[Garbz]

Some of you may have noticed that I have talked a lot about DACs recently, having built one a while ago but never actually posted the design. I was going to post when I was finished but the DAC became a proto-design for anything DAC related. It now has 3 other boards hanging off it (usb, I/V stage, S/PDIF interface) with another to come. However with all the recent talks on output stages and the various designs I have seen using opamp I/V stages I thought I should post something about my recent experiments.

My most recent experiment, a few months in the making, has been a separate power-supply and modular I/V section which would allow me to compare various designs of I/V stages available and determine the “best” output. The original DAC had an OPA627 followed by an emitter follower, and incorporated a 2-pole low-pass filter tuned at around 40khz running from a +/-15V PSU. The new power-supply design is +/-40V and allows me to test a few stages like the Passlabs D1 stage, a cascoded variation of Jocko Homo's stage, Pedja Rogic's stage, and Hawksford’s Stage presented in the paper “Current steering transimpedance amplifiers for high-resolution digital to analogue converters”, (AES, 2000), which I will not post but feel free to contact me for it. Unfortunately the project was put on the backburner because of studies, an upcoming holiday, and a lack of funds. I hope to pick it up again next year but so far the only stage finished is the Passlabs D1 stage, which I chose first because I had IRF610Bs laying around.

There are various ways I/V can be done after the DAC, the simplest being passive I/V. Just put a resistor on the output and the current creates a voltage drop across it. The problem with this method comes down to output compliance of the DAC. Most DACs out there are designed with a perfect I/V stage in mind. They perform at their best with a direct short to their common voltage which allows them to dump all their available current with ease. Resistor I/V stages present an output impedance to the DAC which prevents this, and the most common complaint sound wise is a loss of dynamic range. That said some DACs work rather well in this configuration.

The most common design and indeed the design recommended in most datasheets are opamp I/V stages. Opamps present the voltage connected to their positive terminal to the source. So with that connected to the source’s VCom pin it becomes an ideal I/V stage. The other advantage is the ability to create complex multi pole filters, however this is usually done at a gain stage after I/V anyway. The problem with opamps is their slew rate and bandwidth combined with negative feedback. Opamps need to be very fast (reads expensive) to work properly in this application. A discussion on that is currently happening http://www6.head-fi.org/forums/showt...=194150&page=2. Further information is also available in the Hawksford paper.

Transformer I/V stages are another option however at $200+ for a set of decent transformers they are prohibitively expensive. While they form their own filter they also suffer the same flaws as the resistor passive I/V.

The final option is discrete I/V stages which do not come without their set of problems, the main being complexity and difficulty to design. The general designs appear to be variations of a trans-impedance amplifier which presents an ideal low-impedance load to the source (usually setup as common gate or common base amplifiers with the input going into the emitter) while at the same time providing the compliance necessary for a passive resistor I/V.

The Passlabs D1 stage which I have completed provided a massive improvement over the conventional opamp (OPA627) circuit, and it actually worked out cheaper too. It is a MOSFET based design, with a common gate input and a source follower output I think (FETs are still somewhat alien to me), and it also incorporates a 2nd order low-pass filter though from the looks of things a 3rd pole should be easy to add. Sound-wise the difference mainly lays in the clarity of the sound, not so much with a difference in brightness or bass response or anything. I have only had it for a day but the more I compare the more I find subtle improvements over the conventional opamp stage like instrument separation and depth.

I look forward to completing the other 3 stages too and see how well they work, thus far however I do not think I will ever consider opamp I/V for any future project anymore. Picture:

 

[Pars]

Nice work Garbz. I have been going down the same path and looking at Jocko's stage in particular for implementation into my Rotel. Only problem is, I haven't actually done anything yet

. (Well, I did toss together one channel on a breadboard, but I haven't tested it yet, and it is a real rough approximation).

 

[t52]

could you post a schematic of your actual implementation of the passlabs d1?

 

[Garbz]

The link at the top to Passlabs D1. Page 2, the post by acaudio had the schematics linked.

The exact implimentation is as given with some value changes. R5 and R18 are reduced to 1k. This gives me 1/5th of the distortion and also brings the volume down since the output was too hot.

/EDIT: Sorry no it isn't that is not the original Passlabs schematic!!! The passlabs schematic did not have a CCS on the source follower.
http://www.garbz.com/amp/dac/passd1-mod.pdf

R2 dropped to 1K to reduce volume.
R6 dropped to 1K from 1.5k because for whatever reason I reduced it originally trying to get the volume down thinking it was R2. Volume was the same but bass was punchier. When I corrected the mistake it was back to it's old self so I went in and again dropp this value to 1k. It would probably work best with a constant current source.
R3 dropped to 81k from 100. The problem was that I only had 20k pots on hand. setting these to 20k would not give me 1.9V on the input, and not the 2.48V that the PCM1730 requires as it's VCom voltage.

 

[t52]

thanks for the schematic garbz! i know i'm lazy - on the other hand a correct schematic where i don't have to switch values here and there makes it less error prone...
what does Res1 on your resistors mean - 1w?
how do you sum the two outputs up - opamp?
vcc/vss are +/-40v in your cse, right?

 

[Rescue Toaster]

Quote:

Originally Posted by t52 thanks for the schematic garbz! i know i'm lazy - on the other hand a correct schematic where i don't have to switch values here and there makes it less error prone... what does Res1 on your resistors mean - 1w? how do you sum the two outputs up - opamp? vcc/vss are +/-40v in your cse, right?

I think the voltages are 30 iirc. And actually the D1 has no summing, the output caps go straight to XLR's (or whatever). You could always use a line out transformer, as the D1 has low output impedance some 600:600 output transformers should work fine and would let you have SE outputs as well. Another active stage wouldn't be a good idea IMO.

Also, you can put a log pot/switched attenuator (rheostat config) from the bottom of R2 to the bottom of the inverted side's R2, giving volume control. I forget the exact values, you'd have to sim it to find the range needed, though I think Garbz was messing with that. It's a nice way to do it.

 

[Garbz]

I don't sum the outputs. My system is fully balanced and I use all 4 channels. If you wish to cut costs and not sum anything you could always just ignore 2 channels which is what I do when I connect a single ended device. Opamp for summing will work too. Opamps are bad for I/V not for differential summing

. Also some kind of summing is probably a good idea. My Iout from DAC to I/V stage is only some 3" long but is picking up huge loads of hum which is reduced a LOT because the hum is common to both +ve and -ve.

The original powersupply called for +/-30V but I had +/-30V transformers and did not want a huge voltage drop over the regulators. It seems now though they aren't very warm during operation as I first miscalculated they would be so I will probably drop the voltage a bit from my +/-40V

Res1 is just the package descriptor used in Altium. I normally hide these before I print, but I forgot in this case. All resistors are 1/4watt metal film.

If you use the volume control idea then change R2 to 1500ohm rather than the reduce 1k.

 

[00940]

Before drawing broad conclusions, I'd try other opamps. Many are cheaper and better than the opa627, at least for I/V. Try the lm6172, the opa1632 since you're balanced, the ad8065.

I tried jocko's stage, rbroer simple stage for tda1543 and neither were clearly better than a good opamp (I'm not quite a fan of the opa627 as I/V). And in cdp, as you need a new PS, discrete always end up to be more expensive.

 

[Garbz]

Well actually I am quite content with completely writing off opamps, however as I mentioned this was about I/V stage rolling. Next year before I settle on a final design for my next DAC I will do considerably more testing of various opamp and such stages to draw a more statistically significant conclusion.

Although the reasons I am quite content with writing off opamps is because of the serious flaws they induce. There are quite a few AES papers published about this, one mentioned at the top which lead me to think that no opamp other than the "ideal" opamp is as good for I/V stages as a nofb discrete design. I did breadboard an OPA2604 when I first assembled my DAC. This option used in many cdplayers sounded poor in comparison to the OPA627 stage. I should also mention that the OPA627 is a chip I have enjoyed in a lot of other applications, but it was shamed by the D1 I/V variant I have here.

On the cost side of things a PSU is a PSU. The only difference between the one that was powering the opamps and the one powering the current I/V stage is the opamps psu had a smaller transformer and was more expensive being a jung regulator. Then it comes down to parts. IRF610s are $1.70ea. The OPA627 considerably more expensive. I'm not saying this goes for all stages, but I will say we are audiophiles, since when did cost matter

Btw as a matter of interest which Jocko's stage did you try? There are a few thanks mainly to the original posted by Jocko being workable but far from complete.

/EDIT: Btw will update the schematics with a few minor changes thanks to RescueToaster on the weekend

 

[00940]

I tried the original, but replacing the resistor in a ccs with a pot of course and adding a buffer at the output. You also need to add a cap in parallel with the I/V resistor.

I know there were quite a few articles published pushing against opamps. You can add Jung and Gilbert's to Hawksford's. But considering these thoughts by Andy_c : http://www.diyhifi.org/forums/viewtopic.php?t=262 were not contested by Jocko nor Jung and considering some opamps sound quite good as I/V, I'm getting a bit more skeptical.

edit : I agree about the 2604. Worse than opa627 in this application. Many people seem to like to have a big veil on their music

 

[Rescue Toaster]

Quote:

Originally Posted by 00940 I tried the original, but replacing the resistor in a ccs with a pot of course and adding a buffer at the output. You also need to add a cap in parallel with the I/V resistor. I know there were quite a few articles published pushing against opamps. You can add Jung and Gilbert's to Hawksford's. But considering these thoughts by Andy_c : http://www.diyhifi.org/forums/viewtopic.php?t=262 were not contested by Jocko nor Jung and considering some opamps sound quite good as I/V, I'm getting a bit more skeptical. edit : I agree about the 2604. Worse than opa627 in this application. Many people seem to like to have a big veil on their music

Andy_c's analysis is fundamentally flawed in that not only does 8x oversampling not reduce slew requirements to 1/8, but a Sigma Delta modulator like the pcm1794 that he mentions could actually slew full scale more than once per sample received from the 8x FIR filter.

That said, I'm not saying that people's ears are broken and that all opamp I/V automatically sounds horrible. But there is really no way ir can compete, and there's hardly reason to use opamps when alternatives are not that much more expensive. Except I guess in portables.

 

[00940]

The question is : what kind of signal requires to output such crazy pattern ?

If I understand you well, you're raising the same objection as TCD333 (post 22), answered in post 23.

 

[Rescue Toaster]

Quote:

Originally Posted by 00940 The question is : what kind of signal requires to output such crazy pattern ? If I understand you well, you're raising the same objection as TCD333 (post 22), answered in post 23.

Any signal really. We should expect the S.D. modulator to create all kinds of patterns with the levels of oversampling (64? 128?) & noise shaping we're talking about. And andy_c talked about the sample rate, but incorrectly re-asserted that the step size is limited. The max step size for a Sigma Delta dac is pretty much always going to be full scale, and even a straight multibit on an 8x oversampling filter will see steps of at least 1/2 full scale, probably higher.

 

[00940]

do you have any links describing what's happening /after/ the output of the modulator ? An LPF is part of the DS process, do the DACs fully rely on the analog stage outside the chip to do it ?

 

[jcx]

a interesting looking DAC document that pulls together several good dac internals descriptions and specs:

http://www.iet.ntnu.no/~ivarlo/files...t_audiodac.pdf

AndyC's analysis @diyhifi is on the right track but ignores output Z of real op amps.

for the op amp based I/V converter it is interesting that so many have forgot/never learned what was commonplace app note knowledge ca 1970:

In the early days of monolithic op amps and Iout dacs it was well known that the highly desirable low pass filter forming feedback C in the op amp I/V is a feedforward short circuit for high frequency switching events, this majorly jerked around the asymmetric NPN output stages of early op amps causing distortion, the known solution was to add a low Z Class A buffer in the loop biased to give a low Zout with the worst case current step