DAC <$10k buying advice, 3 choices
Dec 30, 2022 at 2:52 AM Post #76 of 134
Seems like a well thought out pairing for you. I have gone through the r2r with tube phase already and my music preference is very different from yours. Having choice is a good thing.
Aqua is “competitive”. Competitive doesn't mean it's the best, buy only that it should be listened to in your own system and compared with others. This is imperative advice.
 
Jan 3, 2023 at 4:10 AM Post #77 of 134
I value ASR, Golden One and the likes for their measurements. Don’t care for their assessment however. For example, May sounded meh to me on more than 1 occasion. I know it measures extremely well on the standard bench for a r2r dac. Of course not all audio reproduction quality can be measured with just the standard bench tests.

I would also encourage Golden One given there are not enough reviewers that do measurements.
It is still a mislead we have such “influencers” that has a gathering of brainless defenders. Im not thankful for anything they both make.

Love it or not but they are like a cult. This kind of madness should be stopped.

Both.
 
Last edited:
Jan 3, 2023 at 9:45 AM Post #78 of 134
For a cult around measurements I would blame marketing and those who are behind. Of course there are brainless defenders, it is a profitable investment. Those in business who build their reputation for years are under pressure to not lose customers. I respect those who are trying to pass a message, even not in a straight form that would drop a bomb for masses. Measurements can show marketing lies when it is done properly (even against the intention to prove the otherwise), like a case Denafrips and HoloAudio NOS.

Example from the recent story. When John Attkinson from Stereophile encountered a strange behaviour of Holo Audio Springs during tests, he didn't take an attention of irregularities. Next month or two when he tested Holo May which behaved identically, he decided to point out some irregularities. He didn't say in a plain language that roll-off frequency response doesn't conform to the standard NOS sample&hold operation, so then the advertised NOS feature is doubtful. However he gave a hint of the problem, presenting numbers what was expected and what measured. These differences indicate digital filtering with similar characteristic at higher sampling rates, I picked up a message on first reading.

A second irregular behaviour was pre/post ringing on a pulse response test. JA (the same as with Springs) interpreted it as a bandwith limitation of his analyser (200kHz sampling of his A/D converter). It is a valid behaviour in the bandwith limited environment, however the same 1kHz square test signal doesn't show band limitation (ringing) when testing on the same equipment Audio Note DACs (which NOS feature is not questioned). It is indicative that ringing is caused by something else. Someone asked: "JA, when did you change input filters?", there is no response. I think JA was not able to find a culprit using his equipment. Hats off for what he did.

I was thinking that digital filtering (as detected by JA) was a reason of ringing, I was wrong. Filtering is mild, not a brick wall type, just cut-off frequency is only slightly affected. During a discussion @GoldenOne came with his new analyser featuring 3GHz A/D sampling rate. It didn't show ringing, but ringing appeared when sampling rate was reduced to the value used in JA tests. Proving NOS feature? According to @GoldenOne it did. I looked close, comparing a plot at these two sampling rates of the analyser and found that various spikes at ~200kHz test disappeared at 3GHz and contributed in increasing noise floor.

It comes to discovery that Holo Audio use ultrasonic scrambling in order to randomise ladder imperfections. This is the energy that cause ringing in JA tests. Note, it is not a precise error correction as they claim of using a secondary correction ladder. In fact I couldn't identify a secondary ladder from photos. It doesn't exists, there is something else. The added ultrasonic noise shifts the moment where individual bits are activated. A final effect is similar to noise shaping (integral part of DS decoding). When a scrambling energy level gives more voltage variations than ladder imperfections, the analyser display an averaged smooth curve. Brainless defenders interpret that such equipment is better, more advanced, forgetting a fact that the real ladder accuracy is no more than 12-14 bits, in the best case 16 bits.
 
Last edited:
Jan 3, 2023 at 1:03 PM Post #79 of 134
Holo Audio dacs do very well on basic measurements. Good basic bench measurements do not necessarily mean good sound reproduction. The problem with Chinese brands is they don’t have a good dealer network in the US so it’s really hard to compare without buying blind. Among the few times I heard Holo and Denefrips (different setup), I actually thought Denefrips sounded pretty good for the price and worth a listen for any potential buyer.

Also the business practice of Holo makes me nervous.
 
Jan 3, 2023 at 3:30 PM Post #80 of 134
According to @GoldenOne it did. I looked close, comparing a plot at these two sampling rates of the analyser and found that various spikes at ~200kHz test disappeared at 3GHz and contributed in increasing noise floor.
Please provide the plots you are referring to here, thank you

It comes to discovery that Holo Audio use ultrasonic scrambling in order to randomise ladder imperfections. This is the energy that cause ringing in JA tests.
No it is not, the ringing as discussed and evidenced several times before is because if you view a signal with content above the nyquist frequency of the analyzer (including any square wave or NOS impulse), you get ringing, due to the ADC filter.
If you look at a higher bandwidth to capture the full signal, it does not occur, because it is not part of the signal coming from the DAC.
Please stop repeating this, it's simply not true
 
Jan 3, 2023 at 3:31 PM Post #81 of 134
Jan 3, 2023 at 4:19 PM Post #83 of 134
Jan 3, 2023 at 4:22 PM Post #84 of 134
For a cult around measurements I would blame marketing and those who are behind. Of course there are brainless defenders, it is a profitable investment. Those in business who build their reputation for years are under pressure to not lose customers. I respect those who are trying to pass a message, even not in a straight form that would drop a bomb for masses. Measurements can show marketing lies when it is done properly (even against the intention to prove the otherwise), like a case Denafrips and HoloAudio NOS.

Example from the recent story. When John Attkinson from Stereophile encountered a strange behaviour of Holo Audio Springs during tests, he didn't take an attention of irregularities. Next month or two when he tested Holo May which behaved identically, he decided to point out some irregularities. He didn't say in a plain language that roll-off frequency response doesn't conform to the standard NOS sample&hold operation, so then the advertised NOS feature is doubtful. However he gave a hint of the problem, presenting numbers what was expected and what measured. These differences indicate digital filtering with similar characteristic at higher sampling rates, I picked up a message on first reading.

A second irregular behaviour was pre/post ringing on a pulse response test. JA (the same as with Springs) interpreted it as a bandwith limitation of his analyser (200kHz sampling of his A/D converter). It is a valid behaviour in the bandwith limited environment, however the same 1kHz square test signal doesn't show band limitation (ringing) when testing on the same equipment Audio Note DACs (which NOS feature is not questioned). It is indicative that ringing is caused by something else. Someone asked: "JA, when did you change input filters?", there is no response. I think JA was not able to find a culprit using his equipment. Hats off for what he did.

I was thinking that digital filtering (as detected by JA) was a reason of ringing, I was wrong. Filtering is mild, not a brick wall type, just cut-off frequency is only slightly affected. During a discussion @GoldenOne came with his new analyser featuring 3GHz A/D sampling rate. It didn't show ringing, but ringing appeared when sampling rate was reduced to the value used in JA tests. Proving NOS feature? According to @GoldenOne it did. I looked close, comparing a plot at these two sampling rates of the analyser and found that various spikes at ~200kHz test disappeared at 3GHz and contributed in increasing noise floor.

It comes to discovery that Holo Audio use ultrasonic scrambling in order to randomise ladder imperfections. This is the energy that cause ringing in JA tests. Note, it is not a precise error correction as they claim of using a secondary correction ladder. In fact I couldn't identify a secondary ladder from photos. It doesn't exists, there is something else. The added ultrasonic noise shifts the moment where individual bits are activated. A final effect is similar to noise shaping (integral part of DS decoding). When a scrambling energy level gives more voltage variations than ladder imperfections, the analyser display an averaged smooth curve. Brainless defenders interpret that such equipment is better, more advanced, forgetting a fact that the real ladder accuracy is no more than 12-14 bits, in the best case 16 bits.
I'm going to write this up again for the benefit of those reading.

The graph which @sajunky is referring to is this one:
820HoMayfig01.jpg

This is from John Atkinson's measurements of the May which can be found here: https://www.stereophile.com/content/holoaudio-may-level-3-da-processor-measurements

Sajunky is asserting that what is being shown here is evidence of DSP because there is ringing visible on the graph, and a true NOS impulse should not have any ringing (it should be a square).
John Atkinson quite clearly states just before this graph:

"Ignore the very small amount of symmetrical ringing before and after the single full-scale sample, which is due to the antialiasing filter of Audio Precision's A/D converter operating at a sample rate of 200kHz"

Which is exactly what is happening here. Though Sajunky has chosen to ignore this and is using it as evidence of some sort of measurement cheat.

We can show this behaviour with some other devices quite easily, let's use a different brand and product just to be sure.
Here's the NOS impulse response of the Schiit Bifrost 2/64 recorded on the same analyzer with the same sample rate config that John Atkinson used:

1672779386516.png

Seems familiar eh?
Let's zoom in a bit (same measurement, just zoomes on the X axis)

1672779427350.png


Clear ringing! So it can't be NOS right?
Well, no, as John Atkinson stated, this is simply because the ADC is running at 200khz. A true square wave has an infinite bandwidth (in practice it's limited by the slew rate of the device), and so we aren't running the ADC fast enough to capture most of the signal.
Let's run it again, exactly the same, but this time run the ADC at 2.5Mhz to capture a higher bandwidth:
1672779651281.png


Lots less ringing!
So does this mean the DAC is ringing less?
No, the DAC is performing exactly the same. It's just that we're now capturing more of the signal.
The remaining 'ringing' seen on this particular impulse is then just because of impedance matching usually and is why in RF systems cables with characteristic impedances of 50 Ohms are used.

We can also repeat this very same behaviour with the analyzer's own square wave generator.
Here's an 11.025khz square wave for example:

1672780482628.png

No ringing, great! What if we measure this but with the ADC sampling rate reduced from 2.5Mhz to 200khz like with the earlier impulse response tests?
1672780525639.png

Ringing galore! Not because the signal ACTUALLY has any ringing on it (it doesn't), but because the ADC isn't sampling fast enough to capture the full signal. You cannot show a square wave if you have a limited bandwidth.
If we reduce the ADC sample rate to say 44.1khz, it now looks like this:

1672780616293.png

Looks like a sine wave now, not because it's actually a sine wave, but simply because we don't have enough bandwidth to capture above 22.05khz. And because even the first harmonic gets cut out by the filter, what we are left with is what LOOKS like a single sine.

Capturing (almost) the full bandwidth, over 1Mhz, we see a very square looking result:

1672780775177.png



Reduce your sample rate to ~200khz and therefore your capture bandwidth to 100khz, and you cut off most of the higher frequency components, meaning you cannot fully describe the square wave and are left with just the lower frequency components, which shows as ringing:

1672780861626.png


Cut it down further, to a 44.1khz ADC sample rate, and now you only have enough bandwidth to capture that first component. Everything else gets cut off. So it LOOKS like a sine on your display even though it's actually a square wave.

1672780938537.png


The ringing shown on the May measurement from JA is as JA himself said, simply because of the ADC being ran at a low-ish sample rate. When sampled higher, it does not ring. Nor is there any evidence of any unusual high frequency content or 'ultrasonic scrambling'.

He didn't say in a plain language that roll-off frequency response doesn't conform to the standard NOS sample&hold operation
He didn't say this because it DOES conform to the expected frequency response for sample & hold operation......
 

Attachments

  • 1672780738623.png
    1672780738623.png
    78.1 KB · Views: 0
Last edited:
Jan 3, 2023 at 4:33 PM Post #85 of 134
I'm going to write this up again for the benefit of those reading.

The graph which @sajunky is referring to is this one:
820HoMayfig01.jpg

This is from John Atkinson's measurements of the May which can be found here: https://www.stereophile.com/content/holoaudio-may-level-3-da-processor-measurements

Sajunky is asserting that what is being shown here is evidence of DSP because there is ringing visible on the graph, and a true NOS impulse should not have any ringing (it should be a square).
John Atkinson quite clearly states just before this graph:

"Ignore the very small amount of symmetrical ringing before and after the single full-scale sample, which is due to the antialiasing filter of Audio Precision's A/D converter operating at a sample rate of 200kHz"

Which is exactly what is happening here. Though Sajunky has chosen to ignore this and is using it as evidence of some sort of measurement cheat.

We can show this behaviour with some other devices quite easily, let's use a different brand and product just to be sure.
Here's the NOS impulse response of the Schiit Bifrost 2/64 recorded on the same analyzer with the same sample rate config that John Atkinson used:


Seems familiar eh?
Let's zoom in a bit (same measurement, just zoomes on the X axis)



Clear ringing! So it can't be NOS right?
Well, no, as John Atkinson stated, this is simply because the ADC is running at 200khz. A true square wave has an infinite bandwidth (in practice it's limited by the slew rate of the device), and so we aren't running the ADC fast enough to capture most of the signal.
Let's run it again, exactly the same, but this time run the ADC at 2.5Mhz to capture a higher bandwidth:


Lots less ringing!
So does this mean the DAC is ringing less?
No, the DAC is performing exactly the same. It's just that we're now capturing more of the signal.
The remaining 'ringing' seen on this particular impulse is then just because of impedance matching usually and is why in RF systems cables with characteristic impedances of 50 Ohms are used.

We can also repeat this very same behaviour with the analyzer's own square wave generator.
Here's an 11.025khz square wave for example:


No ringing, great! What if we measure this but with the ADC sampling rate reduced from 2.5Mhz to 200khz like with the earlier impulse response tests?

Ringing galore! Not because the signal ACTUALLY has any ringing on it (it doesn't), but because the ADC isn't sampling fast enough to capture the full signal. You cannot show a square wave if you have a limited bandwidth.
If we reduce the ADC sample rate to say 44.1khz, it now looks like this:


Looks like a sine wave now, not because it's actually a sine wave, but simply because we don't have enough bandwidth to capture above 22.05khz. And because even the first harmonic gets cut out by the filter, what we are left with is what LOOKS like a single sine.

Capturing (almost) the full bandwidth, over 1Mhz, we see a very square looking result:




Reduce your sample rate to ~200khz and therefore your capture bandwidth to 100khz, and you cut off most of the higher frequency components, meaning you cannot fully describe the square wave and are left with just the lower frequency components, which shows as ringing:



Cut it down further, to a 44.1khz ADC sample rate, and now you only have enough bandwidth to capture that first component. Everything else gets cut off. So it LOOKS like a sine on your display even though it's actually a square wave.



The ringing shown on the May measurement from JA is as JA himself said, simply because of the ADC being ran at a low-ish sample rate. When sampled higher, it does not ring. Nor is there any evidence of any unusual high frequency content or 'ultrasonic scrambling'.
Well look at that: an evidence backed, scientific answer that is sound from an engineering perspective and also repeatable.

Can’t wait for the pseudo-science, ad hominem responses lacking any first-hand evidence to weave wildly unprovable conspiracy theories.
 
Last edited:
Jan 3, 2023 at 4:35 PM Post #86 of 134
I'm going to write this up again for the benefit of those reading.

The graph which @sajunky is referring to is this one:
820HoMayfig01.jpg

This is from John Atkinson's measurements of the May which can be found here: https://www.stereophile.com/content/holoaudio-may-level-3-da-processor-measurements

Sajunky is asserting that what is being shown here is evidence of DSP because there is ringing visible on the graph, and a true NOS impulse should not have any ringing (it should be a square).
John Atkinson quite clearly states just before this graph:

"Ignore the very small amount of symmetrical ringing before and after the single full-scale sample, which is due to the antialiasing filter of Audio Precision's A/D converter operating at a sample rate of 200kHz"

Which is exactly what is happening here. Though Sajunky has chosen to ignore this and is using it as evidence of some sort of measurement cheat.

We can show this behaviour with some other devices quite easily, let's use a different brand and product just to be sure.
Here's the NOS impulse response of the Schiit Bifrost 2/64 recorded on the same analyzer with the same sample rate config that John Atkinson used:


Seems familiar eh?
Let's zoom in a bit (same measurement, just zoomes on the X axis)



Clear ringing! So it can't be NOS right?
Well, no, as John Atkinson stated, this is simply because the ADC is running at 200khz. A true square wave has an infinite bandwidth (in practice it's limited by the slew rate of the device), and so we aren't running the ADC fast enough to capture most of the signal.
Let's run it again, exactly the same, but this time run the ADC at 2.5Mhz to capture a higher bandwidth:


Lots less ringing!
So does this mean the DAC is ringing less?
No, the DAC is performing exactly the same. It's just that we're now capturing more of the signal.
The remaining 'ringing' seen on this particular impulse is then just because of impedance matching usually and is why in RF systems cables with characteristic impedances of 50 Ohms are used.

We can also repeat this very same behaviour with the analyzer's own square wave generator.
Here's an 11.025khz square wave for example:


No ringing, great! What if we measure this but with the ADC sampling rate reduced from 2.5Mhz to 200khz like with the earlier impulse response tests?

Ringing galore! Not because the signal ACTUALLY has any ringing on it (it doesn't), but because the ADC isn't sampling fast enough to capture the full signal. You cannot show a square wave if you have a limited bandwidth.
If we reduce the ADC sample rate to say 44.1khz, it now looks like this:


Looks like a sine wave now, not because it's actually a sine wave, but simply because we don't have enough bandwidth to capture above 22.05khz. And because even the first harmonic gets cut out by the filter, what we are left with is what LOOKS like a single sine.

Capturing (almost) the full bandwidth, over 1Mhz, we see a very square looking result:




Reduce your sample rate to ~200khz and therefore your capture bandwidth to 100khz, and you cut off most of the higher frequency components, meaning you cannot fully describe the square wave and are left with just the lower frequency components, which shows as ringing:



Cut it down further, to a 44.1khz ADC sample rate, and now you only have enough bandwidth to capture that first component. Everything else gets cut off. So it LOOKS like a sine on your display even though it's actually a square wave.



The ringing shown on the May measurement from JA is as JA himself said, simply because of the ADC being ran at a low-ish sample rate. When sampled higher, it does not ring. Nor is there any evidence of any unusual high frequency content or 'ultrasonic scrambling'.


He didn't say this because it DOES conform to the expected frequency response for sample & hold operation......
As an addition to this, here's a GIF that provides quite a good visual demonstration of how a fourier series or sum of sines builds a square wave. And why if you don't have enough bandwidth to capture the additional higher frequency components, it's not possible to capture a 'non-ringing' square wave

Fourier_series_square_wave_circles_animation.gif
 
Jan 3, 2023 at 4:38 PM Post #87 of 134
Your opinion is immature. I've visited your website and it says your 24 years young. Outstanding! But it's known in medicine and science that the frontal lobe isn't fully developed till age 25.

Continue on with your work. I'm sure at some point you'll have experience and unbiased opinions to share.

I should also add you clearly have sponsors that support your work.

You are a professional.

State the truth.

well done old chap you really showed him who's boss
 
Jan 3, 2023 at 5:17 PM Post #88 of 134
As an addition to this, here's a GIF that provides quite a good visual demonstration of how a fourier series or sum of sines builds a square wave. And why if you don't have enough bandwidth to capture the additional higher frequency components, it's not possible to capture a 'non-ringing' square wave

Fourier_series_square_wave_circles_animation.gif
This gif is pretty good. I like it a lot. I can see nothing wrong with your explanation of the pre/post ringing AP measurement anomalies.

I continue to believe DAC without a reconstruction filter is a mistake from both an engineering and audio reproduction perspectives. The output can never be the same as the original signal due to under over error / step function. NOS is ancient tech that should just be left to die.

back to Holo May, I am still waiting to hear it in a system that impresses me. So far meh.
 
Jan 3, 2023 at 5:19 PM Post #89 of 134
Please elaborate.
Jeff kind of relies on word of mouth (also including influencers and pro reviewers among actual users) to sell his products - it’s a unique marketing approach for sure. I suspect the reason Holo Audio doesn’t have a website is because it’s not an actual company - not registered at least. I believe he is a one man show which makes me nervous.

For comparison Denafrips is a real company based in Shenzhen China, the company is registered in Guangzhou and Denafrips is a real trademark registered in the US.

Too many red flags with Holo - pass for me from an ownership perspective.
 
Jan 3, 2023 at 5:22 PM Post #90 of 134
I continue to believe DAC without a reconstruction filter is a mistake from both an engineering and audio reproduction perspectives
I agree.
Objectively a reconstruction filter IS necessary for proper playback of PCM audio, and whilst subjectively NOS can absolutely be enjoyable for some genres and I do listen in NOS sometimes too, proper oversampling is indeed the 'correct' way, and the majority of the time I use HQPlayer for this purpose.
 

Users who are viewing this thread

Back
Top