S.MS.L D300 Native DSD DAC Review with Measurements

[Miska]

3 hours ago, Andrew Allen Ballew said:

This filter is a digital FIR filter

 

Analog FIR filter... 😉

 

3 hours ago, Andrew Allen Ballew said:

1 bit DSD signal is delayed by a clock cycle, for as many streams in parallel as you need for your moving average filter.  Some will argue for shorter filters, other for longer.  Same bugaboos of impulse response and ringing apply in this argument.  

 

Please note that these filters don't have any ringing at all. And they are totally linear phase as well.

 

[Miska]

1 hour ago, bogi said:

Maybe @Miska could chime in with his opinion, if newer NAA images contain any changes against older regarding ability of 48k DSD rates playback on XMOS chips.

 

No, that is entirely up to the USB firmware inside DAC.

 

1 hour ago, bogi said:

It delays stopping of HQPlayer engine after stop button (not pause) is pressed in HQPlayer by playing digital silence for a specified time. I don't know if it can improve something on your issue but it is easy to try.

 

If one uses Roon for example, it helps avoiding clicks/pops on DAC start/stop, since Roon is frequently stopping and starting HQPlayer playback and this turns such into "fake stop". So the DAC keeps running. And if you then use fixed output rate the result is that you get far less DAC stops and starts. And also actions are quicker, since the player engine takes some time to get started/stopped.

 

[Miska]

12 minutes ago, satshanti said:

Bogi mentioned that if I'd run Windows on the RPi4 and install SMSL's ASIO driver, that might work, but is that true?

 

No, it is not about the driver. But about the DAC firmware...

 

And you cannot run Windows on RPi4 nor install SMSL ASIO driver there.

 

12 minutes ago, satshanti said:

maybe not surprising considering it's possibly more accurate to reconstruct waves to multiples, rather than from 48k to 44.1k.

 

No, it is not any more accurate. At the moment, poly-sinc-gauss group is most accurate and it can also convert between rate families.

 

[Miska]

6 hours ago, bogi said:

Pi4 can run Windows https://www.google.com/search?q=Raspberry+Pi+4++windows

 

At least practically nobody is making drivers for Windows on arm64. So far I have not seen a single DAC manufacturer to ship drivers for such. I may make arm64 Windows build of HQPlayer at some point, but it is mostly useful for running Client on those arm64 based Surface tablets...

 

[Miska]

2 hours ago, bogi said:

I wanted to write:
Maybe a different USB DFU fw version??
his USB DFU fw version is 1.3

 

Usually the issue is USB controller firmware. In many DACs, that same controller is also responsible for dealing with the DAC chip configuration in general.

 

Sometimes it is the DAC chip itself and it's specific configuration, like with ESS chip.

[Miska]

20 minutes ago, satshanti said:

They might show the correct data, but the DAC might still play music at a slower speed, at 44.1/48 = 0.91875x to be exact. With some music you barely notice, especially if you've never heard before how it should sound. At first, it seems to work, but it's not how it should sound. Just a thought... 

 

Yes, for example Holo Spring (1) was doing that. Thus best way to verify is to play for example 1 kHz tone from 48/24 source and check with spectrum analyzer from the analog outputs that the output is actually 1 kHz...

 

[Miska]

26 minutes ago, pkane2001 said:

I've run into this testing SMSL DO100 while using DSD. Everything looks great, except for the actual frequency of the test tone at the output not being 1kHz when using 48k-based rates. Instead it's at 1kHz * 44.1 / 48 = 0.918kHz:

 

 

 

Yes, quite typical with stock reference code XMOS USB Audio Class implementation. It fails to switch clocks and instead always sticks to the 44.1 family clock with DSD inputs. I kind of fail to understand who bothered to actually implement the extra code to do that with DSD sources, since it switches clocks

 

26 minutes ago, pkane2001 said:

EDIT: I lied about all else looking great. This is the difference between 44.1k and 48k with DO100 - check out the noise floor difference:

 

Many ESS chips do that when the ASRC is active. I think the chip would need to have ASRC disabled which, IIRC, requires to run the chip in so called "128fs clock" mode with synchronous clocks. Instead for example typical fixed 100 MHz clock.

 

Same also causes the chip go nuts with certain input data (PCM inputs) that causes huge noise peak centered around 1 MHz in the output.

 

So you have two separate issues here. One with the XMOS USB implementation and another one with the buggy DAC chip.

 

That noise slope lift from 40 kHz on doesn't belong to DSD256 data. It is the ADC? Or maybe it's the ESS chip in question, not sure.

 

[Miska]

7 minutes ago, pkane2001 said:

Same ADC for both (RME ADI-2 Pro FS at 768k), so I doubt ADC is at fault. It knows nothing about DAC DSD rates.

 

ADI-2 Pro has noise floor rising a bit before 100 kHz, but I don't remember how it looks like exactly. You can try to check how it looks like with input open. I just remember the noise floor looks the same with both 705.6k PCM and DSD256 recordings. It is certainly not flat beyond 100 kHz.

 

7 minutes ago, pkane2001 said:

Seems to be something in the implementation of DO100 or the ESS chip, like you said. In any case, 48k rates appear very broken with this DAC.

 

Those mobile Q2M models are less tricky to deal with.

For example with SMSL DL200 (ES9039Q2M):

 

Main difference here is that the images around 44.1k are naturally gone in this 705.6k upsample.

 

 

So you get very clean output at DSD256. But then DSD512 seems to get decimated to lower rate and as result it triggers again those ESS modulator spurious tones and some noise aliasing at the top of the frequency band.

 

[Miska]

There is certainly reason why HQPlayer has that "48k DSD" check box with comment "Only enable this if you are absolutely sure your DAC works correctly at 48k DSD multiples!" to enable DSD output at 48k multiples!

 

[Miska]

1 hour ago, pkane2001 said:

I'm not as concerned with noise floor above 100kHz as the noise floor being 20dB higher in the audible range :) 

 

You mean 44.1k-base vs 48k-base? For 44.1k-base the different rates should give same audio band noise floor (like they do for example on above DL200, and from earlier experience I know also for ES9038Q2M such as on Pro-Ject PreBox S2 Digital where also DSD512 works fine and which is synchronously clocked).

 

[Miska]

19 minutes ago, pkane2001 said:

I was referring to the DO100 256x48k noise floor being 20dB higher than with 128x48k or 256x44.1k. I believe that rate is not recognized correctly by that convertor.

 

I think 256x48k exceeds the ASRC capabilities and then it goes nuts. ES9028PRO for example does the same in certain configurations.

 

It is also possible to trigger similar crazy behaviour with certain PCM inputs in similar configuration.

 

With some DACs it is possible to work around the USB input deficiencies if they have I2S inputs, by using external DDC. For example I did that with SMSL M400 by running it from I2S output of Holo Red.

 

[Miska]

21 hours ago, bogi said:

Thanks, at least it is now clear. I think they are not aware of a feature, because something like 48k based DSD is not known to them.

 

Be it Topping or SMSL, they don't really seem to test much else as inputs than 44.1k PCM, or maybe up to 192k PCM. Not higher PCM or DSD. Only reason I can think of why anything else seems to be completely random hit or miss when it comes to some feature actually working correctly or not. Maybe that is part of the reason for the low price as well...

 

[Miska]

29 minutes ago, bogi said:

but take reference boards with reference software and only change the names there

 

29 minutes ago, bogi said:

But many DACs from western companies don't support it either

 

For this same reason. Because the original (very buggy) XMOS reference code doesn't seem to support it. Thus it gets propagated into various devices. iFi for example has done a lot of their own development on the XMOS code to fix various things. And the person who designed Pro-Ject Pre Box S2 Digital had quite a lot of harsh words about XMOS experience overall.

 

XMOS can be made to behave properly, but it requires some notable effort and firmware development capabilities.

 

Lot of companies have long since moved from XMOS to STM32 controllers for implementing USB interface. Maybe not as cheap, but more powerful and better quality development tools.

 

37 minutes ago, bogi said:

I perceive some difference between DSD256 and DSD512. At DSD256, instruments may sound more expressive. Maybe lower noise floor?

 

Yes, IIRC that is the case. And remember to set the DSD filter to lowest frequency which will give you lowest distortion.