Comments on PCM1794A DAC design

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Finally! I've designed the boards for the DIR9001 -> SRC4192 -> PCM1794A DACs I started on i this thread, but being a self-taught and inexperienced designer, I'd love some input on the design.

First - power supplies and decoupling.
Every supply starts off with a separate secondary - C - LM317 - C - (wire to board) - R - C - REG113 - C - R - C - chip. All caps are Panasonic FC bypassed with a 100 nF 0805 X7R, larger ones to start off with and 82 uF closest to the chip. The two cap banks after the REG113 also have a 100 nF MKP1837 film cap. I know it's totally overkill, specially considering the awesome REG113, but I'm wondering if that has any other adverse effects than board space and part cost.

Second - PCB layout.
Decoupling caps are as close to the chips as possible, which in some cases means putting the 0805 cap underneath the board. I've made sectioned ground planes that connect under each chip, or under the last R for the onboard power supplies. All other routing is done by my fantastic logic and uncommon sense (unfortunately not education nor experience), while trying to keep the top ground plane as unbroken as possible, with a bunch of vias.

I've skipped the input transformers, as these DACs will only be fed balanced AES/EBU from the already transformer coupled output of a Lynx AES16 sound card.

Output from PCM1794 in mono mode goes to an LC Audio Zapfilter and into a Twisted Pear JoshuaTree relay attenuator.

Hours of reading and a decent sense of logic has done me well in the past, but not necessarily so in a complex design like this. I'd really love it if some of you could look through the design and give me some pointers before I ship it off to the board house.
 

Attachments

  • board.png
    board.png
    126.1 KB · Views: 1,795
  • sch_dac.png
    sch_dac.png
    23.7 KB · Views: 1,778
  • sch_psu.png
    sch_psu.png
    7.7 KB · Views: 1,717
  • eagle.zip
    35.2 KB · Views: 220
Last edited:
Dear,

I am not an expert, but from all the tutorials I followed, I learned from the experts like Guido Tent and Bruno putzeys, that one should just use a single unbroken ground-plane for analog and digital all together, despite what TI says in it's data-sheets.

Your ground-plane also get interupted multiple times by many data tracks. Use a bottom layer for the groundplane. Make the board 4 layer if you don't have the space.


A quote from Guido "One final word to finish. Very often the groundplane is intersected by a slot, between the analogue and digital part of the PCB. The two planes are then connected beneath the AD or DA converter chip, or even worse at the supply-lines entry on the PCB (as advised by some IC manufacturers). With the planes separated, signal currents that cross the intersection are forced to return via the groundplane and as such need to describe a huge loop, around the intersection. This leads to high emission, and higher chance on inter-system pollution."

the rest you can find here: http://www.tentlabs.com/InfoSupport/page35/files/Supply_decoupling.pdf

With kind regards,
Bas
 
Last edited:
I read that too (among others), and I've think I've followed the essence of it. The bottom ground plane is pretty much unbroken, and where it's had to be broken, it's closely coupled to the top plane.

I was unclear in my first post, but the different parts of the ground plane are sectioned, not separated. That means each signal group has its own strict ground path to follow, but there's a beefy return path where one signal group connects to another, like digital and analog sections of a chip. I can't find it now, but I read an excellent paper on it not too long ago.

Thanks!
 

Attachments

  • bottom.png
    bottom.png
    47.5 KB · Views: 1,560
Ooops, I obviously hadn't read that document afterall, at least not very well... After reading it, I might redesign the decoupling to get the caps closer to ground pins, as I've focused on getting them as close as possible to power pins. And if his words about ditching SMD bypass caps are correct, I'd be a happy man :) There are quite enough 0805 soldering on four of these boards already.

I won't venture into the inductor world with this, but Guido seems to support my idea of putting a resistor in the supply path. I've chosen 10 Ohm resistors, due to the relatively high current (~50 mA), but higher values might be better.

As for the ground splits, it pretty much follows line of sight (so to speak), so I can't see any huge ground loops forming. Grounds will not be connected at the power supplies.

Is there any chance Guido might read this? ;-)
 
the ZapFilter 2 circuit http://www.lcaudio.com/images/ZapFilter/zf2dia.gif will not handle the current from the PCM1794 - certainly not the summed mono mode current max of -20mA - not even the average of -6mA at zero in dual channel mode

I suspect the 20 mA combined mono mode current source requirement is why the BB/TI app circuit spreads the I/V across 4x NE5534 and then sums the Vout

since the PCM1794 doesn't appear to have a negative bias I'd be very leary of higher Zin, Vos "no feedback" I/V - without more info I'd really want to keep the Iouts accurately pinned to gnd

for op amp I/V I'd look at ADA4898-1 due to the high GBW, unique linearized, low Vnoise diff pair input, and 40 mA "Linear Output Current" spec - I think this chip could be a real "game changer" in I/V apps

and I'd build I/V into the DAC board for minimum trace lengths


also I don't think 10 Ohms series R in all of your chip supplies is necessarily a good idea - you need to know each supply pin's current load frequency spectrum and psrr sensitivities to determine if/how low DC Z is needed
 
Last edited:
That's exactly the kind of feedback I was looking for, even if I don't like it :) Now that I think of it, I should have realized the zaps might not handle the pressure. Compared to the PCM1702 in my current CDP, they were very loud with my previous PCM1738, which only outputs 2.48 mA.

The thing is that I'd very much like to keep the zaps. For one, I bought another two zaps just for this DAC (six channels total). Two, I like the sound of them. And three, I'd be very happy if I didn't have to design new output stages, too. But I do agree it would be best to have it all on one board.

The dual mono DACs are not a must for me, but I have the chips, so I figured why not go all out :)

The series resistor comes from an old design PWatts almost finished for me, but I can't find any reference to the value, and I can't get a hold of him anymore. To be quite honest, I'm not even 100% sure it's a resistor, it might have been an inductor... I'm tempted to just slash it and simplify everything - theAnonymous1's MicroDAC is proof that a simple design works fine.
 
Last edited:
novec
You can lower output current by raising the 10K IRef resistor.
I'm using 2 pcm1794's dual mono into a zapfilter at present.
I changed the IRef to 20K but it's still needs to go higher.

In the data sheets it says 7.8mA for the 1794, and double that for dual mono.

allan

Also I don't know if changing the IRef resistor value will or will not affect the spec's out.
I did look for some info as to what/how this will affect the THD etc but couldn't find anything.
 
Last edited:
You meant to say decreasing Iref?
That would lower the signal/noise ratio of the DAC, since the internal noise will stay the same. Halving the output current will decrease the final S/N ratio by 6dB. Is that a price that you are willing to pay just to use the "zap" I/V convertors instead of good old OpAmps?
 
cutting current by half isn't enough to use the PCM1794 with the ZapFilter in current mode - the Zap input Q are working in folded cascode with 3-3.5 mA each bias
even 1/2 the PCM1794 6 mA BPZ current in dual channel mode gives you no signal range before it cuts off one of the Zap's input Q


the ADA4898 isn't just a good old op amp - the multi-tanh input stage linearization hasn't been available before for anyone to appreciate its unique ability to handle large deltaV with 1 nV/rt(Hz) noise in a modern bipolar process

The mix of specs are better than any previous monlithic op amp for audio DAC I/V - but why limit yourself to just one?: http://www.diyaudio.com/forums/digital-source/168721-other-hawksford-iv-nested-loop-op-amps.html
 
Last edited:
SoNic_real_one: Like awpagan says, increasing the Iref resistor would decrease Iref and thus Iout to a level the zaps can handle. And I am annoyingly aware that'll increase S/N at the same ratio... Right now I'm torn between accepting that fact and using the zaps that I know and like, and have paid good money for, or start working on another output that may or may not be better.

jcx: Your solution looks very interesting, the only caveat is the dreaded T&M - time and money. Do you have any idea when you might have a working and tested circuit i can "borrow"? Despite being stubborn and annoying, I am open to other solutions, as long as they're better ;-)
 
That's me - writing half a sentence and thinking the rest :)

I tend to think a discrete output stage has the potential of being the best, but has a greater potential of being a complete failure. The only opamps I have any kind of experience with are the LM4562, which have fared pretty well with the PCM1794 for a friend of mine. And as far as good goes, those tend to come at a price, and I need 10+ channels for five channel active XO. But let's not worry about that just yet...

jcx: I'll start ordering samples right away! With 50-ish opamps it's gonna take a while ;)
 
I suspect the ADA4898 in the single op amp transimpedance I/V per the DAC app note will be very good even at the PCM 1794 current levels - I should think a significant improvement over the 5534 - the low Vin noise allows the noise gain input mod that shaves a little off the peak error by directly shunting switching glitch energy to gnd

the ADA4898 multi-tanh linearizes the input to 1 ppm gm change at ~ 4 mV diff Vin

the multiloop Hawksford I/V would be a beast - I'd probably use a composite for the final follower to reduce its errors too

but op amp expense is only the start - you will be spending $5-10 ea for bulk metal foil resistors - its the only right thing to do when the active devcie distortion is driven so low
 
Ok, now you're getting me seriously interested... Potentially broke, but interested... Where do I sign up? :)

Do you have a rough estimate on when you'll have tested the Hawksford with ADA4898? I'm leaning towards making a simple opamp output for now, and go for a better one once I've played with the DACs some more. I'm still not sure whether the SRC will be necessary with the Lynx AES16 as the only source.

Would you recommend ADA4898 for summing too, or just I/V? after a lot of back and forth I've decided to go single ended.
 
I did have a great experience with LM4562 in summing/filter, so good that I replaced all the OpAmps in my Denon AVR3805 receiver, DVD2930 universal player, Toshiba universal player (SMD's) and even the I/V stage in a Phillips CD 374 (8DIP) with those...

I am interested in the ADA4898, as a matter of fact I am waiting for a couple samples right now. But in Denon's I have PCM1791A, that are voltage-output.
But already the LM4562 is so good that I cannot envision making a big improvement - but who knows?
 
Last edited:
Use the one from the reference PDF page 20,21 - resistor values are ok for 4562. Just ditch the 22pF capacitors, 4562 is internally compensated (unity gain stable).
You also can take a look at how it is done for PCM1792 in Denon DVD-5910...
 

Attachments

  • DAC.jpg
    DAC.jpg
    210.2 KB · Views: 839
  • DAC.zip
    288.9 KB · Views: 221
Last edited:
... considering the awesome REG113...

I'd not come across that part before, certainly does look high performance. But then, buried deep in the datasheet I read this :

The REG113 uses an internal charge pump to develop an
internal supply voltage sufficient to drive the gate of the
DMOS pass element above VIN. The charge-pump switching
noise (nominal switching frequency = 2MHz) is not measur-
able at the output of the regulator over most values of IOUT
and COUT.


Are you really sure sensitive analog circuits are going to be better off with multiple built-in 2MHz oscillators? Especially as each reg has its own and they won't be synchronised.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.