iFi audio Ships Pro iDSD DAC/Streamer

[asdf1000]

11 minutes ago, opus101 said:

 

Whilst not directly audible, it has strong potential to cause audible effects. No downstream circuitry will have zero IMD and IMD performance for all amplifiers falls with increasing frequency.

 

But as mentioned further up, does up-sampling to DSD256 or DSD512 make this a non-issue?

 

[opus101]

2 minutes ago, Em2016 said:

But as mentioned further up, does up-sampling to DSD256 or DSD512 make this a non-issue?

 

I'm not knowledgeable about DSD256 etc but I would surmise that higher oversampling moves the noise further away from the audio band. So given a textbook-operating filter at the output, there will be less noise in practice with DSD256 than DSD64. The caveat being of course that no filters in practice are textbook, they will themselves contribute IMD and have non-monotonic stop bands.

[asdf1000]

2 hours ago, opus101 said:

I'm not knowledgeable about DSD256 etc but I would surmise that higher oversampling moves the noise further away from the audio band. So given a textbook-operating filter at the output, there will be less noise in practice with DSD256 than DSD64. The caveat being of course that no filters in practice are textbook, they will themselves contribute IMD and have non-monotonic stop bands.

 

Jussi's @Miska measurements of the micro iDSD seem to show improvements with DSD256 and DSD512 up-sampling.

 

https://www.computeraudiophile.com/blogs/entry/428-ifi-idsd-micro-measurements/

 

 

Once question for those much smarter than me, what is that 'hump' that you see at DSD256 and is much smaller at DSD512?

 

The hump between 0 and 1MHz for DSD256 - and why do you want that hump to get lower in amplitude? 

 

DSD256:

 

 

DSD512:

 

[mansr]

5 hours ago, Em2016 said:

Once question for those much smarter than me, what is that 'hump' that you see at DSD256 and is much smaller at DSD512?

 

The hump between 0 and 1MHz for DSD256 - and why do you want that hump to get lower in amplitude? 

 

DSD256:

 

 

DSD512:

The hump is the residual modulator noise after the DAC filter. It is smaller with DSD512 because there the noise starts rising at a higher frequency and the DAC uses the same lowpass filter for both. You want to get rid of it because, as discussed earlier, high-frequency noise can be detrimental to amp and speaker performance.

[mansr]

8 hours ago, Brinkman Ship said:

It has never been audible to me.

It isn't with a competent DAC. Sadly, not all DACs fit that description.

[Miska]

And note those plots are with "peak hold", so it is maximum peak spectrum of 0 - 22.05 kHz sweep, not averaged or anything.

 

Those older measurements are with the original micro iDSD. The newer Black Label version improved especially DSD512 performance, likely because of better analog stage.

 

[asdf1000]

1 hour ago, mansr said:

The hump is the residual modulator noise after the DAC filter. It is smaller with DSD512 because there the noise starts rising at a higher frequency and the DAC uses the same lowpass filter for both.

 

Thanks. Interesting that there's still a hump at DSD256 (~11Mhz), so so far above the 80kHz analogue filter.

 

But Mr DSD says the ideal is between DSD128 and DSD256:

 

http://positive-feedback.com/audio-discourse/raising-the-sample-rate-of-dsd-is-there-a-sweet-spot/

 

"With a clear improvement from doubling the sample rate of single DSD, it seems natural and, of course, tempting to quadruple the sample rate or go even higher. We should expect the same or similar improvement as from simply doubling the sampling rate again, right?

 

Not so fast! It turns out there are physical limitations such as electronic component speeds, finite clock slopes, etc., that limit the amount of performance gain we could expect from raising the sample rate above a certain threshold. Such limitations are indeed starting to affect performance with quad DSD in D/A converters. Side effects in the form of audible noise and distortion are creeping into our audio band with quad DSD, greatly overshadowing the small benefit of the noise shaper curve starting at 80kHz. This, of course, is only apparent in D/A converters that convert the DSD signal directly into analog without any conversion to PCM or other digital filtering. To avoid this the quad DSD signal would have to be low-pass filtered and/or converted to PCM before converting it to analog."

[Miska]

1 hour ago, Em2016 said:

Thanks. Interesting that there's still a hump at DSD256 (~11Mhz), so so far above the 80kHz analogue filter.

 

Those analog filters are not so aggressive... However level of the noise hump is very low, especially compared to level of images around multiples of 352.8 kHz you can see in the plots with PCM inputs. And that hump is completely uncorrelated while those images with PCM input are fully correlated.

 

The hump naturally also depends on modulator design, how much and how aggressively it prefers to shovel noise from audio band to the ultrasonic range.

 

Any delta-sigma DAC (all modern DACs) have a noise hump by definition. How much of it is left in the output depends on modulator output rate, design of the conversion stage and the used analog filter following the conversion stage.

 

For comparison, if you look same sweep at DSD512 from Holo Audio Spring DAC:

 

There is no noise hump.

 

By looking at wide-band sweep output you get some sort of idea of frequency response of the conversion and analog filter stages. I think I used 32-bit 1.4 MHz sampling rate PCM as source, so the end is partially limited by the sampling rate - so the sweep goes from 0 to 705.6 kHz.

 

 

1 hour ago, Em2016 said:

But Mr DSD says the ideal is between DSD128 and DSD256:

 

http://positive-feedback.com/audio-discourse/raising-the-sample-rate-of-dsd-is-there-a-sweet-spot/

 

"With a clear improvement from doubling the sample rate of single DSD, it seems natural and, of course, tempting to quadruple the sample rate or go even higher. We should expect the same or similar improvement as from simply doubling the sampling rate again, right?

 

Not so fast! It turns out there are physical limitations such as electronic component speeds, finite clock slopes, etc., that limit the amount of performance gain we could expect from raising the sample rate above a certain threshold. Such limitations are indeed starting to affect performance with quad DSD in D/A converters. Side effects in the form of audible noise and distortion are creeping into our audio band with quad DSD, greatly overshadowing the small benefit of the noise shaper curve starting at 80kHz. This, of course, is only apparent in D/A converters that convert the DSD signal directly into analog without any conversion to PCM or other digital filtering. To avoid this the quad DSD signal would have to be low-pass filtered and/or converted to PCM before converting it to analog."

 

I don't agree about the rate, but you cannot endlessly increase the sampling rate for various reasons. For example ~50 MHz clocks frequently used for todays DAC chips have about 10 dB worse low frequency phase noise figures than the clocks needed for chipless DSD512 (~25 MHz). And ESS Sabre uses 100 MHz clocks where you get almost 20 dB worse low frequency phase noise.

 

DSD512 can give you flat noise floor to over 200 kHz. DSD256 can give you flat noise floor to over 100 kHz. So at DSD256 you can already exceed capabilities of 192 kHz 24-bit PCM.

 

[asdf1000]

12 minutes ago, Miska said:

And ESS Sabre uses 100 MHz clocks where you get almost 20 dB worse low frequency phase noise.

 

And this worse phase noise manifests itself in the analogue output jitter measurements? Or other measurements also?

 

When I look at your Pro-Ject S2 DAC (ESS Sabre) measurements, the jitter measurements seem better with DSD512 compared with native PCM RBCD file.

 

So where/how does this 20dB worse low frequency phase noise (for ESS Sabre for example) show up?

 

Thanks!

 

[asdf1000]

For others, here's Jussi's Pro-Ject S2 DAC measurements at DSD512. This looks pretty (very) good, no hump?

 

 

Jitter at DSD512 also looks pretty good.

 

 

Jitter at RBCD:

 

 

So where does the "ESS Sabre uses 100 MHz clocks where you get almost 20 dB worse low frequency phase noise" show up?

 

Or does it show in different measurements?

 

 

[mansr]

3 minutes ago, Em2016 said:

For others, here's Jussi's Pro-Ject S2 DAC measurements at DSD512. This looks pretty (very) good, no hump?

 

There's no hump because the graph stops 50 kHz.

[asdf1000]

10 minutes ago, mansr said:

There's no hump because the graph stops 50 kHz.

 

Nice spot.

 

@Miska why are all your measurement graphs using different scales. I hope that's not deliberate ?

 

Your measurements are awesome but hard (impossible) to compare ?

 

[Brinkman Ship]

5 hours ago, mansr said:

It isn't with a competent DAC. Sadly, not all DACs fit that description.

Completely agree, in fact......

[Summit]

8 hours ago, mansr said:

It isn't with a competent DAC. Sadly, not all DACs fit that description.

 

Which specific DACs in your experience are “competent” as that noise is not audible and which DACs are not competent? I really like to know!

[AMR/iFi audio]

On 6/13/2018 at 7:36 AM, Em2016 said:

The manual states iFi recommend using the volume knob past 12 o'clock (same as their iDSD range) due to the analogue volume control.

 

This is to make it easier for users to set the correct gain for their headphone(s).

 

If the volume control is set near the maximum for comfortable listening levels there is no reserve to turn things up a little and the gain is too low.

 

If the volume control is set near the minimum for comfortable listening levels there is too much gain, so volume is hard to control and noise and distortion are higher than they need to be (excess gain always implies excess noise).

[AMR/iFi audio]

On 6/13/2018 at 11:46 AM, mansr said:

Not a little. A lot.

 

This is the frequency spectrum of an actual SACD. There's a lot of noise energy in the 50-80 kHz band. 

 

The term 'a lot' is hardly a reliable objective qualifier. It would be more reasonable to consider the actual noise levels involved and to consider if they will/can cause problems that will be audible.

 

On 6/13/2018 at 11:46 AM, mansr said:

If you ask me, even 50 kHz is too high a cut-off for this noise profile. 30 kHz would be more reasonable.

 

However, as it so happens, we prefer to refer to available standard documents issued and/or the objective performance of the resultant products.

 

As in: "it is recommended that a Super Audio CD player contain at its output an analog low pass filter with a cut-off frequency of maximum 50 kHz. For use with wide-band audio equipment, filters with a cut-off frequency of over 50 kHz can be used.".

[Miska]

6 hours ago, Em2016 said:

@Miska why are all your measurement graphs using different scales. I hope that's not deliberate ?

 

Your measurements are awesome but hard (impossible) to compare ?

 

Of course it is deliberate, scale is different because it is different measurement! THD plot with 1 kHz fundamental, or standard 19+20 kHz IMD plot would be completely useless on 5 MHz frequency scale! Vertical scale of the wideband plots vary a bit depending on maximum output level from the DAC which determines what kind of gain the analyzer input uses because 0 dBFS is calibrated to the maximum output level.

 

It would be pointless to make all measurements on the same scale. I have three different analyzers for different use cases. Each are good for particular types of measurements.

 

PreBox S2 does have noise hump even at DSD512. So yes, more hump than you get from TI or AKM chips with DSD.

 

And here's micro iDSD BL THD spectrum at DSD512 (Standard filter):

 

And micro iDSD BL IM spectrum at DSD512 (Standard filter):

 

And here's micro iDSD BL THD spectrum at 44.1k input (Standard filter):

Here you can also see spurious tones modulator in the TI chips generates.

 

And micro iDSD BL IMD spectrum at 44.1k input (Standard filter):

 

 

6 hours ago, Em2016 said:

So where does the "ESS Sabre uses 100 MHz clocks where you get almost 20 dB worse low frequency phase noise" show up?

 

This one doesn't! It is not Pro-series chip, it is the mobile chip model in first place... And it uses synchronous clocking (like some Mytek models) instead of the Sabre's ASRC.

 

 

[AMR/iFi audio]

22 hours ago, Miska said:

Hmmh, this is actually wrong. I don't know where people pulled this figure, but I've seen it circulating around every now and then.

 

These numbers come from documents referencing the "scarletbook" SACD standard. The "scarletbook" itself sadly seems to hold 'the unicorn status'. 

 

However other Sony/Philips publications exist in the public domain that state a 100kHz bandwidth for SACD/DSD, e.g.:

 

• http://www.muszeroldal.hu/assistance/sacd.pdf

 

See page 10.

 

We have performed quite extensive listening tests with commercial DSD (1x) recordings and always prefer the most "permissible" filter.

 

We would suggest that anyone who feels that DSD audio should be filtered at 25kHz or 30kHz purchases a product that does so. We do not see any compelling reasons based in fact, rather than opinions bandied around freely in internet forums, to offer such a product.

 

We would also suggest that those who wish to debate the relative merits of different filter strategies for DSD audio would do so in a dedicated thread and via actual tests and actual data (including listening test), instead of theorising from the armchair. Just our 0.02.

[AMR/iFi audio]

19 hours ago, opus101 said:

I'm not knowledgeable about DSD256 etc but I would surmise that higher oversampling moves the noise further away from the audio band. 

 

If using DSD (1x) -> DSD512/1024 upsampling in the iDSD Pro, additional digital filtering is employed which reduces the bandwidth of the DSD signal more than the 'direct' 80khz filter.

 

19 hours ago, opus101 said:

The caveat being of course that no filters in practice are textbook, they will themselves contribute IMD and have non-monotonic stop bands.

 

The filter in the iDSD Pro is purely passive and thus has a performance close to a theoretical ideal well past 10's of MHz.

[beerandmusic]

1 hour ago, AMR/iFi audio said:

 

If using DSD (1x) -> DSD512/1024 upsampling in the iDSD Pro, additional digital filtering is employed which reduces the bandwidth of the DSD signal more than the 'direct' 80khz filter.

 

 

The filter in the iDSD Pro is purely passive and thus has a performance close to a theoretical ideal well past 10's of MHz.

 

since you are here, can you answer a couple other questions....

 

tell me about the clocks used and if there are separate clocks for the Ethernet.

what about the power supply?  is ifi power recommended or is the power that it comes with exceed the ifi power....is the dac usb galvanically isolated?  are any usb "toys" needed for the dac to improve it's sQ?  I want a dac where usb "toys" don't improve the sq...i have always felt that dacs should have the engineering that any usb toys (power or reclocker/regenerator, etc) built into the dac.

 

 

[asdf1000]

31 minutes ago, beerandmusic said:

 

since you are here, can you answer a couple other questions....

 

tell me about the clocks used and if there are separate clocks for the Ethernet.

what about the power supply?  is ifi power recommended or is the power that it comes with exceed the ifi power....is the dac usb galvanically isolated?  are any usb "toys" needed for the dac to improve it's sQ?  I want a dac where usb "toys" don't improve the sq...i have always felt that dacs should have the engineering that any usb toys (power or reclocker/regenerator, etc) built into the dac.

 

 

 

Hey beer, most, if not all, those questions are answered on the product page. Check out the iFi Audio site.

 

They also have a thread dedicated to the Pro iDSD on their sponsored section of this forum where they've answered those questions already.

[asdf1000]

12 hours ago, Miska said:

However level of the noise hump is very low, especially compared to level of images

 

Thanks for all the clarifications Jussi.

 

So the iDSD and Pro-Ject S2 DAC are opposite there.

 

With the iDSD (not BL which is better as you say) at DSD512, the noise hump is low relative to the level of images:

 

But with the Pro-Ject S2 DAC at DSD512, the noise hump is much higher than level of the images (very low level images compared with iDSD):

 

Now there's much MUCH more to how a DAC will perform in the end of course - but of the above 2, which has the better measurement?

 

Is the S2 DAC's noise hump at DSD512 still insignificant overall? Considering it's very low level images?

 

Ideally you want to get rid of both the noise hump and all images of course, but which is the biggest limiter to performance, in the above 2?

 

[Miska]

7 hours ago, Em2016 said:

So the iDSD and Pro-Ject S2 DAC are opposite there.

 

With the iDSD (not BL which is better as you say) at DSD512, the noise hump is low relative to the level of images:

 

But with the Pro-Ject S2 DAC at DSD512, the noise hump is much higher than level of the images (very low level images compared with iDSD):

 

Now there's much MUCH more to how a DAC will perform in the end of course - but of the above 2, which has the better measurement?

 

micro iDSD is better in this particular respect.

 

7 hours ago, Em2016 said:

Is the S2 DAC's noise hump at DSD512 still insignificant overall?

 

Yes, both are so low in level that it doesn't matter. In S2 the peak noise level is -80 dBFS. So if we'd assume a very bad IMD distortion level of 0.1% in downstream component, the intermodulation product would be at level of -140 dB... And since it is random noise, also the intermodulation products would be random noise. So if you hear it, you'd hear background noise similar to what you'd get from analog tape for example.

 

7 hours ago, Em2016 said:

Considering it's very low level images?

 

There are no images at all here! Because the input is DSD512 from HQPlayer which runs proper digital filter with >192 dB attenuation straight to full 512x rate.

 

DAC chips have typically only 8x digital filter and then zero-order-hold aka sample-and-hold (ZOH, SAH, S/H) from there on which doesn't do anything - just copies the same sample multiple times. That is why you get images with PCM input on DAC chips.

 

Images replicate the music signal transformed to higher frequency, with both inverse and normal spectrum and there can be intermodulation between this inverse and normal spectrums which are adjacent to each other. So given 44.1k sampling rate input (through 8x filter that's 352.8 kHz effective sampling rate), for example 1 kHz tone produces images at both 352.8 - 1 = 351.8 kHz and 352.8 + 1 = 353.8 kHz and thus the intermodulation product is 353.8 - 351.8 = 2 kHz. This repeats at multiples of the 352.8 kHz rate, so intermodulation products of every image pair amplify the 2 kHz IMD product.

 

With HQPlayer upsampling effective output rate is 22.5792 MHz, so images would repeat around multiples of this frequency. But at such rate, images are long swallowed in modulator noise and even a very gentle analog filter would anyway manage to remove them.

 

[asdf1000]

Thanks Jussi @Miska !

 

Great explanations.

[Miska]

12 hours ago, AMR/iFi audio said:

These numbers come from documents referencing the "scarletbook" SACD standard. The "scarletbook" itself sadly seems to hold 'the unicorn status'. 

 

However other Sony/Philips publications exist in the public domain that state a 100kHz bandwidth for SACD/DSD, e.g.:

 

• http://www.muszeroldal.hu/assistance/sacd.pdf

 

See page 10.

 

That document is just marketing material...

 

12 hours ago, AMR/iFi audio said:

We would suggest that anyone who feels that DSD audio should be filtered at 25kHz or 30kHz purchases a product that does so.

 

I don't, although I offer such option for PCM conversion though.

 

For comparison, current ADC chips give practically same noise floor for 192 kHz PCM inputs as you get from DSD128 with DSD compliant 100 kHz filter. So they operate just like DSD128 ADC converted to PCM would.

 

12 hours ago, AMR/iFi audio said:

We would also suggest that those who wish to debate the relative merits of different filter strategies for DSD audio would do so in a dedicated thread and via actual tests and actual data (including listening test), instead of theorising from the armchair. Just our 0.02.

 

I agree about dedicated thread. And in my opinion this topic has been discussed to death several times over past years on this forum too. So one could go to any of the earlier lengthy threads on the subject.

 

Not sure if this was really directed to me, but I don't think I'm theorising from the armchair after working 20+ years on audio DSP and DSD and I believe having spent more engineering man hours (years) on DSD than many other companies combined...