New Audio-GD DI-20?

[GoldenOne]

I've posted some measurements of the Jitter and noise from the DI-20HE.
This thing is seriously impressive and is the best DDC I've seen thus far.

https://goldensound.audio/2021/07/22/audio-gd-di20he-measurements/

[GoldenOne]

5 hours ago, Confused said:

@GoldenOne - Excellent review and measurements, great information and waffle free text.  I have to say that I have seen some rather disappointing measurements for some other Audio GD products, so I had somewhat dismissed the DI-20-HE.  So it is great to see some measurements for the product itself that show that is actually performs very nicely.

 

As someone who currently uses a Mutec MC3+USB, I was interested to compare the Audio GD measurements with those performed by Hifi News for the MC3+USB.  To be honest, I am struggling a little correlating the two.  See link:

 

https://www.mutec-net.com/downloads/manuals/MUTEC_MC-3plusUSB_-_HFN.pdf

 

Is it possible to correlate the HFN Mutec measurements to yours for the Audio GD DI-20-HE?

 

Yeah I was.... Well let's just say not expecting much given some of the performance of agd's other products. But this is sublime. 

 

With the mutec they don't have any actual direct jitter measurements there, just a j-test of it feeding a dac. Which can't be directly equated as the dac itself will affect things. 

 

You can compare two sources using the same dac for j-testing but can't really use two different ones. 

 

If I manage to get my hands on the mutec at any point I'll post some measurements of it though. 

[GoldenOne]

3 hours ago, Confused said:

I was looking at this yesterday, and later reading a review of the Auralic Altair G2.1 in Hifi News and Record Review.

 

For the Auralic, HFN's "lab report" measurements quote jitter at 14psec at 48kHz and 14psec at 96 kHz.

 

Looking at the Goldensound measurements of the Audio GD, at 48kHz there is the level / frequency graph, which shows a reading of about 3p level (s) downwards from 20Hz, and below the graph figures (RMS) of 78.74ps and (peak) 229.2ps.

 

I am almost certainly demonstrating my own lack of knowledge here, but whilst I can easily compare one HFN measurement to another, and one Goldensound measurement to another, I cannot fathom how to compare a Goldensound measurement to a HFN measurement.

 

I presume the measurements from something like the Auralic and the Audio GD to be broadly similar, but how should the Auralic's 14psec number be compared to the Audio GD's 3p level (s), 78.74ps / 229.2ps numbers? My simple brain tells me that 3p level s is a lot lower than 14 psec, but 229.2 ps is much higher, so I am at a bit of loss as to how compare these different measurements.

 

Can anyone clarify this? I am slightly puzzled!

Could you link this post? The HFN lab report I found doesn't seem to have the information you mentioned

[GoldenOne]

6 minutes ago, Confused said:

Ha! No I cannot. The HFN article I read was in an actual magazine made of old school paper. 

 

Here is a snip of the lab report:

 

 

 

And the relevant part of the lab report words:

 

 

 

Ahh ok, this cannot be directly equated/compared to my measurements. The ones in the article are 'inferred jitter', measures using a J-Test through a DAC.

My test is measurement of the jitter directly from the digital source. (With FFT spectrum of jitter)
I may start including J-Tests in future as well (particularly cause I can't measure I2S directly so it'd be useful to have some info on that in case it differs from coax performance significantly), but the problem is that even with the same digital source, the dac used will affect the measurement as different DACs have different approaches to PLLs, resampling etc.

Actually the most ideal DAC for the task would be a NOS, no-PLL R2R dac, such that there is no further digital processing or reclocking. The Holo May actually suits this very well given as you can disable the PLL so I'll look into that. BUT, it still can't be directly compared to the HFN measurements as they're using a different DAC.

 

[GoldenOne]

On 8/30/2021 at 12:33 PM, Confused said:

Many thanks for the response, which is clear.

 

So ignoring the HFN numbers and just looking at you own measurements, these is one other thing that puzzles me a little.

 

Taking as an example your 48kHz jitter measurements for the Audio GD D120HE, the graphical representation indicates Level (s) from about 2ps / 3ps down to below 1e-13 ps at very high frequencies. What I am struggling to fathom is how this relates to the RMS level s and Peak level s readings, which are 78.74 ps and 229.2 ps respectively. In other words, 78.74ps is way higher than anything indicated graphically. I am sure that I am missing something fundamental here?

It's basically due to the maths of how FFT works/what it is showing.

Here is an example of an analog signal, with many different tones:
 

 

Now, you can see that on the top right in the FFT, each individual tone is only up to about 200mv (0.2v). But our peak and RMS values are MUCH higher.

This is because as you can see in the scope view in the top left, all these individual tones combine to create a waveform that will have a higher amplitude than just one of those tones. And so that waveform peaks at about 2.89V. 

The FFT is showing us a clearer view of what the signal contains, but it's showing us the level of each individual frequency/part of the signal (Bin), not the level of the signal as a whole.

The exact same thing happens when looking at the spectrum/waveform of jitter. And the problem is worse because the stuff we are looking at is so close to the noise floor.

For a signal where the main 1khz tone etc is going to be >100dB above noise and harmonics, then any noise/harmonics are going to be 0.0001% or so of the whole signal, and will have almost no impact on the peak/rms values.

But when we are looking at stuff where there is lots of content of a similar/close value, like with the above example of lots of tones, or in the case of jitter, where there is a noise floor and some small spikes just a little tiny bit above it, then everything has a noticeable impact.

Here's another analog signal example. 25uV 1khz sine, just slightly above the noise floor:
 

 

Here we can see the 25uV 1khz sine, but we also have a lot of noise not too far below it. Meaning that is having a noticeable effect on the peak/RMS value. You can see on the scope how much it's impacting it. And as a result the waveform is peaking at about 76uV, with an RMS value of 26.48.

And in a situation where we have no main signal and are just looking at noise:
 

 

The noise floor on the FFT might look like it's at about 20nV. But all of that is combining to make a waveform that peaks at almost 9uV, wayyyy higher than the FFT shows.

This is also why you should never judge the noise floor of a device based on FFT, it can be changed/altered depending on how the FFT itself is set up.


TLDR: FFT shows you the level of individual parts of the signal by dividing it into different frequencies. It does not show the absolute peak value of the signal.

Hope that helped!