transformer for I/V conversion in DAC?

[dirkwright]

Well, given the mostly negative attitudes towards what I intend to do, I think I'll try to find somewhere else to talk about it some more. You can bash Analog Devices all you want, but I know two former engineers from that company and they are not foolish or stupid. Making offensive comments towards a well respected company says more about you than about them. Thanks for all of your replies.

 

[Andy Peters]

Well, given the mostly negative attitudes towards what I intend to do, I think I'll try to find somewhere else to talk about it some more. You can bash Analog Devices all you want, but I know two former engineers from that company and they are not foolish or stupid. Making offensive comments towards a well respected company says more about you than about them. Thanks for all of your replies.

Wow, you completely misunderstood, and mis-took, JD's comment. You might want to re-read it.

I am sure the Gearslutz crowd will be glad to discuss your ideas at such a detailed level as you've seen here. Good luck.

-a

 

[Samuel Groner]

You can bash Analog Devices all you want, but I know two former engineers from that company and they are not foolish or stupid.

No one was bashing ADI for this app note. It is, as has been pointed out, directed towards the implementation of high-frequency (dozen or hundreds of MHz) converters. Such applications have entirely different goals and difficulties than audio. We speaked against the misapplication of the presented information to audio converters.

Samuel

 

[dirkwright]

You can bash Analog Devices all you want, but I know two former engineers from that company and they are not foolish or stupid.

No one was bashing ADI for this app note. It is, as has been pointed out, directed towards the implementation of high-frequency (dozen or hundreds of MHz) converters. Such applications have entirely different goals and difficulties than audio. We speaked against the misapplication of the presented information to audio converters.

Samuel

I can't find where in this app note it says that the discussion is limited to RF frequencies. Do you assume it is limited because the examples show a load impedance of 50 ohms?

 

[dirkwright]

Well, given the mostly negative attitudes towards what I intend to do, I think I'll try to find somewhere else to talk about it some more. You can bash Analog Devices all you want, but I know two former engineers from that company and they are not foolish or stupid. Making offensive comments towards a well respected company says more about you than about them. Thanks for all of your replies.

Wow, you completely misunderstood, and mis-took, JD's comment. You might want to re-read it.

-a

Well, maybe I did. If so, I apologize. Even so, if the app note is ONLY directed at RF DACs, so what? Why does that make a difference? The app note is so abstract that it appears to be applicable to ANY DAC.

 

[Samuel Groner]

I can't find where in this app note it says that the discussion is limited to RF frequencies. Do you assume it is limited because the examples show a load impedance of 50 ohms?

Indeed it is not explicitly stated that this makes only sense for RF, but if 70% of a text is concerned with impedance and impedance matching you can be quite sure that this is not 20 Hz-20 kHz.

Why does that make a difference? The app note is so abstract that it appears to be applicable to ANY DAC.

Because a RF DAC and an audio DAC are quite different beasts, and because the performance requirements for RF and audio are different too. Of course you can, in principle, apply this to at least some audio converters (and it has been done), but that doesn't mean that it makes much sense. Surely distortion, but probably also noise, will be *way* higher compared to a good active IV-converter. Audio DACs are designed to see a very low load impedance, and if you don't provide this, it will not perform as well as it could.

Samuel

 

[Gold]

Again that sounds like an effect that would benefit from a bypass switch.

A good transformer is more like a blend control than an In or Out device. Take it easy with level and the good ones are plenty clean. Drive them hard and you get stuff that sounds better than a hard clip.

 

[dirkwright]

Because a RF DAC and an audio DAC are quite different beasts, and because the performance requirements for RF and audio are different too. Of course you can, in principle, apply this to at least some audio converters (and it has been done), but that doesn't mean that it makes much sense. Surely distortion, but probably also noise, will be *way* higher compared to a good active IV-converter. Audio DACs are designed to see a very low load impedance, and if you don't provide this, it will not perform as well as it could.

Samuel

My simulations in spice show just the opposite. Noise and distortion are far lower with a simple step up transformer (non-linear spice model of the transformer) followed by a discrete buffer vs. an active I/V stage copied from the DAC data sheet. The active I/V stage shown (as well as many others) appears to be a modification of a normal instrumentation amplifier. I take it from this discussion that most people here prefer to use op amps and other chips. Is that correct?

My simulations show a S/N of 180dB @ 1kHz for my circuit. Distortion is 0.001% with 1.2 Vrms out and 7.8mA input. Obviously this is an ideal situation and would not be realized in reality.

 

[JohnRoberts]

I am not a fan of relying too heavily on simulations (i'm old).

What bandwidth is your 180 dB S/N specified over?  Is it possible you used a 1 Hz bandwidth? That number sounds much too good for any realistic bandwidth. 

If you don't like the advice you get here, go ahead and built one and let us know how it works out.

JR

 

[abbey road d enfer]

The active I/V stage shown (as well as many others) appears to be a modification of a normal instrumentation amplifier.

Although it is a three opamp configuration, it has in common with an instrumentation amplifier only the final differential stage. The input stages are totally different. They are current-input and they are not floating with regard to ground, which is the most important feature of an instrumentation amplifier, but of no value in an I/V converter.

I take it from this discussion that most people here prefer to use op amps and other chips. Is that correct?

That's the way the chip's designers have developped it, and the only way they will approve a design.
For DIY, you don't care about approval. As far as most of us are concerned, you could use whatever you want, it's your privilege.
Now, I don't see the point in introducing xfmrs if you must have buffers anyway.

My simulations show a S/N of 180dB @ 1kHz for my circuit.

As JR said, I think there must be a mistake somewhere.

 

[ricardo]

My simulations show a S/N of 180dB @ 1kHz for my circuit.

That's awesome!

The best a properly dithered 24b channel can do is -141.5dB and only an experimental Panasonic (which was declared unmanufacturable) has got to within 1dB of this.  The very best commercial DACs today are maybe a dB better than 20b at -117.4dB.

You'd better patent this immediately.  Then build one and tell us how it performs in real life.  8)

Noise over 20kHz wrt to FS sine wave.

 

[Andy Peters]

My simulations show a S/N of 180dB @ 1kHz for my circuit.

Dynamic range for 24 bits is 144 dB.

-a

 

[ricardo]

dynamic range for 24 bits is 144 dB.

Actually it's 141.5dB from the rms value of optimal TPD dither to a FS sine wave measured over the Nyquist bandwidth.

The figure for 16b (cos you can find real life 16b converters that do this) are 93.3dB and NOT 16x6.01=96.16dB.  See the earlier papers by Lipsh*tz & Vanderkooy on on dither.

If you measure a 16b 44.1kHz channel and find less than -93.3dB noise over 22.05kHz, it is improperly dithered and will exhibit clearly audible digital distortion well before you reach the noise level.

 

[JohnRoberts]

The rule of thumb is 6dB per bit, which is close enough for guvmint work and to figure that it ain't 180 dB  which would be like 30 bits .  . That 180 dB number is closer to the noise of a 50 ohm resistor in a 1 Hz bandwidth thus my question about measurement BW.

How many angels do fit on a pinhead?

JR

 

[abbey road d enfer]

Woah! We're all shooting at poor dirkwright.
I think he means the performance of his i/v converter, excluding the DAC noise.
Indeed, he can't be right, just the DCr of his xfmr would cause more noise than -180dBu.
In order to achieve -180dBu noise, the cumulated resistance from input to output would have to be less than 0.02 ohm, not taking into account any active element.

 

[Samuel Groner]

My simulations in spice show just the opposite. Noise and distortion are far lower with a simple step up transformer (non-linear spice model of the transformer) followed by a discrete buffer vs. an active I/V stage copied from the DAC data sheet.

First of all, I think you're not understanding the problem. What we are concerned with is the interface between the DAC and the IV converter. You cannot simulate/measure the IV converter alone and conclude that all is well. The voltage at the output terminals of the DAC must not see any apprechiable voltage swing, or excess distortion will be generated. This is because the output Z of the DAC is word- and output voltage dependent.

Second, it must be appreciated to which extent distortion is simulated by which models. Standard opamp models do not include nonlinear behaviour (beyond basic input/output limitations such as clipping, output current limiting and slew-rate). I have no experience with the non-linear transformer model of LTSpice, but unless you have compared simulated and measured distortion figures of a specific transformer you have no clue if you're on the right track. How do you determine the parameters of this model?

This means that you simply cannot simulate the performance difference between an active IV converter and a transformer implementation. You have to build it and measure it.

Samuel

 

[dirkwright]

My simulations in spice show just the opposite. Noise and distortion are far lower with a simple step up transformer (non-linear spice model of the transformer) followed by a discrete buffer vs. an active I/V stage copied from the DAC data sheet.

First of all, I think you're not understanding the problem. What we are concerned with is the interface between the DAC and the IV converter. You cannot simulate/measure the IV converter alone and conclude that all is well. The voltage at the output terminals of the DAC must not see any apprechiable voltage swing, or excess distortion will be generated. This is because the output Z of the DAC is word- and output voltage dependent.

Second, it must be appreciated to which extent distortion is simulated by which models. Standard opamp models do not include nonlinear behaviour (beyond basic input/output limitations such as clipping, output current limiting and slew-rate). I have no experience with the non-linear transformer model of LTSpice, but unless you have compared simulated and measured distortion figures of a specific transformer you have no clue if you're on the right track. How do you determine the parameters of this model?

This means that you simply cannot simulate the performance difference between an active IV converter and a transformer implementation. You have to build it and measure it.

Samuel

Spice models for the DAC are not available. Therefore, the output has been simulated with a current generator with 1Mohm source impedance. That is the best that this software will do. Spice models are not available from the transformer manufacturers.

OK, so you're basically telling me that I'm wasting my time since it is not possible for me to purchase the required instrumentation to measure the distortion in an actual circuit to these levels. Therefore, I don't see a reason to continue this discussion.

 

[ricardo]

OK, so you're basically telling me that I'm wasting my time since it is not possible for me to purchase the required instrumentation to measure the distortion in an actual circuit to these levels. Therefore, I don't see a reason to continue this discussion.

No. No No Dirk.  This is just us showing our prejudices & opinions.  None of us have any FACTS about this subject as none of us has tried it.

What we'd REALLY like is for you to make one and report on its performance.  Only then will the store of human knowledge have increased.

RMAA, if used (and interpreted) carefully, can certainly tell you whether its viable.

But at the end of the day, you may find its no better than a simpler cheaper standard solution.

 

[JohnRoberts]

This means that you simply cannot simulate the performance difference between an active IV converter and a transformer implementation. You have to build it and measure it.

Samuel

Spice models for the DAC are not available. Therefore, the output has been simulated with a current generator with 1Mohm source impedance. That is the best that this software will do. Spice models are not available from the transformer manufacturers.

OK, so you're basically telling me that I'm wasting my time since it is not possible for me to purchase the required instrumentation to measure the distortion in an actual circuit to these levels. Therefore, I don't see a reason to continue this discussion.

If course you "could" simulate anything given enough computer power and an accurate model. My early reluctance to use simulations was because back then you needed to pretty much understand the behavior to build an accurate model. So why bother for one-off designs (while i did find it useful for designing complex multipole filters using practical real world values).

You are trying to sim a circuit that the model does not accurately describe, as evidenced by your impossible (30 bit equivalent?)  result. You may be surprised, but when designing anywhere near the bleeding edge of technology, designers often turn off the computer and melt solder to get a real world result. At the end of the day it is the real world behavior that matters.

You can continue arguing the hypothetical, or melt solder and see what happens. While it may appear that we are ganging up on you, you are receiving the benefit of many decades of experience that are pretty much in agreement, while inclined to have too much fun with this.

You don't need to convince us, it's your time and money.

JR

PS: Do not ignore the earlier suggestion http://www.groupdiy.com/index.php?topic=48909.msg618046#msg618046  to look at transformer in virtual earth configuration as perhaps more appropriate for your application. 

 

[Samuel Groner]

OK, so you're basically telling me that I'm wasting my time since it is not possible for me to purchase the required instrumentation to measure the distortion in an actual circuit to these levels. Therefore, I don't see a reason to continue this discussion.

If you build it, you can send it to me and I'll do the measurements. Promised.

Samuel