Review and Measurements of JDS Labs EL DAC

[amirm]

There is a debate among so-called audiophiles who claim they can hear phase distortion right below Nyquist, where it is greatest. But frankly, I doubt that a human can hear above 20 kHz (I certainly can't, I'm far from it). Therefore, phase distortion at the DAC's output is very probably of very little concern.

There is good research done in this area by two of our signal processing luminaries in peer reviewed Journal of AES:
On the Audibility of Midrange Phase Distortion in Audio Systems
Authors: Lipshitz, Stanley P.; Pocock, Mark; Vanderkooy, John

I might also ask why someone cares about THD+N being at ridiculous and very probably inaudible low levels while not worrying about phase distortion. It makes no sense to me. I think if phase distortion is of no concern then THD+N at e.g. -90 dB is of no concern, too, and Amir would not need to make THD+N measurements any more .

They are different though. We have research that shows even copious amount of phase distortion/shift is inaudible. To wit, here is the report from Clark: Measuring Audible Effects of Time Delays in Listening Rooms

 

[andreasmaaan]

Yes, I would, science says audibility requires at least one cycle at any frequency. I don't think a reasonable minimum phase filter achieves that within the audio band.

There is a debate among so-called audiophiles who claim they can hear phase distortion right below Nyquist, where it is greatest. But frankly, I doubt that a human can hear above 20 kHz (I certainly can't, I'm far from it). Therefore, phase distortion at the DAC's output is very probably of very little concern.

But what happens if such a signal reaches the analog stage? A headphone amplifier for example? I think it's best to avoid audio band phase distortion right from the start.

I might also ask why someone cares about THD+N being at ridiculous and very probably inaudible low levels while not worrying about phase distortion. It makes no sense to me. I think if phase distortion is of no concern then THD+N at e.g. -90 dB is of no concern, too, and Amir would not need to make THD+N measurements any more .

I have to acknowledge your point about THD. I think -90dB is adequate (more, if of a more benign kind), or -100dB to be “safe” with a DAC, given it’s easy to achieve and the amp and speakers will no doubt add more.

For similar reason, I’d also prefer the linear phase filter in the case of this AKM chip, since all ringing will be well outside the audio band and there will be no time domain error whatsoever within it.

In both cases, I think we’re overdoing it, but given this is so easy, why not...

Not sure I understand your point about the significance of it reaching the analogue stage, however?

 

[gvl]

Yeah, DAC phase distortion is a moot point given the crossovers do their share of phase distortion in speakers. Even full-rangers that don't use crossovers must have some phase distortion due to the inductance of the voice coil.

 

[restorer-john]

'm pretty sure JDS uses it since AKM themselves call it the most "natural" filter, although their whole filter flavor thing is bullshit of course.

It is absolute BS.

Please see the graphs below. I upsampled an impulse at + full scale / 32-bit float / 44.1 kHz to 88.2 kHz with iZotope RX 7 using its internal Resample module (highest steepness, cutoff at Nyquist). Graph 1 is the spectrogram for linear phase, graph 2 for minimum phase. For minimum phase you can see phase distortion in the audio band (effectively starting at DC, the higher the frequency the later its arrival). By contrast, linear phase is not phase distorted. As an amendment: graph 3 is time domain for linear phase, graph 4 minimum phase.

That's all very well, but you aren't actually measuring anything there- the output of an actual D/A is somewhat different due to various factors.

Here are some actual impulses (single sample 0dBFS) on actual IIR (NOS 16 bit ladder D/A) and FIR D/A converters in commercial products.

Zoomed:

FIR

Zoomed:

The square wave in the review shot is a mess. Here is a square wave from the world's first CD player- The Sony CDP-101.

Notice how well the ringing is controlled in all examples?

 

[Grave]

According to Headfonia...
https://www.headfonia.com/review-jds-labs-el-dac-amp/2/
"The DAC on the Element is slightly bright, and on the thin side, though it is definitely good enough to be the built-in DAC on a $350 amp. That the EL DAC is a different design is immediately apparent upon first listening. The sound kicked out by the EL DAC adds a noticeable amount more body to the sound. It provides a much more full sound, and, due to the added body in the bass and mids, brings a touch of warmth to the proceedings as well. It does give the DAC a somewhat laid back feel to it."
and on the El Amp
"While the EL DAC is a new creation, the JDS Labs EL Amp IS the amp from The Element, just without the built-in DAC."

I'll probably just order the combo so I don't wonder if I should have...

That headfonia guy is an ignoramus. Do not read anything we writes.

 

[gvl]

It is absolute BS.



That's all very well, but you aren't actually measuring anything there- the output of an actual D/A is somewhat different due to various factors.

Here are some actual impulses (single sample 0dBFS) on actual IIR (NOS 16 bit ladder D/A) and FIR D/A converters in commercial products.

View attachment 16476

Zoomed:

View attachment 16477

FIR

View attachment 16478

Zoomed:

View attachment 16479

The square wave in the review shot is a mess. Here is a square wave from the world's first CD player- The Sony CDP-101.

View attachment 16480

Notice how well the ringing is controlled in all examples?

What do you mean by "controlled"? Shorter duration? Yours is 99Hz and Amir's is a 1kHz square wave, considering it the absolute duration of post-ringing is about the same.

 

[solderdude]

.... so here is S/PDIF averaged 16 times as I normally do:
...And no averaging:

Thanks Amir, I wanted to know if the 'poles' get smaller when averaging. Seems like they are there all of the time, like the stimulus.

 

[March Audio]

There is still debate whether this is audible to humans or not. You can test for yourself with the examples.

Sounds with a short attack time (such as drums, especially snare drum) distort due to phase distortion as the energy pack that builds up that short sound diverges (for example, the higher the frequency the later it comes in comparison to the original).

Personally, I would not want a DAC's digital upsampling filter to additionally distort phase in the audio band. Perhaps it is not consciously audible with certain test signals. But it may be subconsciously audible with actual music.

I'm afraid this is a bit of a myth. Drums dont have the high frequency content you might imagine. The phase linearity of dacs in band is easy to measure and will be extremely good.

As talked about elsewhere this testing with single samples and square waves is misleading as these by definition are not real signals. Digital audio is band limited. What you are reproducing was band limited at the ADC. So testing replay filter with single samples tells you nothing useful about in band behaviour of an actual music recording .

 

[derp1n]

I'm afraid this is a bit of a myth. Drums dont have the high frequency content you might imagine. The phase linearity of dacs in band is easy to measure and will be extremely good.

As talked about elsewhere this testing with single samples and square waves is misleading as these by definition are not real signals. Digital audio is band limited. What you are reproducing was band limited at the ADC. So testing with single samples tells you nothing useful about in band behaviour.

Are you saying a minimum phase filter is phase linear in the audio band?

 

[solderdude]

from stereophile: https://www.stereophile.com/content/ringing-false-digital-audios-ubiquitous-filter-page-2

 

[March Audio]

I'm afraid this is a bit of a myth. Drums dont have the high frequency content you might imagine. The phase linearity of dacs in band is easy to measure and will be extremely good.

As talked about elsewhere this testing with single samples and square waves is misleading as these by definition are not real signals. Digital audio is band limited. What you are reproducing was band limited at the ADC. So testing with single samples tells you nothing useful about in band behaviour.

Sorry I wasn't clear, answering in a hurry whilst doing something else.

What I was alluding to here is that the system isnt output only. What has the ADC done at the front end?

 

[March Audio]

I disagree. I would even say it is a false general statement. If you are interested, I could prove it with spectrograms showing the attack time of kick drum, snare drum, and tom tom. These can easily go beyond 20 kHz.
But you can see it for yourself. Just download the demo version of iZotope's RX7, load an audio file with real drums, and check the spectrogram.
By the way, I have seen attack times of acoustic guitar strings (both metal and nylon) go way beyond 20 kHz. For example, a 192 kHz high resolution version of the song "Hotel California" by The Eagles. If I remember correctly, the frequencies went up to about 70 kHz!
All this is not based on my imagination. This is based on high resolution spectrograms in RX. So pure math.

A generalisation yes but yes I have looked at spectgrams of drums and the high frequency content is limited.

It's no good now shifting the focus to other instruments from your original statement.

Are you really suggesting that hotel California has musical information up to 70khZ? Come on that's not realistic. What microphones did they use to do that? What instruments made that sound? What analogue tape recorder did they use?

I will look at this file later but I am quite confident all you will see is noise.

 

[March Audio]

AFAIK these tests have been done since the beginning of digital audio, or at least since the beginning of the Compact Disc. You really question their usefulness?

They are useful for technically characterising filters however the replay filter is not acting in isolation. You need to consider the whole system from the front acquisition end.

And what's phase coherent about any recording made with a bunch of microphones?

Also, should we really be talking about group delay here?

 

[amirm]

Thanks Amir, I wanted to know if the 'poles' get smaller when averaging. Seems like they are there all of the time, like the stimulus.

Correct. Averaging gets rid of the "grass" around the base so that we can better see the distortion spikes. Any spikes including the original signal remain unchanged from run to run so their value does not change (unless the output changes).

 

[amirm]

With which driver does it happen? Microsoft's generic driver? Or the JDS driver? Or both?
Just in case the developer of ASIO4ALL asks.

The problem is only limited to inbox class drivers in Windows. Installing vendor drivers fixes it. If ASIO4ALL is used on native drivers, it will truncate to 16 bits. At least it has on two DACs I have tested so far.

 

[gvl]

Really? Hotel California remaster (40th Anniversary, 2017, Blu-ray disc, 24 bits, 192 kHz, stereo):

Do you consider a possibility it is an example of bad mastering with artificial ultrasonic junk that shouldn't be there? I mean, no way tape recorders of the 70s could capture that high of a frequency.

 

[March Audio]

I did not do that. I just gave another concrete example.

Really? Hotel California remaster (40th Anniversary, 2017, Blu-ray disc, 24 bits, 192 kHz, stereo):
View attachment 16493
Zoom: View attachment 16494

What does it matter? The proof is in the spectrogram.

See above.

Sorry but there is nothing but noise above about 25 kHz, which would be consistent with the technology used to record the track.

It's a concrete example supporting what I said. You need to understand what you are looking at.

 

[March Audio]

I don't need to. What do you want to talk about instead?

?

 

[solderdude]

Downsample the recording to say 24/48 (using a known good downsampler) and ABX test it to see if the extra bandwidth is contributing to the sound.

Of course you need a transducer that can reproduce it... specs of transducers merely stating they can reach 40kHz is no guarantee they do within a few dB.

What I find interesting is that the high freq. peaks that are detected and act as evidence in this case are not reflected in the actual waveform. I mean I don't see any high frequencies in the waveform (they could be small enough not to show in a linear scale though)
maybe ... just maybe the analysis of the software is erronious ?

 

[solderdude]

This looks more like actual overtones that have been recorded on tape than the previous example which looks more like an analysis error.
As you said, recorders in those days could record well over 20kHz. Most of these recorders did not have brickwalls in it. At best there was a notch filter around the bias frequency or a gentle filter well in front of the bias frequency.
As the bias usually was between 80kHz and 200kHz that isn't much of a problem.
40kHz is no problem for tape (with enough tape speed). It also isn't for hires formats.

Here too, downsample it to 48kHz and ABX it. When you ace the test there is reason to believe it actually matters to you.