Hello Folks,
Yes this post does have a Pass tie in, but you may have to look for it. 🙂
A year or so ago I took a good look at the THS41xx series of fully differential op amps from TI and though that It would be great if I could scale it up so that it could handle more voltage swing. That way you could put it in front of something like a F4 or any other low/unity gain power amp, or even simply use it as a driver/front end for any number of amplifier topologies.
My design is a clone of sorts, but many of the details are left out of the THS41xx data sheets, so I had to still do a lot of work myself.
I think this design has a lot of potential as a driver/preamp, so I thought it would be good to share it and get some feedback from the gurus. 🙂
Let the fun(or flames possibly) begin.
I have tested the circuit as shown, it does work. 🙂 Some resistors were replaced with pots and adjusted for desired bias. Otherwise, it was built almost exactly as shown.
One key to this amp is the common mode error amplifier which brings the output of each side to 0V (within < 1mv on the prototype) relative to GND.
You don't need an output cap.
I used 20V rails, but it would not at all be difficult to go higher, say 35V-50V.
Also significant is that the design can swing very close to the rails (easily within 1V) which means with just 25V rails you could drive an F4 pretty well. Also note that you can attain extremely low offset without the need for any caps in the signal path.
One more cool factoid. The driver easily tolerates the 3V common mode DC at the inputs I through at it. This means that I can DC couple this driver from my DAC with no cap on input as the 2.5V common mode offset is completely nulled and not just relatively(output to output), but at each output to GND. 🙂
I can't think of a good name for it yet... I will entertain ideas.
Cheers!
Russ
Yes this post does have a Pass tie in, but you may have to look for it. 🙂
A year or so ago I took a good look at the THS41xx series of fully differential op amps from TI and though that It would be great if I could scale it up so that it could handle more voltage swing. That way you could put it in front of something like a F4 or any other low/unity gain power amp, or even simply use it as a driver/front end for any number of amplifier topologies.
My design is a clone of sorts, but many of the details are left out of the THS41xx data sheets, so I had to still do a lot of work myself.
I think this design has a lot of potential as a driver/preamp, so I thought it would be good to share it and get some feedback from the gurus. 🙂
Let the fun(or flames possibly) begin.
I have tested the circuit as shown, it does work. 🙂 Some resistors were replaced with pots and adjusted for desired bias. Otherwise, it was built almost exactly as shown.
One key to this amp is the common mode error amplifier which brings the output of each side to 0V (within < 1mv on the prototype) relative to GND.
You don't need an output cap. I used 20V rails, but it would not at all be difficult to go higher, say 35V-50V.
Also significant is that the design can swing very close to the rails (easily within 1V) which means with just 25V rails you could drive an F4 pretty well. Also note that you can attain extremely low offset without the need for any caps in the signal path.
One more cool factoid. The driver easily tolerates the 3V common mode DC at the inputs I through at it. This means that I can DC couple this driver from my DAC with no cap on input as the 2.5V common mode offset is completely nulled and not just relatively(output to output), but at each output to GND. 🙂
I can't think of a good name for it yet... I will entertain ideas.
Cheers!
Russ
Hi Russ,
Following some discussion with CBS240 here :
http://www.diyaudio.com/forums/showthread.php?postid=1224020#post1224020
I became more and more interested by Common Mode Feedback Loops. Post #40 showed a proposal by Win de Jager for an integrated implementation of CMFB.
Previously, same Wim de Jager who is consistent in his ideas had published an "hybrid power amplifier" (bipolar differential input directly coupled to the valve push-pull output stage) where the valve grid DC voltage is maintained fixed by a CMFB using a simple bipolar transistor :
Electronics World, November 1996, p897-900.
I was thinking to open a new thread about it with something like "A new input stage ?" as title.
I will show Jager's input stage in a following post.
I then became aware of your Twisted Pair Audio site and the existence of IC like THS41xx and OPA1632 and its incredible performances.
By pure coincidence, I spent the three last days playing with a simulator and the whole idea. The feature of having a stable DC output without resorting to an overall DC feedback is very interesting. However, as being a feedback applied to a load, how can it be called ?
By the way, looking at your schematics, I am a bit puzzled to see capacitors C1, C2 and C3 connected to ground. I also have been thinking of a more traditionnal two stage architecture rather than a folded cascode scheme, with two CMFB loops, one at the load of the input stage like Jager's circuit and one around the output stage.
I would be glad to see some power supply rejection and distorsion figures.
Regards
Following some discussion with CBS240 here :
http://www.diyaudio.com/forums/showthread.php?postid=1224020#post1224020
I became more and more interested by Common Mode Feedback Loops. Post #40 showed a proposal by Win de Jager for an integrated implementation of CMFB.
Previously, same Wim de Jager who is consistent in his ideas had published an "hybrid power amplifier" (bipolar differential input directly coupled to the valve push-pull output stage) where the valve grid DC voltage is maintained fixed by a CMFB using a simple bipolar transistor :
Electronics World, November 1996, p897-900.
I was thinking to open a new thread about it with something like "A new input stage ?" as title.
I will show Jager's input stage in a following post.
I then became aware of your Twisted Pair Audio site and the existence of IC like THS41xx and OPA1632 and its incredible performances.
By pure coincidence, I spent the three last days playing with a simulator and the whole idea. The feature of having a stable DC output without resorting to an overall DC feedback is very interesting. However, as being a feedback applied to a load, how can it be called ?
By the way, looking at your schematics, I am a bit puzzled to see capacitors C1, C2 and C3 connected to ground. I also have been thinking of a more traditionnal two stage architecture rather than a folded cascode scheme, with two CMFB loops, one at the load of the input stage like Jager's circuit and one around the output stage.
I would be glad to see some power supply rejection and distorsion figures.
Regards
I have been thinking very similar thoughts.
See http://www.battletonphoenix.co.uk/linebal.pdf - the design would run happily off +/- 20V with transistors shown.
Only simulated, not built.
See http://www.battletonphoenix.co.uk/linebal.pdf - the design would run happily off +/- 20V with transistors shown.
Only simulated, not built.
Hi, Russ,
Nice design 😀 Is it intended for classA output stage? I'm not sure, but maybe if it is used for classAB, the folded cascode is not supplying enough gain for fixing output stage non-linearity. For classAB, common emitor for 2nd stage is more like it (but you need to interchange the collector takeoff point).
Your DC servo is nice, how about replacing R10 with current mirror like Extrema power amp (Bruno Putzey's)?
Nice design 😀 Is it intended for classA output stage? I'm not sure, but maybe if it is used for classAB, the folded cascode is not supplying enough gain for fixing output stage non-linearity. For classAB, common emitor for 2nd stage is more like it (but you need to interchange the collector takeoff point).
Your DC servo is nice, how about replacing R10 with current mirror like Extrema power amp (Bruno Putzey's)?
Thanks for all the information and feedback so far.
My intent is to drive 4 F4 amps in 2 pairs differentially for 2 bridged channels.
One idea I have is to take feedback after the output of the F4s and thus wrap them in the feedback loop. I am wondering what sonic effects this will have. I am imagining I will be substituting even order harmonics for some odd, but hopefully at much diminished levels.
I do not have any test gear to speak of, but I will simulate CMRR and PSRR when I get a chance.
I may add a simple class A buffer stage at each output. Not sure yet. 🙂
Right now I have the prototype driving two simple LM3886 based class AB amps with the amps enfolded in the feedback as described before. This is working and sounds very good indeed. Each power amp stage is setup for a voltage gain of 6X. I would say this setup sounds no worse then the THS4131 in the same configuration. I will listen for a bit longer before I say it sounds better. 🙂
I also thought about a current mirror at the servo, but when I simulated it I could not keep it stable without compensation, and I hated the idea of adding a cap if I did not have to. I may try it again later. 🙂
Cheers!
Russ
My intent is to drive 4 F4 amps in 2 pairs differentially for 2 bridged channels.
One idea I have is to take feedback after the output of the F4s and thus wrap them in the feedback loop. I am wondering what sonic effects this will have. I am imagining I will be substituting even order harmonics for some odd, but hopefully at much diminished levels.
I do not have any test gear to speak of, but I will simulate CMRR and PSRR when I get a chance.
I may add a simple class A buffer stage at each output. Not sure yet. 🙂
Right now I have the prototype driving two simple LM3886 based class AB amps with the amps enfolded in the feedback as described before. This is working and sounds very good indeed. Each power amp stage is setup for a voltage gain of 6X. I would say this setup sounds no worse then the THS4131 in the same configuration. I will listen for a bit longer before I say it sounds better. 🙂
I also thought about a current mirror at the servo, but when I simulated it I could not keep it stable without compensation, and I hated the idea of adding a cap if I did not have to. I may try it again later. 🙂
Cheers!
Russ
Power amp modification
This has only been simulated, and I would probably change some devices to beefier types (cascode and CCS transistors). I would also probably parallel output devices, and possibly use drivers for the FETs so I can lower the current for the folded cascode.
One could bias the output devices class A or class A/B.
Anyway, it simulates very well.
Cheers!
Russ
This has only been simulated, and I would probably change some devices to beefier types (cascode and CCS transistors). I would also probably parallel output devices, and possibly use drivers for the FETs so I can lower the current for the folded cascode.
One could bias the output devices class A or class A/B.
Anyway, it simulates very well.
Cheers!
Russ
Attachments
Neat.
A couple of details. I would bump the gate stoppers up from 47Ohms to maybe 220 - depending on layout etc, you may be marginal for stability there.
You have a big cap (C4) limiting the BW of the symmetrising circuit, so it just does offset control. In my variant, I run that full BW; it then wipes out asymmetry dynamically. If you go this way, you can drop the values of R7 and R8, and remove R15, so bias current in that path no longer gives offsets.
If you put emitter resistors under Q5 and Q6 you should get more OL linearity and reduced capacitance (I think). R10 would have to go up a bit to compensate. S
A couple of details. I would bump the gate stoppers up from 47Ohms to maybe 220 - depending on layout etc, you may be marginal for stability there.
You have a big cap (C4) limiting the BW of the symmetrising circuit, so it just does offset control. In my variant, I run that full BW; it then wipes out asymmetry dynamically. If you go this way, you can drop the values of R7 and R8, and remove R15, so bias current in that path no longer gives offsets.
If you put emitter resistors under Q5 and Q6 you should get more OL linearity and reduced capacitance (I think). R10 would have to go up a bit to compensate. S
That is the sort of thing.
I notice you have gone to a current mirror in the front of the symmetriser loop. This presumably puts the open loop gain up a lot; as this loop basically runs at unity gain, do you need/want this much gain?
I haven't played with it in the simulator so I don't have a feel for the pros and cons.
On the other side, the base of Q7 and Q8 is biased by a LED, to presumably round about 1.4V below the rail voltage. But the base voltage of the folded cascode transistors Q3 and Q4 is set by two Si diodes to be roughly the same voltage - shouldn't you swap the 1N4148 for green LEDs, so that there is a sane VCE on Q7 and Q8. Or have I misread the circuit?
I notice you have gone to a current mirror in the front of the symmetriser loop. This presumably puts the open loop gain up a lot; as this loop basically runs at unity gain, do you need/want this much gain?
I haven't played with it in the simulator so I don't have a feel for the pros and cons.
On the other side, the base of Q7 and Q8 is biased by a LED, to presumably round about 1.4V below the rail voltage. But the base voltage of the folded cascode transistors Q3 and Q4 is set by two Si diodes to be roughly the same voltage - shouldn't you swap the 1N4148 for green LEDs, so that there is a sane VCE on Q7 and Q8. Or have I misread the circuit?
PigletsDad said:
Yes on both fronts. 🙂
I have since changed back from the current mirror on VCM error amp. I also found that the amp was not stable unless I added some resistance on the grounded side of that diff pair. I am not sure why.
I will change the bias scheme a bit, but interestingly it does work as shown.
Cheers!
Russ
Thanks to everyone for the replies, I spent some more time simulating in a power amplifier configuration driving an 8ohm resistive load. Here is are the distortion figures I get:
Fourier components of V(+out,-out)
DC component:3.55247e-009
Harmonic Frequency Fourier Normalized Phase Normalized
Number [Hz] Component Component [degree] Phase [deg]
1 2.000e+04 4.538e+00 1.000e+00 -2.57° 0.00°
2 4.000e+04 3.817e-09 8.410e-10 -76.54° -73.98°
3 6.000e+04 1.294e-05 2.852e-06 37.95° 40.51°
4 8.000e+04 2.235e-09 4.925e-10 -164.64° -162.07°
5 1.000e+05 9.788e-08 2.157e-08 155.91° 158.47°
Total Harmonic Distortion: 0.000285%
That at 5Vpp output at 20khz. Notice that odd order harmonics prevail as expected, but they are still very low.
This is with the amp biased at ~150ma per FET. 300ma total.
Here is the circuit I am simulating now. The pot allows me to adjust the servo until it is perfectly zeroed.
I needed to add some compensation because of the added open loop gain of the output stages.
I will post those next
Fourier components of V(+out,-out)
DC component:3.55247e-009
Harmonic Frequency Fourier Normalized Phase Normalized
Number [Hz] Component Component [degree] Phase [deg]
1 2.000e+04 4.538e+00 1.000e+00 -2.57° 0.00°
2 4.000e+04 3.817e-09 8.410e-10 -76.54° -73.98°
3 6.000e+04 1.294e-05 2.852e-06 37.95° 40.51°
4 8.000e+04 2.235e-09 4.925e-10 -164.64° -162.07°
5 1.000e+05 9.788e-08 2.157e-08 155.91° 158.47°
Total Harmonic Distortion: 0.000285%
That at 5Vpp output at 20khz. Notice that odd order harmonics prevail as expected, but they are still very low.
This is with the amp biased at ~150ma per FET. 300ma total.
Here is the circuit I am simulating now. The pot allows me to adjust the servo until it is perfectly zeroed.
I needed to add some compensation because of the added open loop gain of the output stages.
I will post those next
Attachments
Russ,
You might want to take a look at the OPA551/552. Those are rated +-30V, can put out 200mA and are quite fast, regarding bandwidth an slewrate. Of course this would take the fun out of designing a discrete front end...
Regards,
Klaus
You might want to take a look at the OPA551/552. Those are rated +-30V, can put out 200mA and are quite fast, regarding bandwidth an slewrate. Of course this would take the fun out of designing a discrete front end...
Regards,
Klaus
KSTR said:
Oh sure, those look fine as do the National driver(LM4702 etc) but they are not fully differential or super symmetrical like this(and the THS41xx) is. 🙂
Cheers!
Russ
Ah, I see... (and guessed you would bring this up).
Well, one could make them a fully diff / SuSy design externally (NP has proposed some circuits). That might likely be flawed in performance compared to an intrinsic solution "down at the core".
Klaus
Well, one could make them a fully diff / SuSy design externally (NP has proposed some circuits). That might likely be flawed in performance compared to an intrinsic solution "down at the core".
Klaus
KSTR said:
Well, I think what you are really doing then is "wrapping" them in the feedback of a super symmetrical/fully differential amplifier. Which is precisely what my TXO projects do. 🙂
http://www.diyaudio.com/forums/showthread.php?postid=1020886#post1020886
My intent here to scale the fully differential amp up. Including as a full blown power amp. 🙂
Cheers!
Russ
Perhaps the better way of looking at it, is that this power amplifier only has one voltage gain stage.
Cheers!
Russ
Cheers!
Russ
Hi Russ,
Yes, a single voltage gain stage is a good concept. I tend to see things somewhat different than many people, as there are no true voltage gain devices. All devices steer current (either voltage or current controlled) and the voltage is developed across an active or passive load...
Two minor details regarding your driver schem:
1) You decoupled the control voltages for the basic CCS's and the folded-cascode bias to GND, not to the supplies. Doesn't that introduce a backdoor for modulated currents when the supplies ripple? I'm not sure about a siginficant effect on the cascode bias, though.
2) Q5 & Q6 don't have degeneration resistors in their emitters. This might give a stronger effect of device mismatches, resulting in unequal currents.
Ah, did you bode-plot open loop gain and phase in the sim?
Overall, the driver looks great, especially the clever DC control loop.
Regards,
Klaus
Yes, a single voltage gain stage is a good concept. I tend to see things somewhat different than many people, as there are no true voltage gain devices. All devices steer current (either voltage or current controlled) and the voltage is developed across an active or passive load...
Two minor details regarding your driver schem:
1) You decoupled the control voltages for the basic CCS's and the folded-cascode bias to GND, not to the supplies. Doesn't that introduce a backdoor for modulated currents when the supplies ripple? I'm not sure about a siginficant effect on the cascode bias, though.
2) Q5 & Q6 don't have degeneration resistors in their emitters. This might give a stronger effect of device mismatches, resulting in unequal currents.
Ah, did you bode-plot open loop gain and phase in the sim?
Overall, the driver looks great, especially the clever DC control loop.
Regards,
Klaus
Klaus,
I think (hope) I addressed all of your concerns in the schematic this post:
http://www.diyaudio.com/forums/showthread.php?postid=1258523#post1258523
Does it look any better? 🙂
Thanks very much for you feedback. I actually tend to agree with you about voltage vs current, but I was trying to use practical terms.
Cheers!
Russ
I think (hope) I addressed all of your concerns in the schematic this post:
http://www.diyaudio.com/forums/showthread.php?postid=1258523#post1258523
Does it look any better? 🙂
Thanks very much for you feedback. I actually tend to agree with you about voltage vs current, but I was trying to use practical terms.
Cheers!
Russ
- Status
- Not open for further replies.