[00940]

Here is what I'm planning to build this summer. If you have any suggestions or improvements, I'm willing to hear.

Digital section:

The digital section is nothing revolutionary and has been already seen more than once. It uses only proven TI ICs. The REG101s are a bit hard to get though but as I've a few left from another project.... The PCB is one fourth of an euroboard (50/80mm).

A PCM2707 is used as USB-to-I2S receiver. It is powered by its own REG101-33. A 12MHz crystal is used. A single AND gate takes care of the USB power's pull-up.

The ASRC chosen for jitter rejection is the SRC4192 (which is pin compatible with the AD1896). It shares a reg101-33 with the digital filter of the PCM1798. See why the src4192 rather than the AD1896 here : diyAudio Forums - Any feedback on new CS8421 high-res ASRC?

The DAC is the PCM1798. While its digital filter is not as good as the PCM1794's, it has the advantage of a weaker analog output. It thus makes it easier to use a discrete I/V. Its digital filter is set up to slow roll-off (is it the best option ? I'm still undecided). A REG101-5 powers the analog section.

A 24.576MHz clock feeds both the ASRC and the DAC. It can be either DIP8 (such as a Tent clock), DIP14 or SMD. A fourth and last REG101 powers it.

The PCM2707 outputs 44.1KHz/16bits. The ASRC turns that into 96KHz/24bits and the digital filter of the PCM1798 oversamples it 8 times. Aliases should thus appear at more than 700KHz, allowing for a gentle analog filtering.

schematic: http://membres.lycos.fr/sst00940/usbdac/dacsch.png

I/V section:

Each analog output will be connected to a discrete I/V stage. Based on a design by Cauthemoc, it should keep distortion quite low while allowing me to get rid of opamps. The idea behind this design (the current conveyor) is similar to what was found in Wadia products.

After each I/V, there is a simple emitter follower with a 2nd order Bessel Sallen Key filter. It provides 55dB of attenuation at 720KHz (while the digital filter already provides 80dB). At 20KHz, there is only 0.3dB of attenuation. It also buffers the output of the I/V.

They will be wired P2P.

Gilmore Dynalo:

The Dynalo sections are used as balanced to unbalanced converter. A 4-pole switch controls gain: either around 2.2VRMS (line out) or 7VRMS (headphones amplifier). A pot wired in between the hot and neutral allows to control volume.

Power supply:

The DAC will use two transformers. One is 18-0-18 @ 30VA for the I/V boards and the Dynalo. The other is 0-9 @10VA for the DAC board.

A pair of LM317-337 powers the Dynalo at around 18V. Before the regulators, there is 3* 4700µF per rail and a pi filter. To improve the performances of the LM*, zeners are used to set the voltage out rather than the typical resistors. (see Using 3-pin regulators off-piste: part 1 )

The 4 I/V sections will be powered from the regulated supply. They will be isolated from the Dynalo by zener-based open-loop regulators. The operating voltage will be around 15V. Those regulators will be wired P2P.

A stereo Dynalo draws around 180mA per rail. Each I/V section draws around 30-35mA per rail. The total dissipation will thus be around 15 to 16W for the analog section.

The DAC section has its own regulators onboard. However, each REG101 will be fed from a pre-regulator similar to those used for the I/V sections, set at 6VDC.

To compare it to well-known stuff, it’s a bit like TPA’s usb receiver + metronome + COD with a zapfilter on steroids.

 

[error401]

A bit off topic, but I was wondering how you felt about doing SRC after the PCM2707. I've got some SRC4392 at hand, but no board for them. I was considering it as a next project, but I've never really had the opportunity to compare SRC and without. Being a bit skint at the moment, do you think it's worth the trouble?

About your design, I do have a couple comments, but they're minor. Mainly it looks sufficiently over the top to me

. Should be a great system.

- It's probably not a bad idea to damp the digital lines (I2S) with a small value resistor (100R or so)

Okay, I only have one comment. I'm not really up on the analog side of things so I can't really comment there. I don't think I like the pre regulator design, but I don't know enough to comment intelligently. Why does the I/V need so darn many transistors? I really can't make heads or tails of that design, though I admit I suck at the analog side. A link to more details?

 

[00940]

- about the ASRC: many people have good things to say about them. Considering also that the src4192 should stay in "slow mode" with the PCM2707 (the measurements of pcm2707's jitter shown on diyhifi aren't great but ok enough), it shouldn't hurt much either.

- about the regulators: well, they aren't perfect, that's for sure, but beside the fact they're cheap, they're also great at rejecting high frequency noise, something the LM317-337 are not doing well.

- about the I/V stage: it's coming from here: http://www.head-fi.org/forums/f6/diy...97/index6.html
It could use a lot less transistors if you removed all the cascodes. To be honnest it took me a long time playing with it in LTspice to understand how it works. And I'm not sure I really understant all of it

 

[Pars]

Very nice! I'll have to spend some more time to digest it all. I need to learn how to use SPICE myself

 

[ezkcdude]

I would be interested in obtaining the I/V board - I could use it with my ezDAC (PCM1794/8). I do question your source on the SRC4192 vs. AD1896 over at diyaudio. I only see a little bit of anecdotal evidence. To my knowledge, nobody has actually provided these measurements publicly. Have you seen something else? (It's not a huge issue, since they are pin-compatible. I use whichever one I have on hand, actually.)

 

[00940]

I'm sorry, but there will be no PCBs made for the I/V. At least not by me. The pics above are layouts for wiring the I/V and regulators point to point on protoboard.

Wrt the AD1896 vs SRC4192. The jitter measurements of the pcm2707 can be found on diyhifi.org. I don't have exact measurements about the SRC4192 but in an email exchange with Bruno Putzeys, it was confirmed that the SRC4192 wouldn't go in fast mode before "tens of ns of jitter". The PCM2707 should thus be allright. I wouldn't dismiss anything written down by Putzeys as anecdotal. He doesn't post that much and never in vain. Anyway, as been said, the AD1896 and SRC4192 are pin compatible (but you have to be careful about clocks, the AD1896 doesn't support all the frequencies supported by the SRC4192. The TI parts is also a tad more power hungry).

An interesting reading about him btw: IEEE Spectrum: Bruno Putzeys: The Sound Of Music (Extended Play)

 

[ezkcdude]

Quote:

Originally Posted by 00940 /img/forum/go_quote.gif I wouldn't dismiss anything written down by Putzeys as anecdotal. He doesn't post that much and never in vain. Anyway, as been said, the AD1896 and SRC4192 are pin compatible (but you have to be careful about clocks, the AD1896 doesn't support all the frequencies supported by the SRC4192. The TI parts is also a tad more power hungry). An interesting reading about him btw: IEEE Spectrum: Bruno Putzeys: The Sound Of Music (Extended Play)

Yeah, I definitely don't want to come across as if I'm insulting anyone. I am just skeptical by nature. BTW, in my latest ezDAC build I'm using AD1896 with a 24.576 MHz XO (Abracon half-size TTL), and it's set to resample at 96 kHz (just as yours will). I'm using the "slow" rolloff filter on PCM1794.

 

[holland]

Quote:

Originally Posted by 00940 /img/forum/go_quote.gif An interesting reading about him btw: IEEE Spectrum: Bruno Putzeys: The Sound Of Music (Extended Play)

A nice read. Thanks for the link. I like his views.

 

[00940]

ezkcdude: I didn't feel you were insulting anyone. I'm quite skeptical by nature too but Bruno Putzeys is on a short list of people I'm willing to trust (maybe because he's Belgian

).

Which abracon clock did you pick btw ? I'm considering this one (3$ at digikey) : http://www.abracon.com/Oscillators/ASV.pdf

 

[Cauhtemoc]

Good job.

 

[FS2]

This is inspring me to get back to work on my own DAC, which has been sitting on the back burner for way too long now...

I have a question about ASRCs though. Hope nobody minds as I haven't really been able to find a clear answer even after sifting through forums: What is an ASRC meant to acheive? And does it really make a difference?

 

[error401]

Quote:

Originally Posted by FS2 /img/forum/go_quote.gif This is inspring me to get back to work on my own DAC, which has been sitting on the back burner for way too long now... I have a question about ASRCs though. Hope nobody minds as I haven't really been able to find a clear answer even after sifting through forums: What is an ASRC meant to acheive? And does it really make a difference?

Jitter rejection. Both S/PDIF and USB are pretty poor clock transfer mechanisms, so you don't end up with the most jitter-free clock when recovering from them. Using an ASRC keeps the data, but should remove the jitter from the signal and use a local clock.

Whether it makes a difference or not...

 

[FS2]

Should remove the jitter... but is that the primary purpose of an ASRC, or are they meant to resample the signal and jitter clean-up is a nice side effect?

 

[00940]

Well, they are called asynchronous sample rate converter, no ? So indeed they are at first meant to resample the signal. The original aim of such parts was to use only one local clock for various incoming frequencies (typically multiple of 44.1khz or 48khz). It simplifies quite a lot some designs, considering they also let you switch between 16 and 24bits. Whatever you feed the ASRC with, the output is always the same.

However, they're also asynchronous. It means that the time domain at the output has no relationship with the clock at the input. Thus, if your output clock is perfect, you should have (almost) no jitter at the output. It's actually a bit more complicated and some of the input jitter will find itself at the output, transformed into errors of amplitude (signal noise and distortion). If the ASRC is well-designed, that part that goes through should be so attenuated that it doesn't matter.

As such, jitter cleaning is not what the ASRC was designed for, but it's not exactly a side effect either.

Now, does it make a difference ? Technically speaking, it should reduce jitter from a few ns to a few ps. Does it really matter ? Many people i'm willing to trust say it does. I'll see.

It will of course turn a typical 44.1khz/16bits signal into a 96khz/24bits signal. You don't gain any new informations this way, so I don't expect to gain anything soundwise there (despite all the marketing claims about upsampling and so on). However, I don't expect to lower the sound quality either, considering how low the distortion and noise figures are for the SRC4192. At least, I get some 300KHz extra for analog filtering (somehow going from 8x oversampling to 16x).

The typical link on the topic: diyAudio Forums - Asynchronous Sample Rate Conversion

 

[00940]

Quote:

Originally Posted by error401 /img/forum/go_quote.gif ... Okay, I only have one comment. I'm not really up on the analog side of things so I can't really comment there. I don't think I like the pre regulator design, but I don't know enough to comment intelligently.

Well, I've been thinking about this... What do you guys think about this supply instead ?

The +15 negative supply on the schematic is just there since I didn't simulate also the negative supply. A definitive version would use the ne5532.