HOLO Audio MAY DAC

[Miska]

3 hours ago, asdf1000 said:

 

@Miska do you have any DSD related pops with your Spring2, using your HQP OS image as NAA?

 

When using your headphones setup?

 

Spring 2 is on my loudspeaker system, but the preamp has headphone socket as well. Spring 1 is on my headphone system.

 

If you mean start/stop pops/clicks, absolutely nothing on either. I cannot check again right now.

 

[Miska]

23 hours ago, louisxiawei said:

I also don’t have start/stop pops. The pop I meant only occurres when switching to another track manually. Dac8 used to have this kind of pops and it might still have it on Windows OS.

 

That is not very clear case, because what happens depends on the tracks being played. If they are same sample rate it is pretty much gapless operation. If they are different sample rate, then it depends if output rate family is being switched or not. If output rate is being switched between tracks, then the DAC is stopped, reconfigured and started again. If the output rate/format doesn't change, then it is pretty much gapless operation from the DAC's point of view.

 

I have not noticed pops with DAC8 DSD on Windows either...

 

If you are on Windows, make sure you have HQPlayer's Buffer Time set to "Default" and that you have large/safe buffers set from the driver's control panel so that you don't get dropouts at the point when CPU's can hit 100% load momentarily.

 

[Miska]

7 hours ago, louisxiawei said:

When I dragged the same track time bar (e.g. from 1:20 to 2:45), there is a pop occurred on MAY (or DAC8 on windows). I remember I sent an email to T+A regarding this and they confirmed this pop noise and told me it's normal.

 

Ahh, I don't do that. But that is discontinuation in the waveform and will result in some kind of a glitch. Nothing big, something  within the normal playback volume.

 

[Miska]

On 1/2/2020 at 8:41 PM, mevdinc said:

That's what I did when I used DACs with no volume control but a good DAC with a decent pre section gives much better results than using software volume control.
My current Musicbook has a very good analog volume control. I may just use it as a pre if I go for the MAY DAC.

 

Well, technically properly done digital volume control is much better than anything you can achieve with analog.

 

But how about something like this as pre?

https://www.stereophile.com/content/benchmark-la4-line-preamplifier

 

[Miska]

54 minutes ago, mevdinc said:

Sure, I agree re digital volume.
But I also need just one analog input for home cinema use so hence my preference for a good DAC with a decent pre and one analog input such as what I have now.

Thanks for the pointer, Benchmark LA4 does indeed look like good.

 

I rather get a good pre and a good DAC, than try to bundle the two into some strange beast. My current pre in the listening room system has 8 inputs of which 3 are balanced. I would prefer to have 8 balanced inputs though. But for analog sources I have ADC since I want to have digital room correction also for such sources. So such sources go through the same digital playback chain as digital files/streams.

 

For my headphone rig I'm planning to get a Benchmark HP4 when my budget permits... It is like LA4, but with great headphone amp.

 

[Miska]

And with some DSP help you can boost these R2R's to even better performance (20+ dB distortion improvements at low levels)...

 

[Miska]

6 hours ago, GoldenOne said:

What sort of DSP tweaks would you use with a dac like the May?

 

Same as Spring 2, so if you use PCM mode, set DAC Bits 20 and use noise shaper like LNS15 at 1.5 MHz sampling rate. I personally stick to DSD256 output using ASDM7EC modulator.

 

[Miska]

2 hours ago, GoldenOne said:

Yeah I think it was more a case of the original USB implementation wasn't great, and so people used I2S.

 

So far nobody has shown objective evidence of such. I would be interested to see though.

 

[Miska]

4 minutes ago, GoldenOne said:

What is the reasoning for those settings? (Other than the dac bits which I understand)
HQplayer is fantastic, and i'm curious to learn about why these settings benefit the spring.

I'd be using a network streamer that only supports 768khz. Would the desired settings other than sample rate differ in that case?
 

 

I've not seen any either, just that I'd seen a lot of subjective results that i2s was an improvement.
In any case the usb module has since been upgraded, and the May L2 and above uses a further improved version so I can't imagine there would be issues anymore.

 

It is due to realities of physical world, ladders have limited resolution, but can be helped enormously by suitable DSP to correct inaccuracies that are due to manufacturing tolerances and such.

 

13 minutes ago, GoldenOne said:

I'd be using a network streamer that only supports 768khz. Would the desired settings other than sample rate differ in that case?

 

Same settings are fine.

 

16 minutes ago, GoldenOne said:

I've not seen any either, just that I'd seen a lot of subjective results that i2s was an improvement.
In any case the usb module has since been upgraded, and the May L2 and above uses a further improved version so I can't imagine there would be issues anymore.

 

I2S is designed for connections between chips on the same board. But even there, the board layout design has massive impact on performance.

 

Between devices it has fundamental problem with clocking because in most implementations (since there is no standard), clock is going in wrong direction. So my default assumption is that it will make things worse. Since data itself is the same, only difference is in clocking.

 

[Miska]

AFAIK, the two are technically very different...

[Miska]

12 hours ago, skipspence said:

Interesting to know what differences are between them, could you please enlighten?

 

Holo Audio has pretty steep analog reconstruction filter, while Denafrips doesn't seem to have much if any (maybe first order)? So the stuff between the conversion section and output differs quite a lot.

 

[Miska]

2 hours ago, fds said:

@MiskaWondering about the sonic implications of a steep analog reconstruction filter? Does it mean it will roll off earlier in the frequency response and thus sound less open?

 

Sonic implications are mostly more accurate analog signal reconstruction. You always need analog reconstruction filter after the conversion stage. Combination of digital oversampling filters, noise-shaper/modulator, possible conversion filter and analog reconstruction filter are supposed to remove all images and noise from the analog output, leaving just accurate reconstruction of the original analog signal.

 

Because digital sampling rate cannot be infinite, you always need analog reconstruction filter that should effectively make the sampling rate disappear from the analog signal. Oversampling/upsampling exists for the purpose of making this job easier for the analog filter and helps it perform better. It is also important part with high speed noise shapers.

 

2 hours ago, fds said:

Looking at the Stereophile measurements of Mays NOS mode, one sees a somewhat early roll off.

 

In NOS mode the roll-off depends on the input sampling rate, since in OS mode the conversion stage runs at fixed rate. If you run a NOS DAC at 44.1k or similar low rate, you will certainly have early roll-off. This is mathematical nature of the conversion itself. It mostly rolls off between 1/4th and half of the sampling rate. If you use NOS mode and input for example 768k, this roll-off is between 192 and 384 kHz.

 

Even with oversampling there can be still a bit of roll-off depending on the analog filter design, where it's -3 dB corner frequency has been set at design time. Higher the oversampling ratio, higher this corner frequency can be designed and less negative implications such as early roll-off and phase-shift there are in the audio band.

 

2 hours ago, fds said:

Wondering how much HQP could help here, e.g., by upsampling to 706.5kHz or to DSD256 or higher?

 

Of course it does, because the system is not "NOS" anymore. The OS is in HQPlayer. So the frequency response is same or better than in OS mode.

 

2 hours ago, fds said:

Also since the Stereophile measurements of JA do not consider the DSD DAC inside the May, I am wondering how the corresponding results would look like? Any experience maybe from measurements of your Holo Audio Spring (2) that you would be willing to share with us?

 

My guess is that the results look pretty similar to the Spring 2 ones I've published in the corresponding thread.

 

[Miska]

4 hours ago, Khalondir said:

Is there any benefit in using the I2S input instead of usb in holo may? Thinking if the d/d denafrips gaia can bring some gain along with the holo audio may dac.

 

server > usb > denafrips gaia > I2S > Holo May

 

Problem with all those I2S interfaces is that clocking is wrong way. From the I2S source, while it should be the opposite direction, from DAC towards the source...

 

[Miska]

3 hours ago, GoldenOne said:

With pll off, usb and i2s are neck and neck on the may, couldn't pick a clear winner but usb is my go to due to pll locking instantly once I turn it on. 

 

There shouldn't be any PLL involved when using asynchronous USB....

 

[Miska]

11 minutes ago, GoldenOne said:

I don't believe there is no. Sorry, probably could have worded that post better. 

 

What I meant is that the pll puts basically any digital source on equal footing with the may. Be it an expensive streamer or basic optical from my pc. 

 

USB sounds the same regardless of if the pll is on or not. 

 

USB doesn't need any PLL because the clocking is at DAC side. PLL is only needed when DAC needs to regenerate clocks from external source (S/PDIF, AES or optionally I2S). Which is not the case with USB where clock is solely at the DAC side...

 

There's not much point in DAC regenerating clock from it's own clock with PLL. It would only degrade performance.

 

[Miska]

43 minutes ago, Flextreme said:

Is there a objective way to find out (other than subjective listening) which USB port gives best results?

 

I prefer to measure DACs analog output, that is what tells the real story of what you are listening to.

 

And yeah, annoyingly there's quite a lot of variation from product to product depending on the USB interface and source implementation...

 

[Miska]

6 hours ago, ThenewGearPPK said:

NOT sure what the difference between Filter 1x & Filter NX is? I just use the same filters on both of them.

 

1x is for "1x sampling rates", meaning source rates below 50 kHz. Nx is for source rates above 50 kHz. So like separate selection for hires and non-hires content.

 

6 hours ago, ThenewGearPPK said:

For dither, almost all the options are for specific sample rates. i.e. NS4 is for 88.2kHz & LNS15 is for 768kHz etc..

TPDF is the standard for all type.

Gauss1 is a high quality dither that should be used for sample rates below 96kHz.

 

Since May can do high PCM rates, LNS15, NS9 and NS5 are suitable options for it.

 

[Miska]

14 minutes ago, Lightwave said:

What is taps? How does the # of taps affect the upsampling?

 

It is one technical aspect, but pretty useless number alone.

 

14 minutes ago, Lightwave said:

And is there filters that I should avoid using for R2R DAC like the May?

 

No, choice of filters is up to you. But I would recommend to get started with the default poly-sinc-ext2

 

16 minutes ago, Lightwave said:

 I see ASDM5 (which is the default for SDM/DSD is delta-sigma modulator. Is that less "compatible" with R2R DAC?

 

May is like two DACs in one box. It has great SDM/DSD section too. I always run my Spring 1 and Spring 2 in DSD mode, at DSD256 using ASDM7EC modulator.

 

So you can run it at maximum PCM rate possible, or DSD256.

 

But note that with DSD the output level is half (6 dB) lower compared to PCM mode. This is normal and due to technical reasons. Just something to take into account if you want to compare the two sides.

 

[Miska]

45 minutes ago, Lightwave said:

 

@Miska Thank you for your help! A couple more questions if you don't mind. What's the difference between 32k x512, 44.1k x512 and 48k x512 bit rate limit? My guess is that if I use 48k x512 as the bit rate limit, 44.1k PCM source should be resampled to 44.k x512 DSD, right? So, for the May DAC, I should set it to 48k x512 (or 48k x256 if I want a lower bit rate)?

 

If you want DSD256 output, set rate limit to 48k x256. For May you can also enable 48k DSD.

 

With default settings with adaptive output rate grayed, for the filters that can, you'd get 48k x256 output also for 44.1k sources and for filters that can't you'd get 44.1k x256 output. You can make output rate follow input rate by checking the adaptive output rate. Since it is possible to use it with May.

 

51 minutes ago, Lightwave said:

Since DSD has half output level, should I set the Vol Min to -6db and Vol Max to -3db?  I'm connecting the Mac to a pre-amp. (Not directly to a power amp like many do.)

 

It is unrelated, the half output level only affects your preamp volume setting if you want to compare PCM vs DSD outputs, you'd need to turn up volume for DSD mode.

 

You can use for example that range, and check that you have volume control initially set to -3 dBFS.

 

[Miska]

10 minutes ago, Lightwave said:

This may or may not be a May DAC related issue. My apology if this is specific to the HQPlayer NAA.

 

@Miska Does HQPlayer NAA only support PCM?  I have one Mac running HQPlayer Desktop and the other running networkaudiod NAA, but all DSD got converted to PCM before streaming to the NAA node. It seems like that HQPlayer Desktop only see the networkaudiod node as a PCM device. And the console output of networkaudiod seems to suggest that CoreAudio can only see the May DAC as a PCM device. Will a Rasperry Pi running Linux with a proper USB driver be able to see the May DAC as DSD device?

 

 

That should work since you have DoP enabled. On macOS there's only PCM interface between the OS and DAC because CoreAudio doesn't support DSD.

 

On your NAA Mac, make sure the May is NOT selected as default audio output device for the OS or any other running application.

 

Do you have some specific reason to use such a NAA here instead of direct USB connection?

 

[Miska]

43 minutes ago, Lightwave said:

Another I observed is that the bit depth set to 16-bit when streamed to the NAA. Is this a CoreAudio issue?  I've set the bit to 20 but it still got converted down to 16-bit when streaming to NAA.

 

Ahh, yeah, could be that CoreAudio driver has changed a little if you have Big Sur. That would explain why DoP is not working, it needs 24-bit (DAC Bits set to 20 doesn't affect this).

 

[Miska]

10 minutes ago, Lightwave said:

The Mac running networkaudiod NAA is Catalina, not Big Sur. The Mac running HQPlayer Desktop is Big Sur. If I get a Sonore microRendu or Raspberry Pi 4 as NAA, would I get the 16-bit problem?

 

You can try with RasPi4 NAA for starters, it is pretty inexpensive and you can boot my OS image on so it is very straightforward to setup. You just write the image on microSD and boot it up. Unless this is some USB firmware feature in your device...

 

But in order to enable full capabilities of May, you need Intel based NAA. Assuming it has the USB firmware allowing 1.5M PCM and DSD1024.

 

[Miska]

2 hours ago, Lightwave said:

Do you have a list of known RasPi4 NAA that are known to work with the full capabilities of May?

 

It won't... Should work up to 705.6/768k PCM and DSD512 though.

 

There's not much to list, there are only couple of different memory configurations of RasPi4 hardware. (2, 4 and 8 GB). 4 GB is well enough for NAA use.

 

2 hours ago, Lightwave said:

Not sure how to go about finding out if the USB firmware on a RasPi4 is compatible.

 

No, I mean May's USB firmware. Trying it out will tell.

 

[Miska]

2 hours ago, mrkoven said:

What's the best way to set up HQP with an Innuos server? Is it possible to retain the Innuos as Roon core and player while outsourcing the HQP processing to a more powerful processor? In other words can the Innuos roon core point to HQP elsewhere, but still output to the DAC as a player via USB?  Or does the HQP host need to connect directly into the DAC?

 

Yes, you can split Roon Core and HQPlayer such way. Sometimes it is beneficial to do so.

 

With HQPlayer, DAC is accessed by HQPlayer, not anymore seen by Roon. It can be connected either to HQPlayer host, or to HQPlayer NAA network endpoint. If the Innuos server can run the NAA software module, the DAC can remain connected to it.

 

[Miska]

6 hours ago, lpost said:

Based on this should we still set DAC bits to 20 or should it be set to 21 or 22 or 23?

 

20... You can actually see that from their linearity sweep measurement Fig 11, where the linearity begins to deviate at -120 dB which is equal to 20 bits:

 

Setting DAC Bits to 20 won't affect noise floor when combined with LNS15, NS9 or NS5 noise shaper and high (>= 352.8k) sampling rates. This linearizes the D/A conversion removing distortion in low level signals while maintaining same noise floor.

 

Note that Stereophile measurements are with flat TPDF dither, not with upsampling combined with noise shapers.

 

You can even set DAC Bits to 16 when using the highest sampling rates and still the noise floor won't change as long as you use suitable noise shaper. In fact, setting DAC Bits to 20 and using for example LNS15 shaper at higher rates, digital noise floor at audio band is lower than it would be with DAC Bits set to 24 and TPDF dither, and without distortions. This is whole point of noise shapers; allows you to use linear region of the DAC while keeping or improving the dynamic range. But for the magic to happen you need higher sampling rates, noise shaper, and then the DAC's analog reconstruction filter does the rest.

 

Here is example from Spring 2 (these were measured before LNS15 existed)...

 

1 kHz tone at -120 dBFS, with DAC Bits set to 24 (at 1.4112M sampling rate):

 

 

Here with DAC Bits set to 20 and NS5 noise shaper at 1.4112M sampling rate:

 

Here again the same, but DAC Bits set to 16, NS9 noise shaper, 1.4112M sampling rate, very little noise floor increase at 90 kHz:

 

With NS5 shaper and DAC Bits set to 16, at 1.4112M sampling rate, you would begin to see noise floor increase above 60 kHz. So if you'd drop the sampling rate to 705.6k instead, the noise floor rise with 16 bits and this relaxed noise shaper, the noise floor rise would begin at 30 kHz instead - still above audio band.

 

With DAC Bits set to 20 and NS9 shaper, at 705.6k sampling rate, you still have flat, unchanged noise floor.