What's inside a Prism Dream AD-2 converter

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

living sounds

Well-known member
Joined
Jul 26, 2006
Messages
3,931
Location
Cologne, Germany
It's impossible to find reliable information how conversion is achieved in this unit. Does it use common sigma delta chips or is actually discrete as claimed in a few articles from usually respactable magazines? The only high end converters since the early 90s rumored to not use sigma delta modulation are the two Pacific Microsonics units and the  old db technologies / Lavry Gold model. No idea how this is practically achieved.

The Prism AD-2 has specs unobtainable with the chips availible at its inception in the late 90s. Did they maybe cleverly combine more than one chip per channel?

Any information is welcome.
 
It would be nice to see internals, but I imagine the warranty sticker on the seam of a $5k+ device holds a fair amount of power over an owner's curiosity.  :)

 
9K+  new.  I have one but have never tried to figure out what’s going on. I know they are hand built. Still sound great.  Vintage digital.
 
living sounds said:
Have you shot it out against other converters? Anything you prefer?

I still think there is nothing better available. I haven't shot it out with anything lately. My test is how a converter sounds with tape as a source. Once audio has gone through a reconstruction filter the difference between converters is very small. Using a music  file as a source to test converters will surely lead you to think  all converters sound the same. They don't.
 
I got one and had a peek inside. I think the actual converter ICs are under a big heat sink connected via thermal grease. I won't touch anything under the hood, so the mystery will stay unsolved for now. It all appears to be mostly late 90s surface mount parts, nothing fancy. Looks really solid, industrial no-corners-cut design.

Now to test how it sounds and measures.  8)
 
I did some tests, the performance is really remarkable, the noise floor is much lower than with any other converter I've ever used. There appears to be active DC offset  compensation that slowly removes static (or at least slow moving) offsets.

 
The real time SRC is also one of the best sounding SRC’s there is. It took others a long time to catch up.
 
The mystery may not be revealed by visual inspection. The good performance is probably a combination of good hardware "and" good software, to slice and dice and post process the oversampled data, and/or multiple conversion data streams.

There is a tendency to think conversions are fully characterized by the codec or chip set used, but execution of that design still matters.

JR
 
JohnRoberts said:
The mystery may not be revealed by visual inspection. The good performance is probably a combination of good hardware "and" good software, to slice and dice and post process the oversampled data, and/or multiple conversion data streams.

There is a tendency to think conversions are fully characterized by the codec or chip set used, but execution of that design still matters.

JR

The SNR is really uncanny. The evaluation boards of current chips don't get close. Maybe there are multiple gain stages combined or something along these lines.

As for distortion, here is an example of someone reducing THD in a DAC by pre-processing the signal:

https://www.diyaudio.com/forums/equipment-and-tools/328871-digital-distortion-compensation-measurement-setup-2.html#post5584029

Don't know if this would work for program material, but it's interesting.
 
living sounds said:
The SNR is really uncanny. The evaluation boards of current chips don't get close. Maybe there are multiple gain stages combined or something along these lines.
Not sure how extra gain stages could help, but poor gain stage execution could hurt.

In over sampled conversions noise shaping can shift wide band noise energy out of the <20kHz bandwidth. I don't know what/if the trade-offs are from gaming this (shifting in band noise away too?). Caveat this is just a WAG. 
As for distortion, here is an example of someone reducing THD in a DAC by pre-processing the signal:

https://www.diyaudio.com/forums/equipment-and-tools/328871-digital-distortion-compensation-measurement-setup-2.html#post5584029

Don't know if this would work for program material, but it's interesting.
Pre-distortion is relatively old technology for correcting known nonlinearities. I am not sure how this applies to digital conversions. (I followed the link but did not find that specific discussion) 

The actual distortion from better modern conversion technology seems acceptably low. Since distortion is often measured as THD+N, it seems like noise floor matters.


JR
 
JohnRoberts said:
Not sure how extra gain stages could help, but poor gain stage execution could hurt.

Maybe gain stage is the wrong term. What I meant was something along the lines of two converters in parallel, one optimized for lower and one for higher amplitude signals, and the post-processing selecting which feed is used for the output depending on amplitude. Just speculating here, I don't know enough to judge the viability of my idea.
 
living sounds said:
Maybe gain stage is the wrong term. What I meant was something along the lines of two converters in parallel, one optimized for lower and one for higher amplitude signals, and the post-processing selecting which feed is used for the output depending on amplitude. Just speculating here, I don't know enough to judge the viability of my idea.
Funny, I was going to mention that several days ago but decided to not clutter up the thread while somebody might still present objective information.

I looked into that decades ago because it seems like an inexpensive way to extend dynamic resolution (level shift audio N dB between two convertors then seamlessly combine the digital output to buy extra bits of resolution. What isn't obvious is that the upper convertor linearity has to far exceed it standard linearity (for that number of bits) so the combined digital output is monotonic. 

Imagine if one LSB of the top convertor represents 4 bits or more (24dB) in the lower convertor.  This nonlinearity some 20dB above the quantization floor, is not likely to sound very bad but as usual no free lunch. There may be other gotcha's but this was enough for me to abandon it a couple decades ago.  :(

===

Bench performance could be gamed with transparent gain scaling (before/after). So bench tests need wiggle all the bits. Gain scaling would show up as noise floor modulation.

JR 


 
living sounds said:
Maybe gain stage is the wrong term. What I meant was something along the lines of two converters in parallel, one optimized for lower and one for higher amplitude signals, and the post-processing selecting which feed is used for the output depending on amplitude. Just speculating here, I don't know enough to judge the viability of my idea.

I remember hearing that there are four discrete converters per channel in parallel.
 
Gold said:
I remember hearing that there are four discrete converters per channel in parallel.
That works too but there are quickly diminishing returns ... Adding two convertors together buys you 1/2 bit(?). Signal adds coherently, noise incoherently so +3dB net gain in S/N... 4 convertors double again so one full bit improvement... +6dB in not nothing and now not as impractical as back when.

JR
 

Latest posts

Back
Top