0.9y-4 - w00t

[earflappin]

I spoke with the guys at Berkeley (formerly Pacific Microsonics) and did some more experimentation with my Berkeley Alpha DAC (BADA). 

Mani, as I mentioned before the BADA has 4 filters (actually 8 as there are 16 and 24 bit versions of each of the 4 filters).  Filter 2 is the same filter as used in your Model 2.  Filter 1 is a more sophisticated filter which required computing power that was not available when your Model 2 was designed/manufactured.  AFAIK, the BADA will not upsample an incoming 176 signal, but will upsample incoming 44.1, 88.2 streams.

Coming out of XXHE with no upsampling (NOS) and no interpolation/filter and using filter 1 I get amazing SQ.  Basically, filter 1 is a form of QAP - i.e. it is an upsampling and interpolation algorithm.

However, while the differences are subtle, I continue to prefer using QAP into the BADA - using either filter 1 or 2 gets me excellent results.  To my ears, with QAP the sound is more "filled in", there is more detail and three dimensionality and that bit of glare at the high frequencies is reduced/eliminated.  BTW, to my ears, the bottom end is more resolute with QAP which some might hear as less powerful.  I think this is because with QAP there is a reduction in time smearing which is what makes the bass sound louder (like a loudness control).

Based on my input and their other customers, they are going to set up XXHE on Win7 and experiment with it.   They told me that while upsamping is obviously optional, every DAC has to have some form of digital filter.  Peter, I got the impression that your DAC was going to be devoid of any digital filter?  Is that right?  They asked me what type of minimal digital filter would you want to see used with QAP?

[PeterSt]

David, about your question :

My DAC doesn't contain any filtering whatsoever, and besides that it has been explicitly made to eliminate all natural filtering which could come from analogue steps (like my first I/V contained over a 1000 parts (not that the signal went through all of them), while right now the count is, say, zero.

And so, this is exactly what the Arc Prediction has been made for : to supply the DAC with easy to test and control custom filtering, that normally being in-DAC. It will obviously work on any DAC which really is 100% filterless, BUT that DAC should be 24 bits, or otherwise it can't work.
That it also works out for the better on OS DACs (or NOS filtering DACs if you want) mist be because the algorithms normally used (sincx) do not get the opportuniy to destroy. Like : the work already has been done, and the very same algorithm can deal with that (something like : the required result = 1.5 while the base is 3 *thus* devide by 2, versus, the required result is 1.5 while the base is 1.5, *thus* do nothing).
This is just my estimation afterwards, but I think the math concerned allows for just that.

Peter

[PeterSt]

I ran a test using an SPL meter.  Going QAP into my DAC increases sound pressure by 5-6 db versus going in with no upsampling.  This occurs independent of whatever filter setting I use on the DAC.

Ok, just spent a few hours on this, but there is no way I see what you see. Not by SPL measuring test tones, not by SPL measuring music, and not by looking at test tones at the scope. But maybe to keep in mind : I can only compare with the other means of upsampling, or no upsampling, and, my DAC doesn't filter either way.

It should -said softly- be the conclusion that when your DAC is allowed to filter effectively itself (read : it hasn't been pre-done so now that math in there goes along with it (as how I estimate it works !)) it "filters" everything down with some 6dB ? hard to believe of course, but theoretically possible.

I can imagine it depends on the music, and for testing SPL with music I used the Yellow Submarine track from the original Yellow Submarine album. Just the first 20 seconds or so. Maybe you should try that one as well, just in case.
If you don't see it happening there, you might mention to me what you used to see it happening. Then I can check that too.

Peter

[manisandher]

Mani, as I mentioned before the BADA has 4 filters (actually 8 as there are 16 and 24 bit versions of each of the 4 filters).  Filter 2 is the same filter as used in your Model 2.

Earflappin,

The following has been posted on the Goodsoundclub forum: "The Berkeley Audio DAC reportedly uses the Delta-Sigma Analog Devices’ DAC AD1955A." Could you pass this by the Berkeley guys and find out for sure? If this is indeed the case, then I can't see how the filters can be the same - the Model Two uses custom, discrete full-ladder converters.

BTW, a 5-6dB increase in volume with QAP is huge! I wonder what's going on...

Mani.

[earflappin]

Mani, since my post I spoke to a friend who owns both the BADA and the Pacific Microsonics Model 2.  He told me the Model 2 is more resolving to his ears and regards it as being superior. 

The Berkeley guys decline to confirm what DAC they use....but I'm pretty sure it does use the Delta-Sigma. 

This comes right out of the BADA manual: "Filter 2 is the same filter used in the Pacific Microsonics PMD-100 and PMD-200 HDCD decoder IC’s."  And they were clear in saying they regard Filter 1 as being superior to Filter 2 and that neither the Model 1 or 2 had the computing power to run it.

Have you had any experience with the Merging Pyramix Mykerinos software/hardware offering? 

[manisandher]

I think things might be a little more complicated. Loads of CD players and DACs use the PMD-100 and PMD-200 filters. Indeed, anything with HDCD replay must use these filters (other than the BADA that is, which I think uses an equivalent - superior? - filter).

For 16/44.1, I can well believe that the BADA's Filter 2 is the same as the filter that the Model Two uses... at these SRs.

But the Model Two uses true multi-bit converters, as opposed to delta-sigma, and I think does not oversample at rates >=176.4. But as you've found, it can be hard getting any definite answers from anyone.

To be honest, I'm still stumped as to why QAP still sounds 'good' with delta-sigma DACs, which by definition, need to use heavy oversampling to work. But many people are reporting that it simply does. At some point, I'll try to find some time to experiment with my FF800...

(Your friend has a BADA and a Model Two? Lucky guy...)

Mani

[manisandher]

Have you had any experience with the Merging Pyramix Mykerinos software/hardware offering?

No I haven't... but I am intrigued.

One thing that I've wondered for a while is how the interface can still affect the sound if it's being slaved to the DAC. It seems likely that different power supplies, etc,  could certainly affect the situation if the clock resides in the interface... but when the clock doesn't sit in the interface but in the DAC instead and the interface is bit-perfect, what else is there that can affect the sound of the interface?

I've been threatening to compare the 3 'interfaces' that I have (AFI1, FF800 and MOTU) in slave mode for a while now. Hopefully, I'll get a chance to do this in the next few weeks.

Mani

[manisandher]

Actually, perhaps getting the clock from the DAC to the interface 'intact' is not as trivial as it seems. Certainly, I can't see how the breakout cables that the Lynx card uses could maintain the required 75 Ohm impedance for receiving the clock 'intact'...

Mani.

[boleary]

Well, it took a while but slow learners do learn..... About three weeks ago I reconfigured my system: got the dac next to the laptop and replaced an 8 ft USB and a 6 foot S/PDIF cables with a .5 meter kimber USB and S/PDIF cables, respectively. sh*t. 9y-4 with double arc prediction makes my delta-sigma sing. I had no idea just how smeared the highs were until now.....

Right now it sounds so good that replacing the laptop with a desktop just has to be on hold for a while longer....

[PeterSt]

Hahaha, all great experiences !!

But ...

and replaced an 8 ft USB and a 6 foot S/PDIF cables with a .5 meter kimber USB and S/PDIF cables

Do note that the big trick here is the conversion from feet to meters. XXHE outputs in meters you know, so this avoids a conversion !

[Moritz]

the explanation is quite easy, I am seeing the same on my BADA and it's by design; to support HDCD decoding, 24bit is 6db louder than 16bit:

"encoding may use reversible limiting called “Peak E xtend” that can reproduce peaks
4 4 .1kH z is the only sampling rate that uses H DCD amplitude encoding. Amplitude
6 dB higher than normal 16 bit data. To reproduce those peaks, the output level to
the DAC must be decreased by 6 dB, or 1 bit, to allow for peak reconstruction if an
H DCD encoded source is played."

I ran a test using an SPL meter.  Going QAP into my DAC increases sound pressure by 5-6 db versus going in with no upsampling.  This occurs independent of whatever filter setting I use on the DAC.

Ok, just spent a few hours on this, but there is no way I see what you see. Not by SPL measuring test tones, not by SPL measuring music, and not by looking at test tones at the scope. But maybe to keep in mind : I can only compare with the other means of upsampling, or no upsampling, and, my DAC doesn't filter either way.

It should -said softly- be the conclusion that when your DAC is allowed to filter effectively itself (read : it hasn't been pre-done so now that math in there goes along with it (as how I estimate it works !)) it "filters" everything down with some 6dB ? hard to believe of course, but theoretically possible.

I can imagine it depends on the music, and for testing SPL with music I used the Yellow Submarine track from the original Yellow Submarine album. Just the first 20 seconds or so. Maybe you should try that one as well, just in case.
If you don't see it happening there, you might mention to me what you used to see it happening. Then I can check that too.

Peter

[PeterSt]

This may be a hard thing to understand by some, and even I am a kind of astonished by an outlay like this.
It would be true though ...

"Peak extension" is exactly what happens, and for the same reason you may get "crack detect" trips because the check for that doesn't anticipate on that yet in 0.9y-4.

Moritz ... great again.
Peter

[manisandher]

the explanation is quite easy, I am seeing the same on my BADA and it's by design; to support HDCD decoding, 24bit is 6db louder than 16bit:

No, 24bit is not louder than 16bit.

However, 24bit is louder than HDCD-encoded 16bit... on an HDCD-capable DAC/player... because of the 'Peak Extend' feature.

It now all makes sense to me now. Earflappin must have been using HDCD-encoded 16/44.1 material when evaluating Arc Prediction using the BADA. With no AP, the BADA would have engaged the HDCD processing and reduced the output by 6dB, in line with the HDCD specs. However, engaging AP would have changed the LSB, and the BADA would have switched the HDCD processing off, increasing the output by 6dB.

This is all down to earflappin using HDCD-encoded files. If he had used 'normal' 16/44.1 files, this should not have happened.

[EDIT: I prefer playing 16bit HDCD-encoded files back without any upsampling, to take full advantage of the HDCD 'Peak Extension' and 'HDCD Limiting' processes.]

Mani.

[manisandher]

Of course, my comments in the previous post might well not apply to the BADA. It could well be that the designers decided to reduce the output for all 16bit material by 6dB... in the event that some of it might be HDCD-encoded, thus requiring this attenuation for the HDCD 'Peak Extension' process to work effectively.

The only rationale that I can see for doing would be to normalise the output between HDCD and non-HDCD material from the BADA's outputs. But this then imposes 6dB of digital attenuation (still a 'no-no' in many people's eyes) on >99.99% of the 16bit material that just doesn't need it...

Of course, this 'imposed' 6dB attenuation is then lifted in the BADA once XXHE's upsampling increases 16bit material to 24bits.

My sincere apologies to earflappin if I came across as arrogant in my previous post.

Mani.

[PeterSt]

But this then imposes 6dB of digital attenuation (still a 'no-no' in many people's eyes) on >99.99% of the 16bit material that just doesn't need it...

Of course, this 'imposed' 6dB attenuation is then lifted in the BADA once XXHE's upsampling increases 16bit material to 24bits.

Hmm ... Going from 16 bits to 24 bits is not much related to a higher volume ... only dynamic range and volume resolution.

Also, if 6dB would be needed to extend the peaks, I don't know of any album that has that headroom, although they might exist. Take for example Crime of the Century (Supertramp) which is an album carying much headroom (and is very soft for that reason), which still can bear some 2dB only.

Don't ask me how much a soft-clipped album needs attenuated first in order to get to the full peaks again, which btw gives me the idea doing just that.
I think I told elsewhere that the Arc Prediction still can be improved, and this is one of the subjects. Another is dealing with the hard clipping which actually requires a similar solution, though even more attenuation. This would be a good thing, because currently AP *causes* hard clipping where soft clipping is just under the hard clipping point.

I dare say that 50% of albums reach the maximum level at more than one sample, and therefore hard clip.

Ha, but wait a minute ...
You guys shouldn't have allowed me to apply some "processing" by liking Arc Prediction. Now the bear is loose;
With some nice Jamaican inspiration I just thought of a way to uncompress compressed albums ...
Yep, should work.
The only reason I see causing it not to workout to how the (dynamic range of the) recording ever was, is that I expect not all compression to work linearly. And if not, how to know how the relations are (between low volume compression and higher volume compression). I think this can be "seen" though. So once I manage to indeed analyse those relations ... watch out for a next version of Arc Prediction ...

Peter