DSD comparison to PCM.

[PeterSt]

DSD Plot

Ok, I was looking for something like this :

(from Philips, 2002)

This is after noise shaping (one form of it; many exist)

So, supertweeters do influence the sound, right ? (I sure believe that but don't use them). Well, then this should sauce the sound.
I'd say that this can be filtered analogly and assuming that we want to have in frequencies of up to 35KHz (because frequencies to up there will exist) this means we filter at 70KHz with a 6dB (per octave) filter, which means that 70 + 70 = 140 = -6dB + 140 = 280 = -12dB +280 = 560 = -18dB + 560 = 1120KHz = -24dB. Well, at 100KHz the noise is 40dB down so now it will be 64dB down in relatoin to the signal. I didn't "math" further, but at 1000KHz (the 1e+06) it will be that much down just the same.

 

[Orb]

Peter,
from what I have seen/read/discussed with designers a fair few of the high end amps/preamps have a wide bandwidth of around 300khz.

Cheers
Orb

 

[Bruce B]

Ok, I was looking for something like this :

So, supertweeters do influence the sound, right ? (I sure believe that but don't use them). Well, then this should sauce the sound.

At SACD rates but not at DSD 128fs

 

[PeterSt]

Plots - Jitter

A bit reluctant to post (and only because it was asked for), here's some jitter plots from the NOS1.

In all cases the FFT is 256K and 6 times averaged.

First a nice confusing one :

See that ? not 11025Hz but 22050. This is because of the way the whole thing has been set up and it can't cope with a few things. So, the test signal *is* 16 bit 229Hz LSB toggled + 11025 (at -3dBFS), but it doesn't show like that. This can be done normally (see below) but here I wanted to show the broader spectrum, including the clock (see right side at 44100Hz). Because this is upsampled to 24/705600Hz now there's no images visible and no aliasing (but see a small one at ~-128dB at ~800Hz).

Now all looks normal and this is upsampled to 24/176400Hz. But, one channel with data (both are active) and you can nicely see what the noise is caused by, say, glitching. In the mean time we can see the channel separation (watch the red in the 11025 signal) which in the "normal" 24/705600 situation will be some 10dB better.

Zoomed in more and caught one of the 229Hz "harmonics" as well.

Again zoomed in somewhat more and with some imagination we can see a spread over 100Hz for this 11025Hz and it would be at close to -150dB. Or 10Hz at -138dB, also good.

Notice the importance of the 256K FFT which in this case will be less deep than most often used (I can't go higher) and which is against us this time. So, the higher the FFT depth, the smaller the root of the signal will be and the better it will look. In this case I'd say "the better it will be" because no side lobes pop up yet although you just start to see them when the lot is redrawn all the time. But the experienced eye can see it in the still picture as well.

Before someone asks, what's used is "Prism 7" windowing, proprietary to Prism.


PS: As a bonus : 0.0043% THD+N on a JTest signal ? haha. But I really don't know how that compares to "others".

 

[DonH50]

Have you tried testing using the IEEE ADC Standard (1241) method to obviate the need for windowing?

 

[LynnOlson]

Just finished reading the Six Moons review of the NOS1. Umm, let me get this right ... the reviewer auditioned a high-performance DAC through a Class D amplifier? Was the reviewer aware that a Class D amplifier is itself a ADC -> DAC, running asynchronously to the preceding DAC, and a RF transmitter in the AM radio band?

 

[Orb]

Just finished reading the Six Moons review of the NOS1. Umm, let me get this right ... the reviewer auditioned a high-performance DAC through a Class D amplifier? Was the reviewer aware that a Class D amplifier is itself a ADC -> DAC, running asynchronously to the preceding DAC, and a RF transmitter in the AM radio band?

I thought the N-core were for active woofer-speakers or you mean the Devialet (which is technically closer to Quad's current dumping rather than Class-D) ?
Late so appreciate I may had misread something.
Cheers
Orb

 

[LynnOlson]

Looking again at the link:
http://www.6moons.com/audioreviews/xxhighend/1.html
they could have used any of a whole flock of amplifiers.

On page 7 there's a reference to the Ncores ... "The Ncore power amps connected to another zone just like the active woofers of the Avantgardes" ... so it's not at all clear what was connected to what. The system is Ground Loop City, based on the setup and connection struggles of the reviewer.
http://www.6moons.com/audioreviews/xxhighend/7.html

 

[LynnOlson]

On a slightly more serious note, it would be interesting to compare 64fs and 128 fs DSD playback with two completely different systems: the Playback Designs operating in native mode, versus Pure Music or JRiver converting DSD to 352.8/24, and going into the NOS1 with its Burr-Brown 1704 converters.

There's noise-shaping either way, but PB is noise-shaping a single-bit converter, while DSD noise-shaping is done in the computer by PM or JRiver, and flash conversion by 1704's.

 

[bbb]

On a slightly more serious note, it would be interesting to compare 64fs and 128 fs DSD playback with two completely different systems: the Playback Designs operating in native mode, versus Pure Music or JRiver converting DSD to 352.8/24, and going into the NOS1 with its Burr-Brown 1704 converters.

There's noise-shaping either way, but PB is noise-shaping a single-bit converter, while DSD noise-shaping is done in the computer by PM or JRiver, and flash conversion by 1704's.

Close to what I suggested. Master in to DSD (via Sonoma or Grimm) and 24/176 PCM (via PM 2 (according to Mani no OS)) then through Playback and Phasure. Perhaps then we could convert via software the DSD and PCM into vice versa and see what happens.

 

[Bruce B]

. Perhaps then we could convert via software the DSD and PCM into vice versa and see what happens.

It would blow up

 

[bbb]

It would blow up

Ha ha. No funny icon this time?

 

[Bruce B]

Ha ha. No funny icon this time?

Sorry.... I typed this on my iPhone!

 

[bbb]

Sorry.... I typed this on my iPhone!

 

[PeterSt]

Just finished reading the Six Moons review of the NOS1. Umm, let me get this right ... the reviewer auditioned a high-performance DAC through a Class D amplifier? Was the reviewer aware that a Class D amplifier is itself a ADC -> DAC, running asynchronously to the preceding DAC, and a RF transmitter in the AM radio band?

Well, at least it's DSD(-like). And at least this one is from BrunoP. And at least it should be the best measuring amps ever. And yes I live by measurement (+ ears).

But I never heard anything good from that MHz switching gear (and did I not try !). If not the TV went off first. Or the lights.
Heard the nCores in their system(s) too. Brrr. But this was before they sorted out all the groundloop and mains stuff.
In the end I don't know and much if not all depends on getting that radiation under control which for a main part is just about the (feedback) process. And then still. A 0.0001% or whatever it is THD figure is nice, but now let that loose on the whole spectrum and look again.
Yeah, well, DSD ...

 

[LynnOlson]

Now that we're talking about measurements, what the reaction around here to John Atkinson's measurements of the Playback Designs MPS-5?
http://www.stereophile.com/content/playback-designs-mps-5-sacdcd-player-measurements

I noticed the comments about "random bursts of higher-frequency noise" and the left channel having a higher noise level than the right. Did a defective transistor (in the analog section) slip through QC inspection, or is this PB's noise-shaping in action? Or did JA measure the MPS-5 wrong?

 

[MarinJim]

Now that we're talking about measurements, what the reaction around here to John Atkinson's measurements of the Playback Designs MPS-5?
http://www.stereophile.com/content/playback-designs-mps-5-sacdcd-player-measurements

I noticed the comments about "random bursts of higher-frequency noise" and the left channel having a higher noise level than the right. Did a defective transistor (in the analog section) slip through QC inspection, or is this PB's noise-shaping in action? Or did JA measure the MPS-5 wrong?

Point well taken.

 

[LynnOlson]

Well, there's four possible scenarios for the JA's measurements of the MPS-5:

1) Defective transistor in analog section. If a bipolar transistor is damaged by heat or a voltage transient, it can become noisy (which precedes failure). The noise is characterized by bursts, growing worse over time. If the following stages are DC-coupled, the power amplifier can be destroyed by the DC pulses, and the woofer in the loudspeaker will also be damaged. (Even pro woofers can only tolerate 1 watt or less of DC offset, since the voice-coil cannot cool itself by moving back and forth.)

2) Unstable regulator; these can slide in and out of oscillation in the 1~10 Mhz range, and scopes don't easily trigger on bursts of oscillations. RF oscillations makes analog sections do bad things that resemble noise but aren't. Another possibility is a noisy Zener diode that's used as a voltage reference; these are usually bypassed with a cap to lower noise, but the cap might have gone bad.

3) Sub-optimum noise-shaping algorithm in the Playback Designs. This is consistent with the high level noted - no better than 16-bit resolution - but is not consistent with the 3 dB difference in noise levels between channels. You'd expect the algorithm to be the same between both channels. What is consistent with this hypothesis are the noise bursts - 1-bit converters are known for instabilities in the noise-shaping algorithm.

4) JA messed up the measurements somehow. There are many ways to induce ground loops in a measurement setup, and the low levels being measured will reveal this kind of thing. Bad connections in particular will show bursts of noise. Not sure if this would account for the 3 dB difference in noise level between channels, though.

In the absence of more information from either Atkinson or Playback Designs, it's hard to tell what happened. If the noise bursts - which is the most unsettling comment by JA - appeared in both channels, it's the noise-shaping algorithm. If it appeared in only one channel, then it would almost certainly be a bad transistor (or Zener diode). Unstable regulation could appear in one or both channels, depending on whether the MPS-5 shares regulation across channels or not.

The power-supply rails would have to be measured to confirm supply stability, and in the absence of a schematic or a board layout guide, not something you'd expect JA (or any reviewer) to do. Poking around randomly on a circuit board without knowing what you're looking at is a great way to short out a regulator and damage the review sample.

 

[MarinJim]

^ At first I did not get you, but after reading all your posts, I am deeply grateful for your insight and knowledge.

 

[Orb]

I do not know Lynn as the manufacturer response is interesting, and I must admit I am concerned with what exactly can be gained just from a simple tone - feel more needs to be considered with the use of a some kind of very specific complex waveform.
This is the response from Jonathan Tinn, Andreas Koch:

Most of the measurement results are generally to be expected from the way they were measured. What differentiates the D/A converter inside the MPS-5 from other, more conventional converters is that it uses all custom algorithms and discrete components that were not designed following classic theories and practices.
A large percentage of your charts show the behavior of the MPS-5 in the frequency domain, and only two charts show the time domain, although with rigid sinewaves as test inputs. While this would be totally adequate in most cases, it isn't for the MPS-5.
For instance, most of the filter algorithms inside the MPS-5 cannot be described or even defined by feeding them periodic test tones such as sinewaves and looking at frequency charts. They were designed for real music signals, and therefore "listen" to the input signal and vary accordingly, to take advantage of how our ear perceives music, which never even resembles periodic test signals. It is common knowledge that such psychoacoustic criteria hardly ever lead to ideal measurements based on steady-state test signals.

Cheers
Orb