HOLO Audio Spring DAC - R2R DSD512

[GoldenOne]

On 3/30/2020 at 8:41 AM, Bimmer100 said:

 

serene will be totl or flagship preamp/headamp. Same form factor as spring/spring2/May dac chassis and designed to pair perfectly with these products. It will be an absolutely epic preamp/headamp with tech similar to azure yet many improvements of course. 
 

Are you able to confirm if the serene will include a headphone amp? Your msg said it will, but the unit that WolfX received to measure didn't seem to have any headphone outs (but did have all 8 modules as in the azure).

Hoping it will have a headphone amp! So excited to get one.

[GoldenOne]

1 hour ago, Bimmer100 said:

There will be two versions. One is preamp serene, the other will be headamp serene.

there is too much to put both into one chassis. The headamp version is not ready yet.

Ooh interesting. 
Does that mean that the headamp version won't additionally function as a preamp?

Also, are you able to give any information as to pricing? My wallet is trembling but I am so ready!

(Also goodness, a quadruple holo-audio stack would look fantastic)

[GoldenOne]

2 minutes ago, Miska said:

OK, so I got my Spring 3 L2 DAC today and have performed some initial testing.

 

Performance is very nice overall! I also like the new standby mode with display off.

 

As before, output level is 6 dB lower with DSD sources just like before. This is sort of natural and leaves headroom for over-modulated sections that can momentarily reach +3 dB level.

 

Some measurements here...

 

First linearity check as before, 1 kHz -120 dBFS sine here, TPDF dithered to 24-bit resolution:

Same tone, but now LNS15 noise-shaped to 20-bit:

As expected, linearity error causing distortion gets corrected.

 

Then first PCM output to 705.6k, LNS15 noise shaped to 20-bit.

 

1 kHz THD test:

 

19+20 kHz IMD test:

 

Multitone-test (analyzer left to same settings, I should have used a custom setting for these):

 

J-test24:

 

Then the same tests with DSD256 by ASDM7EC modulator.

 

1 kHz THD test:

19+20 kHz IMD test:

 

Multitone-test:

 

J-test24:

Nice! 

For the j-test is that a full scale j-test? 

[GoldenOne]

1 minute ago, Zauurx said:

I couldn't find any measurement on my Musician Pegasus (Denafrips like) but I think the scheme produces the same result: configured in 20 bits and a maximum SQ in SDM 256 gauss-xla ASDM7EC (or 5EC... my core i5 does the maximum but...). NOS setting.

Denafrips/pegasus don't actually have a proper dsd converter as far as I'm aware. It converts to pcm and converts on the main ladder

[GoldenOne]

3 minutes ago, Zauurx said:

I suspect that the schema is not totally identical... not really the same price. 

Is it the "Proprietory R2R + DSD architecture" ?
But there is DSD decoding (6bits), otherwise I'd better stick to PCM 1.412 direct ?
The PCM seems more dense but the DSD more dynamic (EC filter ?).

Denafrips was very evasive about it when I asked them. Unfortunately it seems they don't like talking about any of the more inner details of how their dacs work. 

 

But there's no separate dsd converter visible on the board. It seems it just converts to very high sample rate 6-bit pcm and converts this on the main ladder. 

But I'll check when I next get a denafrips dac in. 

 

Rockna to their credit was very open about how their dac does it. It converts to 768khz 24bit pcm and plays on the main ladder

[GoldenOne]

1 minute ago, Miska said:

 

It doesn't mean that DSD would be converted by R2R network, but instead a separate SDM style network. If you want to see schematics of example implementation, check out my DSC1 design. It uses 32-resistor network to convert DSD to analog (producing 33 unique analog levels).

 

Are you aware if this is what denafrips is indeed doing? 

 

Im hoping to get another one in to test but until then can't really go off anything other than the pcb which doesn't seem to show anything that would do this. Whereas on the holo etc its clearly visible

[GoldenOne]

Posted my full review of the spring 3 with some comparisons to May and Spring 2 :) :

 

[GoldenOne]

12 minutes ago, DuckToller said:

Thank you very much for this review @GoldenOne! Kudos!

So you are quite enthusuatic about the Spring with Pre.
Which - for me - seems to be a good point to explorate this product further, especially because with NOS, DSD, External Oversampling plus Pre-Amp  this DAC may fit my key demands I would lay down for a future (endgame/reference) DAC.
Am I mistaken, when I understood that you reviewed the L3 KTE version with Pre-Amp module ?
You mentioned the Titanis module, thus without reference if it is 2.0 ?
However, at another point you were noting 2.5 grand as pricepoint, which made me wonder ...
Could you please clarify?

Thanks in advance
DT

I am!
For the money it's a crazy good option IMO.

L1/L2 spring 3 have the 'normal' usb card. Still has galvanic isolation, but no titanis chip and not as advanced. How much of a difference that makes I cannot say as I've not compared the two.

I said $2500 cause L2 is $2500. (L1 being cheaper and KTE being a bit more).
Having tried the KTE and L2 may though, there is a very subtle difference but not enough to change anything I said in my may video for example.

The spring 3 l1/2/3 have some slight component upgrades but they're the same design, same dac. It's not going to make it a drastically better/worse product or anything.
 

[GoldenOne]

10 minutes ago, Cogito said:

@GoldenOne

 

Thanks for the excellent review of Spring 3.

 

Couple of small issues:

 

1. DSP volume attenuation not only reduces the headroom, but also reduces the resolution of the source material. 

Redbook 16 bit PCM has the headroom of 16 x 6 = 96 db

The max value that can be represented by a 16bit binary number is 65,536. 

DAC's reference voltage (5V?) is divided into 65,536 equal parts to represents all possible values of 16bit binary (PCM sample). Each step is 0.015mv.

 

If the volume is attenuated by 6dB in the digital domain reduces the bit depth to 15.

Redbook 15 bit PCM has the headroom of 16 x 6 = 90 db

The max value that can be represented by a 15bit binary number is 32768. 

DAC's reference voltage (5V?) is divided into 32767 equal parts to represents all possible values of 16bit binary (PCM sample). Each step is 0.03mv. This is much coarser representation that the original redbook source. That result is, you would loose microdynamics and resolution. For this reason audiophiles never use DSP for volume attenuation.

 

2. Preamp compensates for the headroom loss due to DSP attenuation.

This is not true. Preamp has two purposes:

1. Volume attenuation in the analog domain with potentionmeters

2. Analog signal amplification

 

Preamps and Power amps do not change the headroom present in the source material. Preamp can increase the volume at with the audio is played but it does not change the headroom.  If the DSP attenuation reduces the resolution from 16bit to 14bit, preamp can compensate for the loss of volume but the headroom will still be that of of 14bit.

 

 

 

 

 

 

The issue is that whilst most music might be 16 bit, the DAC in most modern examples will have THD, SNR and AES17 Dynamic range higher than that, and oversampling/dsp volume control will likely be being done at a higher bit depth than the source material too. Which is why 96dB isn't enough to call something 'perfect'.

Dithering and noise shaping also have enormous effects

Additionally, I understand that the preamp doesn't affect the source material, but it is part of the playback chain. If you have a DAC with say 120dB dynamic range, and you attenuate 10dB in DSP, you will lose 10dB dynamic range (in most cases, some dacs with non-linear THD+N/level performance might be slightly different).

But if you have say a preamp with 140dB dynamic range. Now, you could attenuate by 10dB, and it's entirely likely that the noise floor of the preamp is still well below that of the DAC. Meaning you might still have the full 120dB dynamic range as you'd be limited by the DAC not the preamp.

I did a few tests with this on the spring 3 and found that I got about 5dB extra headroom before dynamic range was affected (and the spring 3 has exceptionally high dynamic range anyway so could attenuate further before dynamic range reaches the level of many other similarly priced dacs).

The Holo Serene is probably the most drastic example of this due to the quite frankly ridiculous dynamic range it has.
I'll take it out of my chain and do some practical demonstrations of the above in the next few days. Need to wait for a new headphone amp to arrive then I can take the serene out and put it on the measurement bench

[GoldenOne]

5 minutes ago, Cogito said:

Different issues. The specs like THD, SNR provide by DAC measure only the artifacts introduced by the DAC. DACs are not enhancing the digital source.

You can convert 16 bit source info to 24 bit to apply DSP with less loss of accuracy/dynamic range.
There is no reason you have to keep 16 bit as 16 bit (and internally the vast majority of dacs will not)

 

 

6 minutes ago, Cogito said:

You are explaining  passive attenuators, which audiophiles stay away from as they compromise micro dynamics.

In a typical active preamp, about 90% of the original signal is burned off and amplified again. In our original example, 14bit PCM has a dynamic range has a noise floor of -84dB.  In the preamp and amp, all analog signals, which includes noise, is also amplified.  Preamp having lower noise floor only means preamp is not adding noise to the signal.

No I'm not. The serene is an active preamp.
https://www.l7audiolab.com/f/measurements-of-holoaudio-serene-preamplifierpre-retail/

 

As mentioned, I'll do a practical demonstration showing the effect of dynamic range in a preamp in the next few days.

 

[GoldenOne]

10 minutes ago, matthias said:

 

AFAIK, it is a preamp with no gain, a pure buffer one but not a passive one.

 

Matt

It has upto 6dB gain

[GoldenOne]

4 minutes ago, bodiebill said:

 

So you have experience with the Holo preamp module, the Serene as well as The Truth?

Could you summarize their SQ differences?

 

That would be helpful as the MiniDSP in my setup comes very close to The Truth.

I can't compare spring 3 + serene to spring 3 w/ integrated preamp as Id have to have the preamp and non-preamp versions of the spring 3 to do so. 

 

The serene is likely going to be better, but it's also a much more costly option. 

The serene makes more sense if you're getting a may. Otherwise I'd go for the spring 3 preamp. 

 

The hardware preamp does provide some advantages as with dsp vol control you start losing max THD+N as soon as you attenuate. Whereas with the hardware preamp you do not until you start getting limited by the noise floor. 

 

I do not like dsp vol control other than for small (less than 6dB ideally) adjustments. Anything beyond that I'd use a good preamp. 

 

[GoldenOne]

7 hours ago, caesee said:

I will get my new dac holo spring 3 L3 KTE version tomorrow , new dac will support dsd1024 and pcm 1.5m

But hqp only support dsd 256 7ecv2.

Will pcm 1.5m 7ecv2 will better than dsd 256?

The main advantage with dsd for many regular dacs is that you can bypass and effectively 'replace' the modulator in the delta sigma dac itself with hqplayer's high performance once. 

 

With an r2r dac like the spring, there is no modulation needed as it can convert natively at 24 bit (cause its r2r). 

 

Whether you end up preferring dsd or pcm on it will come down to personal preference. 

In some areas dsd is objectively better but subjectively I prefer pcm on holo

[GoldenOne]

3 minutes ago, KenMoreira said:

 

 

Who buys an r2r resistor ladder dac than uses the chip based dsd as their preferred option. 

 

Just get the d90se if that's the way you prefer to listen n save significant money and have better dsd 

 

( my opinion) 

The dsd converter in the spring (and may) isn't a chip it's discrete. 

 

It has separate discrete converters for both dsd and pcm. 

 

Also the d90se is ess so it wouldn't actually convert dsd natively anyway. Ess dacs convert to pcm internally as they utilise a lot of dsp for stuff like their THD compensation

[GoldenOne]

57 minutes ago, KenMoreira said:

I had read the spring uses a proprietary chip for dsd. And their is only one ladder, it's for pcm. 

 

The may having balanced ladder dacs for dsd and pcm both 

No the spring 2 and spring 3 both have discrete DSD converters too.

[GoldenOne]

1 minute ago, Bertel said:

 

Don't think that this is correct, at least not for the ES9038PRO, the data sheet talks about Native DSD Format and how DSD (when not in DoP) is bypassing the FIR filter stage. @Miska when I recently asked that's why you recommended using DSD over PCM in HQPlayer with the D90ES, right?

It could bypass the FIR filter stage by just converting to a high PCM sample rate, rather than being converted to redbook and then 'starting from scratch' etc.

It might (and is likely) doing something different internally to being fed PCM but if it was converting the DSD natively then none of the THD compensation would be working and so we'd see notable performance differences between PCM and DSD with the 9038 which in my testing has not happened.

It's a bit tricky to know for sure though as ESS is understandably fairly tight-lipped about the inner workings of their DACs

[GoldenOne]

1 hour ago, mrwiggles said:

Setting up new headphone rig using Ferrum stack amp and ps. Have pretty much settled on Spring 3 DAC for this setup but wondering if I should order with the preamp option for volume control. I'd read somewhere that you can switch the Orr to 'amp only' and get better dynamics, but it requires a preamp obviously in that case. Wondering if @GoldenOne has a position on this. Or anyone else with opinions would be most appreciated.

The OOR in my opinion does sound ever so slightly better in bypass mode than not in bypass (and actually measures a bit better too). 

 

If the budget allows, I'd get the spring 3 preamp and put oor in bypass. 

[GoldenOne]

On 2/12/2022 at 3:46 AM, Miska said:

 

Why not just use digital volume control instead? No analog distortions... And OOR has gain control, so you can adjust suitable gain for your headphones.

 

The Spring 3 preamp seemingly has higher dynamic range than the DAC itself as I was able to attenuate I think about 15dB before the AES17 dynamic range measurement showed a reduction (I can't remember the exact figures unfortunately but it was a notable amount).

Whereas DSP vol control reduced dynamic range immediately.

Personally I prefer to do vol control on whatever device in the chain has the highest dynamic range to best preserve overall dynamic range of the chain

[GoldenOne]

2 minutes ago, Miska said:

How about distortion and noise of the preamp section itself, compared to without the preamp in chain at all?

I've not tried the Spring 3 without pre so can't comment on that unfortunately. The spring 3's pre can't be taken out so I wasn't able to compare with/without it.

Serene though is about as clean as it gets and you have more dynamic range than the DAC itself:
4v in 4v out:



4v in -60dB attenuation:



So I'd imagine the spring 3 preamp probably has pretty excellent performance too. 
 

 

5 minutes ago, Miska said:

With DSP, analog noise floor of the DAC is only limiting factor and it is (or at least should be) naturally static.

It should be, but it also depends on what additional processing might be done inside the DAC or whether the distortion profile remains the same with level which not all DACs do.
A Rockna Wavedream for example sounds quite different when you put 20dB DSP headroom on cause the distortion profile vs output level is a bit odd:


Though of course with the majority of DACs this shouldn't be a factor, but is something worth mentioning in particular on R2R DACs which usually have more of this sort of effect.


 

[GoldenOne]

11 minutes ago, Miska said:

Surprising amount of mains hum, although this could be related to connections between devices.

This was with -60dB attenuation, so the 0dB point here is -60dB relative to 4v

[GoldenOne]

12 minutes ago, Miska said:

This looks exactly kind of DAC that could be linearized with suitable DSP, just like Holo too.

It'd be pretty cool to see something like that done.
What sort of modelling etc might be needed to achieve this?