familiar with my posts here would agree that I do my best to avoid ad hominem criticisms, so I say this in a spirit of honesty (and exasperation, to a degree), not malice or snark: In my experience and opinion, Chris Connaker is a total dilettante when it comes to these kinds of technical issues in audio. Whenever he comments on them, he’s out of his depth and his tone of dismissive certainty is not backed up by anything approaching sufficient knowledge or evidence - and I’m not even talking about professional credentialed knowledge. Even on the reduced scale of knowledgeable amateurs (like me!), he doesn’t know what he’s talking about, and in my perception it’s abundantly clear that his desire to know is quite limited.
 
Last edited:
Had a bit more time to investigate the phase issue. Here is the outcome of comparing 3 DACs:
View attachment 563390


Comparing to Topping D70s this time against a (newly released) Loxjie D30 shows what looks like random phase differential. In reality there is now a 18 millisecond delay (1000 times longer). The sweep doesn't have enough points to capture it so shows that chopped up output. To see that more clearly, here is the unwrapped phase of the same, this time measured with a CHIRP signal:
View attachment 563391
Amir, the original d70 vs Loxjie D30 phase seemed to have a peak just below 500 Hz. The phase seems to be increasing before it, and decreasing after it. Shouldn't the unwrapping have preserved this?
 
With all respect, this is absolutely unimportant when doing a DBT listening test. We should not go ad absurdum.
Maybe with a very small delay, but with a longer delay difference it can be very similar to the passband ripple that you were able to notice in your DAC recordings ABX.
 
Both are shown, as explained. And the -40 dB copy will be delayed in one and advanced in the other channel (relative to the main signal). That's the important difference.

EDIT: I found a problem with the original vs -3 samples files and regenerated the audio and plots. See updated post above :)
I am still not getting why your A vs B shows ripple and mine does not. Maybe I made an error in my files but I don't see it.
 
I am still not getting why your A vs B shows ripple and mine does not. Maybe I made an error in my files but I don't see it.
Just to be sure: Do both of your "channels" (A & B) contain one echo each?
 
Just to be sure: Do both of your "channels" (A & B) contain one echo each?
Yes.
A is original + (-40dB original shifted backward two samples)
B is original + (-40dB original shifted forward two samples)
 
Yes.
A is original + (-40dB original shifted backward two samples)
B is original + (-40dB original shifted forward two samples)
Not sure what the problem is. I'm pretty sure I fixed the initial mistake in my files. For 60 µs, which is what I aimed for, you should use 6 samples @ 96 kHz.

Logically, it makes sense to me that the ripple would emerge: If you add two time-shifted copies @-40 dB to one track, the difference to the original are the two echoes. If you add one time-shifted copy @-40 dB to each original track, the nominal difference between them is the same: The two echoes. But one will have its sign flipped.
 
Back
Top Bottom