Weird Line Driver

[atavacron]

Y'know, for driving weird lines

I've been trying to figure out a way to do a sub-115dB line driver that doesn't load the preceding stage, doesn't load the inverter, and doesn't require adding gain at the non-inverting output to roughly match the bandwidth / phase rolloff of the inverting output. The circuit below meets those requirements. But I have no idea if it would work IRL.

There is precedent in the literature for returning negative feedback to the non-inverting input, but I haven't seen it done in a unity gain situation like this. And I haven't seen anyone use an inverting OPA1611/211 (40/45MHz) without at least 5.6pF in the feedback loop, but that is not possible here because the thing needs to be at least 5x the speed of the OPA1622 (8MHz) in order to avoid peaking. Even 604R/5.6pF is not an option. So stability is a question.

This circuit (if it works IRL) would be great with an OPA1692 (5.1MHz) and an LM4562 (55MHz) with good decoupling and just the teeniest-tinyest feedback capacitor at the inverter. That's the eye-opening cheap option, since OPA1692s should perform well in this position, but only if they are buffers! Obviously if you wanna make this already-kinda-spendy driver even more so, you could parallel another buffer + inverter, and maybe start coming close to the noise floor of the non-inverting OPA1622 half.

What do people think? Suggestions for improvement? Potential pitfalls?

1. Schematic
2. Curves without cable sim, before LR only
3. Curves with cable sim, before and after LR






PS - I suppose that's a not-quite-28dBu maximum output, as even the OPA1612 won't put out 22dBu as a voltage follower. 21.5dBu maybe. OPA828 buffer + OPA211 inverter could work. Also, re: the LR isolator peaking...I chose 20R/4u7 over the conventional 39R/4u7 because it's less wild with a 10nF/10nF/10nF load. Maybe a +5dB peak as opposed to +8dB or something. Best thing would be 10R/10uH but that's really only gonna work for the OPA1622 (which arguably might not even need the 5.1Ω buildouts); any other amp one might realistically use needs at least 25Ω Riso.

 

[atavacron]

By the way, the feedback loop seems to work just as well (if not better) with a composite inverter made from a medium speed precision amp and a fast amp with gain. OPA210 + THS4031, for instance. The Groner / Polak super low distortion composite worked well too, though I had to extend the bandwidth a bit - I think it's set up for 16MHz in the AES paper (?).

If you omit the feedback buffer (thus loading the negative output as one would in a normal inverting output scheme), there is better flexibility with amp types and compensation options. For instance, an OPA1611 or LM4562 set up with 604R / 5p6 causes no peaking with an OPA1622 as the driver, and you can even raise Rf and Rg a bit. But the buffer is worth it...the biggest advantage of this architecture is that both halves of the driver are unloaded, and that lets you use any slow dual amp with healthy drive capabilities to do the job heretofore only doable by the LME49724 (and now the OPA1633), but with 6dB less noise gain.

 

[ccaudle]

There is precedent in the literature for returning negative feedback to the non-inverting input, but I haven't seen it done in a unity gain situation like this.

It seems like asking for trouble with real devices. It is hard to know how accurate the SPICE models are. Have you run Monte Carlo simulations to see if that configuration stays stable across device and temperature variations? It is usually difficult to keep a feedback loop stable with two additional delays in the loop without some manipulation of the frequency response of the feedback network. The only frequency selective components are the R9/C1 LPF which generally is the wrong direction (adding a lag network usually makes the loop less stable).

a way to do a sub-115dB line driver

Missing some units there. Do you mean noise floor 115 dB below full scale output? -115 dBu RMS noise floor? A line driver that can get to just under 115 dBu? OK, that last one wasn't serious at all, but dB is a comparative measure with no intrinsic units, so you need to be explicit about what you are trying to achieve. 115 dB below full scale output (e.g. 24 dBu - 115 or -91 dBu) is much different than -115 dBu (which is a 139 dB dynamic range if you want the traditional 24 dBu output and is not practically achievable; not impossible, but not practical).

What do people think?

Personally I think you are trying to solve for a non-problem (which might be OK, I have my own windmill I've been tilting as well).

It would help if you were more explicit about what you do not like about the ATI cross-coupled output circuit that That Corp uses in their OutSmarts branded line drivers, and what you do not like about the traditional unity buffer, followed by another unity buffer for the hot out, and a unity inverter for the cold out.

Also why you feel you need to be able to drive 600 Ohm loads, especially when your simulation is only accounting for 10m of cable. The traditional use for 600 Ohm loads was so you could use telco lines to e.g. go from your remote gig back to the radio station studio. Almost all (or maybe all by now) telco lines you would use for that these days are digital, so you aren't really going to be running audio on a mile of 600 Ohm cable with transformers at each end.
I have seen a few people who insist on using 1960's designs without modification, so still using transformer coupled 600 Ohm inputs, but I don't have a lot of tolerance for those kinds of headaches today when every other piece of gear I would use has had 10k Ohm inputs for at least four or five decades.

Speaking of cable simulation, your schematic notes call out coax cable, which doesn't really make sense to connect to a balanced output on pro gear.
Something like Belden 2451 indoor/outdoor cable which is labeled as "stadium" cable (e.g. you might run it from stage to mix position, or stage to remote truck, so it is waterproofed) would be more likely. 110pF/m conductor to conductor, and 220pF/m conductor to others, which I would interpret as an additional 110pF from conductor to shield. So still about 1nF for 10m, but then scale that for the longest you might need to drive. I would want to make sure there are no oscillations even with 100m of reasonable cable, or about half that length of star quad cable.
Of course if this is for your own use and you don't think you would ever need to drive across a large auditorium or stadium you are free to limit the design capabilities to whatever you think makes sense.

Note that your notes in the schematic about currents are a little confusing, because the note is actually peak current, but the level is indicated in RMS voltage, so a little bit of mismatch in units.
Also that current level is only correct for a true resistive load, the current will be a little higher including the effect of the capacitance (which admittedly is less of an effect when designing for 600 Ohm load than it would be if you were designing for e.g. a 5k Ohm minimum load).

 

[JohnRoberts]

build it and see what happens...

JR

PS: Modern canned solutions are too good for me to measure

[edit: you mentioned Sam Groner... IIRC he had to develop his own low distortion sine wave source to measure high linearity gear. /edit]

 

[atavacron]

Have you run Monte Carlo simulations to see if that configuration stays stable across device and temperature variations?

Nope

It is usually difficult to keep a feedback loop stable with two additional delays in the loop without some manipulation of the frequency response of the feedback network. The only frequency selective components are the R9/C1 LPF which generally is the wrong direction (adding a lag network usually makes the loop less stable).

That’s the CM + diff input capacitance of the amp.

-115 dBu RMS noise floor?

Yes.

you need to be explicit about what you are trying to achieve.

A driver design block that I can drop in at the end of any of the chains I am working with across a variety of projects, without much concern about what it follows, or whether the user is using a 10K or 600R piece of kit at the end.

It would help if you were more explicit about what you do not like about the ATI cross-coupled output circuit that That Corp uses in their OutSmarts branded line drivers,

-105dBu noise floor

and what you do not like about the traditional unity buffer, followed by another unity buffer for the hot out, and a unity inverter for the cold out.

Phase difference at 20kHz when using anything but fast amps, multiplied by how many times you go into and out of a channel strip under real world analog mix conditions. Loading of both the buffer and the inverter when using low value resistors.

Also why you feel you need to be able to drive 600 Ohm loads, especially when your simulation is only accounting for 10m of cable.

I could have posted the previous one, which was 100M, but nobody does that. It was basically the same thing, but with bandwidth shrunk and the LR peak more extreme — like 5dB. From my console to patch is like 20’ of cable (40’ round trip), but from room to room in the studio it can be 75’ per jump. Plus 25’ of XLR on one end. Plus distance from patch to outboard, which is anywhere from 10’ to 25’. Anyhoo, driving 100’ of cable is typical and driving 150’ of cable is not out of the question, and it is reasonable to assume that the run will be comprised of cabling of a minimum professional quality but with a variety of capacitances. That’s my standard, rather than the much more common DAC-to-console-to-small-outboard-rig-to-console-to-ADC hybrid mix loop that keeps runs under 30’ish.

you aren't really going to be running audio on a mile of 600 Ohm cable with transformers at each end.

Correct. And there are white papers on how to do so.

Speaking of cable simulation, your schematic notes call out coax cable, which doesn't really make sense to connect to a balanced output on pro gear.
Something like Belden 2451 indoor/outdoor cable which is labeled as "stadium" cable (e.g. you might run it from stage to mix position, or stage to remote truck, so it is waterproofed) would be more likely. 110pF/m conductor to conductor, and 220pF/m conductor to others, which I would interpret as an additional 110pF from conductor to shield. So still about 1nF for 10m, but then scale that for the longest you might need to drive.

That comports. Maybe I put the wrong model #? 31pF/foot conductor to conductor. To correctly model it I’d have to add virtual caps between conductors and to shield at both ends, with resistance between. But it’s just shorthand anyways for “an unrealistic amount of cable.” Using AES cable helps, that’s under 20pF/ft typically. Anyways, we’re talking about the same thing.

Note that your notes in the schematic about currents are a little confusing, because the note is actually peak current, but the level is indicated in RMS voltage, so a little bit of mismatch in units.

Yes, I was skipping some math to show just the important bits.

Also that current level is only correct for a true resistive load, the current will be a little higher including the effect of the capacitance (which admittedly is less of an effect when designing for 600 Ohm load than it would be if you were designing for e.g. a 5k Ohm minimum load).

Great point! I will remember that. BTW the HF response of the sim after 100M of cable is the same regardless of whether it’s 600R or 10K at the end.

 

[atavacron]

[edit: you mentioned Sam Groner... IIRC he had to develop his own low distortion sine wave source to measure high linearity gear. /edit]

Yeah I don’t need the -180dBu THD but as soon as there is an application for an OPA211 + THS4031 composite, it’s common sense to see what that amp does. I suppose it would be smart to accurately model a regular composite and that composite, with common values between the two, and see the difference. I assume that even without the Groner/Polak RC networks, those two amps together are already batting at a pretty good average. I was impressed that I could eliminate peaking in my application by wrapping an 18MHz part (OPA210, which is leagues more stable than OPA211) around a 200MHz part running at a gain of just over 2. Though I have not nailed down a good scenario with the buffer - just the inverting composite in the FB loop. It’s nice in theory to be able to use a more stable amp in the inverting position, but making that loop out of an OPA211, an OPA210, and a THS4031 would just be unrealistic for common deployment.

I’m hoping that maybe someone can suggest a solution for this that just involves two duals, but is cheaper than what I’ve drawn, and doesn’t have the output limitations and Iq of the LM4562 in the feedback role. I did look at the ADA4898 but it appears that it’s way way slower at Av=-1 than at Av=1 (65MHz). And at 8mA Iq there is little advantage over a 4031-based solution. Maybe there is a way to do the feedback loop with discretes that is more to the point — though I generally try to avoid electrolytics.

 

[atavacron]

An updated version. TI sim'd it, seems to work virtually for them as well. Worst case scenario graphed; BW looks a lot nicer with only 3M of cable at the back end.

If anyone can come up with a cheap 'n' cheerful version of this, I'm all ears. OPA2227 + LM4562 could be just the ticket in DIP. Not so cheap, but the speeds are right. Also, OPA1692 + OPA1602. Speed ratios work, and pretty affordable.

 

[atavacron]

@ccaudle Keep tilting at windmills long enough, the windmill tilts.

Couldn't avoid the output loading (at least on one side), but I did get my hi-z input. Iq = 15.7mA. I don't think this is limited to the OPA1622, response is pretty close to the graph with 10MHz and 12MHz amps. The servos bring DC to ~5µV with typical offset all around. Values & authority adapted from the single node servo @gyraf posted once upon a time, which I have come to realize were chosen for phase purposes more than bandwidth.