[00940]

I meant a cap accross the feedback resistor in an opamp based I/V.

If you have some time, please read again this : http://www.diyhifi.org/forums/viewtopic.php?t=262 . According to Andy, the cap accross the feedback resistor has a serious influence on the bandwith requirements. He suggests that for the DAC he used (8Xoversampling, 7.8ma output), a bandwith of 14MHz is actually enough, if you use a largish cap accross the feedback resistor. It'd be worse for NOS and lower output. But even then, do we really need such crazy opamps ?

 

[ezkcdude]

Quote:

Originally Posted by Rescue Toaster Slew rate is absolutely everything in opamp I/V. Go for the LM6171, just be careful of the layout with an opamp that fast. C35 in particular is pretty far away... I'd move R28 if possible so C35 can get closer. Maybe make the feedback resistors SMD?

Quite a bit of activity since my last post

.

Rescue, where were you last week, buddy

? I've already ordered some boards for prototyping, but I can certainly make these types of changes for the "final" version.

However...I thought it's a general design rule to use higher quality (e.g. metal film) resistors, as oppposed to ceramics, wherever cost and space permit. You're suggesting replacing a high quality resistor with a surface mount to save a few mm lead inductance on the bypass cap. Is that a good tradeoff? It looks to me like I can move C35 as close as most of the other bypass caps on the board, just by shifting it to the right and down a few mm, without affecting R28.

Also, as for slew rate, remember that I originally was planning on using either AD8610 which has a slew rate of 50 V/us or AD8065 (180 V/us). It seems to me with either of these op amps, slew rate is not going to be the limiting factor here. Do I have that right?

 

[ezkcdude]

Quote:

Originally Posted by 00940 I meant a cap accross the feedback resistor in an opamp based I/V. If you have some time, please read again this : http://www.diyhifi.org/forums/viewtopic.php?t=262 . According to Andy, the cap accross the feedback resistor has a serious influence on the bandwith requirements. He suggests that for the DAC he used (8Xoversampling, 7.8ma output), a bandwith of 14MHz is actually enough, if you use a largish cap accross the feedback resistor. It'd be worse for NOS and lower output. But even then, do we really need such crazy opamps ?

In that thread, Walt Jung talks about using film caps for decoupling and specifically recommends some Panasonic caps at digikey. I think these must be what he's talking about:

http://www.digikey.com/scripts/DkSea...591549&Site=US

Has anyone here tried these?

 

[Rescue Toaster]

Quote:

Originally Posted by ezkcdude In that thread, Walt Jung talks about using film caps for decoupling and specifically recommends some Panasonic caps at digikey. I think these must be what he's talking about: http://www.digikey.com/scripts/DkSea...591549&Site=US Has anyone here tried these?

He's talking about digital decoupling caps. Panny ECHU right? Whether they're the best or not I don't know. I know of one pcm1794 dac that just uses tantalum & ceramics and everyone says it sounds great, so...

As for that cap placement, I think it might only be an issue with very fast opamps. Since you're doing resistor I/V it probably wont' be an issue. However the one recomendation I'd make is to think about adding a filter pole before your output opamps. That would mean adding 1000-3000 pF silver micas across your I/V resistors. Mouser Part 5982-19-500V3000 could probably be mounted underneath the PCB, directly to the I/V resistor pads.

Also, on another look at the PCB, C32 & C31 are 'backwards'... the VCC should connect to the left pads and ground to the right. That makes more of a straight shot from the ground pads on the right to the ground pin on the 1794. Right now the ground has to loop above & down the right side to get to the pin. Likewise C26 & C27 should be turned sideways with VCC on the left & ground on the right, which will make the traces much shorter. This will also tighten up the loops around your I/V resistors (Bring thier grounds closer to the grounds of the decoupling caps) and improve the noise reduction of those silver mica caps across the I/V resistors.

 

[ezkcdude]

Quote:

Originally Posted by Rescue Toaster He's talking about digital decoupling caps. Panny ECHU right? Whether they're the best or not I don't know. I know of one pcm1794 dac that just uses tantalum & ceramics and everyone says it sounds great, so...

Actually, ECP (digikey# PCF1126CT-ND) come in 0805, I think the other ones are all 1210 and larger.

Thanks, for your advice about the PCB. Sounds like good suggestions to me.

 

[t52]

thanks ez for updating the schematics!

i've got another question: the i/v conversion circuit you used seems to be pretty common out there, though i think there might be a design flaw: as far as i can see the negative voltage across r23 (i look at you schematic, left channel!) will see a parallel resistance of 150 + 1k (r26/r27) - that makes for 1.15k, the positive voltage developed across r24 will see a parallel resistance of only 150ohms (r25 only, as the inverting input of the opamp is a virtual ground), correct?
if that holds true, one signal will be stronger than the other and we don't have true balanced operation anymore, so common mode distortion cannot be removed completely...
i think this is quite important for balanced output dacs like the pcm1794, the current outs each show significant amounts of distortion, which seems to be common mode and are reduce significantly in balanced opertion - sorry, i read that some time go from somone who actually did measurements but i can't find the link anymore

 

[ezkcdude]

Quote:

Originally Posted by t52 i've got another question: the i/v conversion circuit you used seems to be pretty common out there, though i think there might be a design flaw: as far as i can see the negative voltage across r23 (i look at you schematic, left channel!) will see a parallel resistance of 150 + 1k (r26/r27) - that makes for 1.15k, the positive voltage developed across r24 will see a parallel resistance of only 150ohms (r25 only, as the inverting input of the opamp is a virtual ground), correct? if that holds true, one signal will be stronger than the other and we don't have true balanced operation anymore, so common mode distortion cannot be removed completely... i think this is quite important for balanced output dacs like the pcm1794, the current outs each show significant amounts of distortion, which seems to be common mode and are reduce significantly in balanced opertion - sorry, i read that some time go from somone who actually did measurements but i can't find the link anymore

Yes, I was aware of this when I chose that particular design. It's not ideal, but it is much easier to implement than the solution (well, one solution) I saw, which utilized two opamps per channel. I think I saw this on the Elliot Sound Products website some time back, but I just looked again and couldn't find it. I'll keep looking, too... As for how it will affect the sound, I wouldn't get all gloomy about it. I think if the board works, and sounds pretty good now, I will address several issues, including: 1) SPDIF input circuit 2) layout issues as was pointed out above and 3) I/V conversion. However, each time a new IC is added I fear complexity and eventually the board size will need to get larger and more expensive. I'm willing to optimize this initial design, but there needs to be some tailwind for me to sail

. On the other hand, if this prototype sounds terrible, maybe I'll give up DIYing forever

!

 

[ezkcdude]

Boards are in

! I'll put hi-rez versions on my website, but for now...

 

[bubbamc119]

EZ, I'm very interested in your design. Are you doing Electronics Engineering at College? The reason I ask is I need to choose a 4th year project soon and was considering a 24bit 96khz (or higher) hi-fi dac, but I don't know if it would be too easy or too complex. I would be interested in building your DAC.

 

[ezkcdude]

bubba, I was not trained in EE. I've learned everything I know about electronics (well, except for one circuits class in college) over the past year or so. That's probably not saying too much, though. Anyway, once the board is tested and I know it works, I will probably modify the design and get some more boards ordered for people who might want them. You're welcome to have one at that time. Although, I think it would be better for you to design your own DAC. It's quite a good learning experience! Heck, I'm sure you'd be able to teach me a thing or two after taking those EE courses

.

Because you're Australian (correct me if I'm wrong), let me take this opportunity to express my feelings about Steve Irwin's death. It's so tragic. I can't remember the last time I actually felt truly sad about a celebrity dying. He seemed like such a good guy. I'd like to propose that all head-fiers pay a small tribute by hitting pause on their CD players for a minute. I'll certainly miss that guy. As they say in your country, Cheers, mate!

 

[bubbamc119]

Yep Steve's passing on Monday was really sad. Then on Friday Peter Brock (Australias Motor Sport hero) died not far from where I live - crashed directly into a tree and died instantly while on a rally. Been a bad week that's for sure.

 

[t52]

Quote:

Originally Posted by ezkcdude Yes, I was aware of this when I chose that particular design. It's not ideal, but it is much easier to implement than the solution (well, one solution) I saw, which utilized two opamps per channel. I think I saw this on the Elliot Sound Products website some time back, but I just looked again and couldn't find it. I'll keep looking, too... As for how it will affect the sound, I wouldn't get all gloomy about it. I think if the board works, and sounds pretty good now, I will address several issues, including: 1) SPDIF input circuit 2) layout issues as was pointed out above and 3) I/V conversion. However, each time a new IC is added I fear complexity and eventually the board size will need to get larger and more expensive. I'm willing to optimize this initial design, but there needs to be some tailwind for me to sail . On the other hand, if this prototype sounds terrible, maybe I'll give up DIYing forever !

the flaw could be fixed by using a r26/r27 combination that's 150 ohms in series, that would make 19.56ohms/130.4ohms, no plan if there are close values...
but i'll post a circuit in a later post with my approach to i/v...

 

[t52]

here's my approch to i/v conversion:

some explanation:
the i/v conversion resistors r1 and r2 and c1 form a first order lowpass tuned ad 80khz (hope the values are correct, did some quick calc only), followed by another two lowpasses (r4/c3 nd r3/c2). then i use two opamps as buffers, followed by a summing opamp. i think this config could be quite promising as we have a second order lowpass BEFORE all opams, thus removing a lot of the high freq spectrum and lowering the opamp's slew rate requirements...

would like to hear some comments, regardles of ez wants to use this design (it's his dac after all

) of course the filters could need some tuning for pulse response by simulating this circuit - i don't have a sim software though. can anybody recommend some free software?

 

[ezkcdude]

Quote:

Originally Posted by t52 i've got another question: the i/v conversion circuit you used seems to be pretty common out there, though i think there might be a design flaw: as far as i can see the negative voltage across r23 (i look at you schematic, left channel!) will see a parallel resistance of 150 + 1k (r26/r27) - that makes for 1.15k, the positive voltage developed across r24 will see a parallel resistance of only 150ohms (r25 only, as the inverting input of the opamp is a virtual ground), correct?

It seems to me the inverting input is not a virtual ground in this case, as the non-inverting input is not connected directly to ground. That doesn't mean your proposed circuit isn't a good idea (I'm not sure really), but maybe it's not solving the right problem. I think the main problem in my design is matching the resistor pairs. That will do the most to increase common mode rejection. Well, these are my thoughts for the moment. I could be (and believe me, it's happened) completely wrong.

 

[t52]

Quote:

Originally Posted by ezkcdude It seems to me the inverting input is not a virtual ground in this case, as the non-inverting input is not connected directly to ground. That doesn't mean your proposed circuit isn't a good idea (I'm not sure really), but maybe it's not solving the right problem. I think the main problem in my design is matching the resistor pairs. That will do the most to increase common mode rejection. Well, these are my thoughts for the moment. I could be (and believe me, it's happened) completely wrong.

yep, just "simulated" this on paper - the inverting input is a virtual ground only as long as the noninverting stays at ground level. if you input, say 1v into the positive input of the summing amp, the noninverting input's (opamp) potential is at 0.87v - not exactly at ground level

i don't know what this means to the overall performance of this circuit now!