Review and Measurements of the PS Audio Stellar Gain Cell DAC

[amirm]

He goes on to say:

Sigh. Seems like he doesn't understand what jitter is. Likely thinks if a device is not a DAC, it can't have jitter. I suggest reading any text on jitter and failing that, Julian Dunn's great AES paper: JITTER AND DIGITAL AUDIO PERFORMANCE MEASUREMENTS

Jitter is a modulation in time of a signal. Mathematically then it would be v'(t) = v(t-delta(t))
Let's assume v(t) is a cosine wave. And delta(t) is our jitter signal and it likewise is a cosine wave. Our equation then becomes:

Looks complicated but it is just the simple equation we started with and simply substituting and expanding.

For small values of jitter we are dealing with we can use this approximation:

And:

Plugging these two into equation (4) we get this:

In plain English, we have our original cosine wave (Acos(wt)). All jitter did was add two sideband cosine waves, one at sum of our signal and that of jitter. And the other, at the difference. We can see this plainly in an example in one of my articles I wrote for WSR Magazine: https://www.audiosciencereview.com/forum/index.php?threads/audibility-of-small-distortions.67/

We took a 10 Khz tone and added sinusoidal jitter at 2.5 kHz. Results are the two sidebands on each side of our signal.

Even if you got lost in all of this , that is OK. The key here is that nothing I have shown so far has anything to do with digital audio! All math was analog in nature. Indeed, the "jitter" shown here is exactly the same as AM modulation used in every AM radio!

Now, change the jitter source to be noise rather than sinusoidal and it causes broadening of our signal tone proportional to PDS of the noise:

That brings us to your gain cell. Your DC input that you are using in Gilbert Cell is a modulator of the audio signal. If it is anything but DC, it will modulate the main signal just like jitter would in a DAC.

In your case, your "DC" signal fed to the Gilbert Cell is polluted with power supply and random noise. The former causes spikes on each side of our tone. The latter causes broadening of the shoulders of our tone. Here it is again:

See the similarity in the way your pre-amp has broadened the bottom of the 1 khz tone and what I showed in my graph above with random jitter. The random noise could be created by the front panel microprocessor for example. Now you see why I call it "jitter like" distortion.

Because you don't measure your designs, and don't understand the fundamentals of the circuit topologies you are using, you are confused by my measurements and terminology. I suggest using this opportunity to learn by instrumenting the signal going into your Gilbert cell. I am pretty sure you will find the noise and power supply spikes riding there. Clean that up and these measurements will improve and will not do anything to deter from the "sound" you think this gain stage has. Lest you want to tell us now all of a sudden that "jitter is good!"

Conclusion
Math, audio science and engineering plus measurements enable us to understand behavior of audio gear. Without it, you are shooting in the dark, producing suboptimal designs. Without scrutiny like we are providing here, you would get away with it much like a restaurant can by not washing their dishes well. But not anymore. Please spend less time being flippant online and more time building better products. Heaven knows I can do other things than being your test and verification shop.

 

[amirm]

He also posts:

That is very simplistic view of THD. The spectrum of THD tells us if the harmonics are audible or not, not a single number like that. This is why I show the FFT. This is why I mention in jitter spectrums if they are audible.

But let's put that aside for now. If the waiter spits in your soup on the way out to your table, it likely will taste the same. I don't know about you, but I don't want anyone to spit in my soup, much less when I am going to a high-end restaurant!

That is the business you are in Paul: building high-end audio products. This means impeccable attention to detail. Engineering away distortions well below what anyone could hear.

Some of you have completely screwed up this business to the detriment of us the customers. WIth faulty notions of what sounds good, combined with nary a proper test and analysis of your product performance, you produce equipment that would be unacceptable in mass consumer products let alone pass for "high-end."

For you to say, "you don't hear this anyway so let me dish out all the "poo" I can" in our signal chain is just irresponsible. There are $10 audio dongles that run rings around the performance of your amplifier for heaven's sake. How illogical can you be to tell us that we should not care?

So no, don't insult us please. Don't say we should not care that you distort our beautiful music so that you can have a talking point about your pre-amp over your competitor. We want excellence in engineering and we are going to demand it, damn it!

 

[amirm]

And there is more:

If you have those measurements, just post them Paul. It can be for any unit. Post an FFT of 1 kHz tone with 32k points. We can start with that and go from there.

For now, we both know you did not run this test. Because if you had, and saw all the problems and still released the product, there would be even more riots in the streets.

By the way, did you listen to each piece before release? You know, after 800 hour of break in to make sure the unit is manufactured correctly? Can't be that you rely on measurements just for that, can it? It is kind of confusing what you say...

 

[BillG]

I may actually set this up for myself when I get home just as experiment.

Yeah, it actually works just fine: I've got my tablet set up as a remote controllable local renderer within Bubble, and it shows up as such on my phone, which is also running Bubble. I've taken the headphone output from the tablet, connected it to my amp, and am streaming away from my server currently...

My tablet's configuration

A screenshot of another streaming application on my phone referencing my tablet as a playback device.

 

[StevenEleven]

So if I am reading correctly he says that the aspects of his equipment that do not measure particularly well are not audible and are not a big deal, but that if he tried to make the equipment measure better, the gear would sound like "dog poo,” as does the better-measuring gear of some of his competitors.

To state the obvious, if, as he claims, the aspects of his gear that do not measure well (as compared to competitors) still perform well enough so as to be below audible thresholds, then making the gear measure better would not make it sound worse, or any different at all.

To paraphrase Uptain Sinclair, perhaps what appears to be a lack of understanding on his part is due to the fact that his livelihood depends on it.

 

[restorer-john]

I didn't say that it was not a proper listening test, I said that it was done with a dubious test file (single sine tone, level at -120dB). I'm also not claiming this device is transparent.

What I proposed is that you do a DBT with normal music file played at normal listening level.

It's a joke to take a -120dB signal and amplify it to audible and then use that for ABX testing. Not a remotely meaningful, clever, or useful test. I sure hope that wasn't done in this case.

 

[Frank Dernie]

It's a joke to take a -120dB signal and amplify it to audible and then use that for ABX testing. Not a remotely meaningful, clever, or useful test. I sure hope that wasn't done in this case.

Quite so.
If I have a -120dB tone and have to amplify it by 60dB to hear it it is no longer a -120dB tone - it is -60dB

 

[pozz]

@amirm Do you still have the unit to measure the headphone output?

 

[scott wurcer]

As to the first comment, let's cut through this: I will come to your place at my expense. Bring one of those low distortion DACs you say sound like "dog poo." If you can determine that against this product in a blind test, I will give $10,000 to your charity of choice.

Should be easy Paul. You come out a hero and give money to a needed charity to boot.

Since the product in question is probably audibly "flawed" this offer could be dangerous to make if in fact your intent is to compare the PS DAC to one that measures well. IMO Randi's cable challenge was poorly managed since he left the door open for (IIRC Mr. Fremer) to choose a pathological cable/speaker combo.

Paul should take a lesson in politeness from Nelson Pass who offers some designs that don't have conventionally low distortion numbers as "entertainment" without the insults to the other side.

 

[ahofer]

I do think it’s a safe bet that everyone can tell the difference between a Topping or Benchmark and dog poo. What would the dog poo measurements look like?

 

[Soniclife]

I will come to your place at my expense. Bring one of those low distortion DACs you say sound like "dog poo." If you can determine that against this product in a blind test, I will give $10,000 to your charity of choice.

Have you done your own listening test yet?

 

[pozz]

It kind of makes sense. If you're interested in audible differences you're going to create designs with characteristic high-level artefacts. No point in messing around with -100dB spurs.

 

[graz_lag]

I do think it’s a safe bet that everyone can tell the difference between a Topping or Benchmark and dog poo. What would the dog poo measurements look like?

It is simply scandalous that an entrepreneur, who claims to be a reference in his industry, uses that noun to define the SQ of electronics that have been built from top notch design & engineering.

It seems the horrible measurements associated to his DACs make him a bit nervous.

 

[amirm]

Have you done your own listening test yet?

No, I just don't have time. The unit is packed ready to go to its owner....

 

[amirm]

Quite so.
If I have a -120dB tone and have to amplify it by 60dB to hear it it is no longer a -120dB tone - it is -60dB

There was no other amplification in the path. Audio captured using AP, and simply played on my Topping DX3 Pro.

 

[amirm]

It's a joke to take a -120dB signal and amplify it to audible and then use that for ABX testing. Not a remotely meaningful, clever, or useful test. I sure hope that wasn't done in this case.

Maybe a joke to you but I took the time to capture the tones and perform the ABX test and post the results. Not seeing any of you lifting a finger to advance our understanding of threshold of hearing.

And as I noted, there was no extra amplification. So don't go accusing me of corruption here without facts.

The results may be a surprise to you but wouldn't be to Fletcher-Munson after their 1933 listening test research to determine threshold of hearing:

You see the bottom set of curves where it says "threshold?" That is the lowest level tone we can hear. In the case of 1 kHz it is at 0 dB SPL. In the case of 3 to 4 kHz, it is actually -10 dBSPL (harmonic distortion of 1 kHz tone land here as does noise). If you pick a playback level of 110 dB, then the latter translates into 120 dB of dynamic range. In absence of TSS, and in quiet situation as in my closed headphones, you can hear such low level tones.

You need to get up much earlier than this to call my ethics into question.

 

[Frank Dernie]

There was no other amplification in the path. Audio captured using AP, and simply played on my Topping DX3 Pro.

Was the gain set on your amplifier such that a 0dB signal would give a 120dB sound level? Then the test 120dB below that would be at a sound level of -120dB. I am pretty sure I wouldn't be able to hear anything much then.
If you increase gain in the system above this in order to hear a tone you are amplifying the distortion too and saying it is -120dB is no longer true.
I don't believe I could hear something which was 120dB below full scale if the system was set at a level to listen to music, and certainly not an artefact 120dB below the music whilst it was playing.
I have been using the Fletcher -Munson equal loudness curves as well as recording music for over 50 years btw.
I know -120dB is near the threshold of hearing but I also know it has never existed during a music recording I have made.
120dB of dynamic range would be useful for an environmental noise nuisance recorder without AGC. For music pointless IME.

 

[amirm]

Was the gain set on your amplifier such that a 0dB signal would give a 120dB sound level? Then the test 120dB below that would be at a sound level of -120dB. I am pretty sure I wouldn't be able to hear anything much then.

That math would not give you dBSPL. You need to include the headphone and no, I don't have a combined acoustic measurement to know that.

 

[Frank Dernie]

That math would not give you dBSPL. You need to include the headphone and no, I don't have a combined acoustic measurement to know that.

My point entirely.
If you don't know what the overall amplification is when listening you don't know that you are listening to a -120dB tone.

 

[amirm]

My point entirely.
If you don't know what the overall amplification is when listening you don't know that you are listening to a -120dB tone.

But I know that is the level I asked the DAC to reproduce. And it created distortion that in my setup was audible. And that setup was ordinary, with no extra amplification.