Let me present some design ideas for 1ppm power amps using parallel opamps.
Each opamp is fitted with a Vlemincq output stage made of three inexpensive BJTs. There are two parallel branches for implementing a complementary symmetry. Have a look to the two Vlemincq output stages.
The Kuroda topology got chosen for getting rid of the power supply stabilization one might expect using integrated opamps or integrated buffers.
This way, any unregulated PSU voltage can be used, anywhere from 22V to 42V per supply rail depending on the voltage rating of the transistors used in the Vleming output stage. The simulation got made using ideal +36V and -36V rails.
The regulated +18V and -18V local supply feeding the opamps can be shared by all cells.
For avoiding the Kuroda return current flowing in the ground wire, a virtual ground gets locally generated. Each virtual ground opamp must be capable of handling 15mA.
Single opamps need to be used in the voltage amplification stages, as their instantaneous power supply current get used as signal in the Kuroda topology.
Dual or quad opamps may be used in the virtual ground opamps. Their supply can be the shared +18V and -18V supplies.
On simulation, each cell delivers 1.50 ppm THD (0.000150 %) when outputting a 24V peak-peak-sinus at 1kHz into a 64 ohm load. You need to parallel at least 8 cells for driving a 8 load with the same quality.
There are no expensive parts in this design. All parts are widely available, multisource and multivendor. There are big savings made in the power supply department as the power supply doesn't need to be regulated.
Output transistors should be as fast as possible.
There are medium power BJTs in SMD nowadays with excellent specifications. A radiator should be clipsed on top of them. Don't allow the PCB to heat up, as this may impact the opamp specs.
For the ultimate compacity and speed, SOT-89 models are extremely fast (300 MHz Ft) but they are limited to 50V and 2 amp like : 2SC5569-2SA2016, 2SC5566-2SA2013, 2SC5964-2SA2125, 2SC5994-2SA2153. Their dissipation can't exceed one or two watt. Don't expect power with these !
Very compact also, and fast, are some SOT-223 models (130 MHz Ft) but they are limited to 100V and 4 amp like : FZT853 - FZT953. Their dissipation can't exceed 3 watt. Don't expect massive power with these !
If ultimate compacity is not targeted, the wise choice are fast DPAK TO-251 models able to dissipate 20 watt, cope with 100V and handle 3 amp or 4 amp.
Ft > 120 MHz NPN : 2SD1815, 2SD1816
Ft > 120 MHz PNP : 2SB1215, 2SA1593
If the above fast DPAK models are not available, a decent second choice are the 80V and 8 amp devices :
Ft 50 MHz NPN : MJ_MJD44H11
Ft 50 MHz PNP : MJ_MJD45H11
In through-hole, a wise choice would be fast TO-126 models able to dissipate 20 watt, cope with 100V and handle 3 amp or 4 amp.
Ft > 120 MHz NPN : 2SD1725
Ft > 120 MHz PNP : 2SB1167
The Vlemincq output stage looks very attractive with output transistors like the above ones, having a Ft > 120 MHz.
See attached .zip
Each opamp is fitted with a Vlemincq output stage made of three inexpensive BJTs. There are two parallel branches for implementing a complementary symmetry. Have a look to the two Vlemincq output stages.
The Kuroda topology got chosen for getting rid of the power supply stabilization one might expect using integrated opamps or integrated buffers.
This way, any unregulated PSU voltage can be used, anywhere from 22V to 42V per supply rail depending on the voltage rating of the transistors used in the Vleming output stage. The simulation got made using ideal +36V and -36V rails.
The regulated +18V and -18V local supply feeding the opamps can be shared by all cells.
For avoiding the Kuroda return current flowing in the ground wire, a virtual ground gets locally generated. Each virtual ground opamp must be capable of handling 15mA.
Single opamps need to be used in the voltage amplification stages, as their instantaneous power supply current get used as signal in the Kuroda topology.
Dual or quad opamps may be used in the virtual ground opamps. Their supply can be the shared +18V and -18V supplies.
On simulation, each cell delivers 1.50 ppm THD (0.000150 %) when outputting a 24V peak-peak-sinus at 1kHz into a 64 ohm load. You need to parallel at least 8 cells for driving a 8 load with the same quality.
There are no expensive parts in this design. All parts are widely available, multisource and multivendor. There are big savings made in the power supply department as the power supply doesn't need to be regulated.
Output transistors should be as fast as possible.
There are medium power BJTs in SMD nowadays with excellent specifications. A radiator should be clipsed on top of them. Don't allow the PCB to heat up, as this may impact the opamp specs.
For the ultimate compacity and speed, SOT-89 models are extremely fast (300 MHz Ft) but they are limited to 50V and 2 amp like : 2SC5569-2SA2016, 2SC5566-2SA2013, 2SC5964-2SA2125, 2SC5994-2SA2153. Their dissipation can't exceed one or two watt. Don't expect power with these !
Very compact also, and fast, are some SOT-223 models (130 MHz Ft) but they are limited to 100V and 4 amp like : FZT853 - FZT953. Their dissipation can't exceed 3 watt. Don't expect massive power with these !
If ultimate compacity is not targeted, the wise choice are fast DPAK TO-251 models able to dissipate 20 watt, cope with 100V and handle 3 amp or 4 amp.
Ft > 120 MHz NPN : 2SD1815, 2SD1816
Ft > 120 MHz PNP : 2SB1215, 2SA1593
If the above fast DPAK models are not available, a decent second choice are the 80V and 8 amp devices :
Ft 50 MHz NPN : MJ_MJD44H11
Ft 50 MHz PNP : MJ_MJD45H11
In through-hole, a wise choice would be fast TO-126 models able to dissipate 20 watt, cope with 100V and handle 3 amp or 4 amp.
Ft > 120 MHz NPN : 2SD1725
Ft > 120 MHz PNP : 2SB1167
The Vlemincq output stage looks very attractive with output transistors like the above ones, having a Ft > 120 MHz.
See attached .zip
Attachments
Last edited:
output transistors from STMicroelectronics in DPAK TO-252
Ft=150 MHz 60V 5A 15Watt
NPN = STD1805
PNP = STD2805
Ft=130 MHz 60V 3A 15Watt
NPN = 2STD1360
PNP = 2STD2360
driver transistors from ROhm in small SMT packages (smaller than DPAK TO-252)
Ft=100 MHz 80V 1A 1Watt
NPN = 2SD1733 and 2SD1898
PNP = 2SB1181 and 2SB1260
Ft=150 MHz 60V 5A 15Watt
NPN = STD1805
PNP = STD2805
Ft=130 MHz 60V 3A 15Watt
NPN = 2STD1360
PNP = 2STD2360
driver transistors from ROhm in small SMT packages (smaller than DPAK TO-252)
Ft=100 MHz 80V 1A 1Watt
NPN = 2SD1733 and 2SD1898
PNP = 2SB1181 and 2SB1260
hi jcx, the floating supply bridge looks attractive. I got it simulated. Works like a charm. Many thanks. From where is the first opamp getting his negative feedback ? What's the magic ?
Attachments
Last edited:
hi jcx, could it be the right side of the schematic of the floating supply bridge can't be implemented using parallel opamps ? With this schematic, there is a major current sharing issue in the right side of the schematic.
Attachments
Hello all
I would like to thank you for the interest you have shown in the opamp-poweramp design. I'm afraid I'm showing up rather late for this party, but pressure of business, you know...
I should say at once that while this is a wholly serious project, that does exactly what it says in terms of performance, it also has its light-hearted aspect. I don't think it is going to revolutionise amplifier design (though at least one kind person has suggested it, in another forum!) but I do believe it at least provides a new way to look at the subject.
I can't attempt to answer all the points raised in this discussion, but here are a few thoughts that occur:
As suggested, it would probably be possible to get lower distortion by operating the output stage opamp array with shunt feedback. The problems are:
1) It adds two resistors per opamp section (ie four per package) and that is a lot of extra components. The voltage-follower could hardly be simpler.
2) The input impedance of the output stage opamp array will now be MUCH lower, if reasonably low feedback resistor values are used, and it will be harder to drive it with very low distortion.
3) The voltage-followers have very accurate unity gain. Using shunt feedback adds in two resistor tolerances and might cause trouble with current-sharing.
An idea that was put forward in the Elektor forum was the possibility of closing a feedback loop around the output stage voltage-follower array. The distortion of the voltage-followers is very low, and I did not consider it necessary- in a first design at least- to try to close another feedback loop around them. That eliminated possible stability problems and made the design very straightforward. In fact, the major challenge was designing the gain stage so it had distortion low enough to be comparable with that of the voltage-followers. That job could probably be done with a single LM4562, which has extremely low distortion, but that does rather lose the "all-5532" philosophy.
IMPORTANT! Please be aware that my website is now at The Douglas Self Site (douglas-self.com)
Please pass this info on to anyone you think might be interested. The address should not change again.
I would like to thank you for the interest you have shown in the opamp-poweramp design. I'm afraid I'm showing up rather late for this party, but pressure of business, you know...
I should say at once that while this is a wholly serious project, that does exactly what it says in terms of performance, it also has its light-hearted aspect. I don't think it is going to revolutionise amplifier design (though at least one kind person has suggested it, in another forum!) but I do believe it at least provides a new way to look at the subject.
I can't attempt to answer all the points raised in this discussion, but here are a few thoughts that occur:
As suggested, it would probably be possible to get lower distortion by operating the output stage opamp array with shunt feedback. The problems are:
1) It adds two resistors per opamp section (ie four per package) and that is a lot of extra components. The voltage-follower could hardly be simpler.
2) The input impedance of the output stage opamp array will now be MUCH lower, if reasonably low feedback resistor values are used, and it will be harder to drive it with very low distortion.
3) The voltage-followers have very accurate unity gain. Using shunt feedback adds in two resistor tolerances and might cause trouble with current-sharing.
An idea that was put forward in the Elektor forum was the possibility of closing a feedback loop around the output stage voltage-follower array. The distortion of the voltage-followers is very low, and I did not consider it necessary- in a first design at least- to try to close another feedback loop around them. That eliminated possible stability problems and made the design very straightforward. In fact, the major challenge was designing the gain stage so it had distortion low enough to be comparable with that of the voltage-followers. That job could probably be done with a single LM4562, which has extremely low distortion, but that does rather lose the "all-5532" philosophy.
IMPORTANT! Please be aware that my website is now at The Douglas Self Site (douglas-self.com)
Please pass this info on to anyone you think might be interested. The address should not change again.
Last edited:
your R1,2 senses the midpoint of the floating supply which the 1st op amp moves by "tugging" on the rails through its ps pins when its output sources current into "gnd"
this inverts the sign of the gain so R1,2 midpoint provides negative loop feedback - so it should go only to 1 op amp - you can't "share" feedback in parallel op amps
to parallel output current just use the 1st op amp as the "master" and use ordinary unity gain connection parallel op amps mirroring the Vdrop of the feedback controlled output current sharing/sense resistor "A47" style
I consider 100 Ohms too large for the current sharing/sensing Rs - Cload gives phase shift which destablizes the loop, 10 Ohms gives good enough current sharing with mV offset op amps
Last edited:
Thanks for the explanation about the 1st opamp feeback path. As advised, I used the 1st opamp as the "master" then buffered it using unity gain parallel opamps. However, I don't see how to mirror the Vdrop of the feedback controlled output current sharing/sense resistor. What is the "A47" you are referring to ?
p.s. 1 : At the moment I'm using 100 ohm current sharing resistors in the simulation, but of course with better opamps and more opamps in parallel the value is going to be decreased to maybe 1 ohm.
p.s. 2 : What is the role of R5 (mid point R6/R4) in your post #17 ?
Attachments
Last edited:
"A47" is pretty specific to headphone land but the circuit is common in various op amp app notes: Comparing The A47 and This Headphone Amp Spec'd in OPA2227 Datasheet - Head-Fi.org Community
seen in opa2604 datasheet fig 9
will be more stable with +V gain, even a gain of 2 helps
the sim below runs with Gear integration - which is known to give silightly different numerical stability in the sim vs the default Trapezoidal
maybe your transistor level op amp models will be happier
U3, R10 copies R9 current, likewise U4, R11 copies R12 current; U3,4 "slave" op amps need rail-to-rail inputs to work well in this circuit
seen in opa2604 datasheet fig 9
will be more stable with +V gain, even a gain of 2 helps
the sim below runs with Gear integration - which is known to give silightly different numerical stability in the sim vs the default Trapezoidal
maybe your transistor level op amp models will be happier
U3, R10 copies R9 current, likewise U4, R11 copies R12 current; U3,4 "slave" op amps need rail-to-rail inputs to work well in this circuit
Attachments
Last edited:
hi jcx,
thanks a lot, I'm getting the "A47" style (bootstrapped buffered) floating supply bridge working nice on simulation using the transistor level TL071 model.
Steph
thanks a lot, I'm getting the "A47" style (bootstrapped buffered) floating supply bridge working nice on simulation using the transistor level TL071 model.
Steph
Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.