Hi Pawel,
Here's a pure current drive version for 200W into 4 ohm with Rout of about 500 ohms and 0.02% at 1W - just what you wanted😉
There is about 6dB of current negative feedback using R14 and R45 which are 1W rating.
The total idle current is bumped up to 770mA for lower distortion at full power with 65W of idle dissipation, still half worst case sinewave dissipation. And with typical music the heatsink temperature will stay almost constant😎. But if you reduce the rail voltage to say 55V or 60V you can use a smaller heatsink and power supply.
Adding an input resistor to Vin will allow some voltage feedback; you can have mixed current and voltage feedback and reduce the output resistance over the range of 500 ohms to 5 ohms with Rin of 0-1k.
Here's a pure current drive version for 200W into 4 ohm with Rout of about 500 ohms and 0.02% at 1W - just what you wanted😉
There is about 6dB of current negative feedback using R14 and R45 which are 1W rating.
The total idle current is bumped up to 770mA for lower distortion at full power with 65W of idle dissipation, still half worst case sinewave dissipation. And with typical music the heatsink temperature will stay almost constant😎. But if you reduce the rail voltage to say 55V or 60V you can use a smaller heatsink and power supply.
Adding an input resistor to Vin will allow some voltage feedback; you can have mixed current and voltage feedback and reduce the output resistance over the range of 500 ohms to 5 ohms with Rin of 0-1k.
thank you Ian!
I spoiled it a little:
do you think it has a chance to go?
I spoiled it a little:
do you think it has a chance to go?
- TH parts,
- discrete,
- easy to build,
- cost: worse specs...
Hi Pawel,
Looks good to go. Thru hole parts for DIY👍. Remember to thermally link Q5,6 to D1,2 (likewise 2nd stage).
Note D1 and D2 are SMD so change to MUR1615 or equiv THT diodes. In sim change D's name to MUR1615_. Add two in parallel for each diode since MUR1615 is a dual pack. Then re-bias by changing R2=R22=R37=R38 (since 2 diodes in parallel). I find 0.2% THD of mainly 3rd is not noticeable - it should sound great😎.
BTW biasing for Q15,Q16 can be simplified since they are identical stages. EG remove C1,C3 and R7,9,16 and wire base Q15 to base Q11, and base Q16 to base Q12. Please repost your update. Edit the filename (remove OPA1656 from the filename), also from the circuit heading.
You should remove my copyright (so it can be copied to anywhere), or if you want add your name. As you are probably aware copyright does not protect the circuit itself, only the reproduction of the image. Any circuit on this forum becomes public domain and cannot therefore be patented. Anyone can make and sell it.
Cheers, IanH
Looks good to go. Thru hole parts for DIY👍. Remember to thermally link Q5,6 to D1,2 (likewise 2nd stage).
Note D1 and D2 are SMD so change to MUR1615 or equiv THT diodes. In sim change D's name to MUR1615_. Add two in parallel for each diode since MUR1615 is a dual pack. Then re-bias by changing R2=R22=R37=R38 (since 2 diodes in parallel). I find 0.2% THD of mainly 3rd is not noticeable - it should sound great😎.
BTW biasing for Q15,Q16 can be simplified since they are identical stages. EG remove C1,C3 and R7,9,16 and wire base Q15 to base Q11, and base Q16 to base Q12. Please repost your update. Edit the filename (remove OPA1656 from the filename), also from the circuit heading.
You should remove my copyright (so it can be copied to anywhere), or if you want add your name. As you are probably aware copyright does not protect the circuit itself, only the reproduction of the image. Any circuit on this forum becomes public domain and cannot therefore be patented. Anyone can make and sell it.
Cheers, IanH
It is amazing that this thread is still active after years. I watched this thread from the beginning, but I lost the interest as soon as I found out “Current Dumping” can hide almost all crossover distortion. The output stage of “Current Dumping” does not have to be Class C(zero biased). It could be a fine tuned class ab. You can get 0.001% thd out of it without much trouble.
Hi jxdking,
Are you aware it has been applied to Class-D also? I saw it used for an SMPS for ripple reduction in Wireless World May 1978 by Divan and Ghate - PDF attached. But I haven't seen any Class-D amps using Current Dumping. Maybe you (or others on this forum) would like to develop one?
It could be applied to an Opposed Current (OC) half-bridge (Class-D) output stage, since they can give very low open loop distortion and there's no dead-zone distortion of standard Class-D - see PDF snip attached. Then adding Current dumping allows a lower switching frequency and smaller inductors! with good efficiency at low average output power like we need with typical music😉. The best of both worlds for audio.
Maybe it's been done - I just haven't looked hard enough at the Class-D world? Meanwhile, I'm happy to continue developing my Class-AB.
And thanks for chiming in.
P.J. Walker's knighthood for inventing Current Dumping was justified IMO. It is a marvelous technique for Class-AB audio amp distortion reduction.
Are you aware it has been applied to Class-D also? I saw it used for an SMPS for ripple reduction in Wireless World May 1978 by Divan and Ghate - PDF attached. But I haven't seen any Class-D amps using Current Dumping. Maybe you (or others on this forum) would like to develop one?
It could be applied to an Opposed Current (OC) half-bridge (Class-D) output stage, since they can give very low open loop distortion and there's no dead-zone distortion of standard Class-D - see PDF snip attached. Then adding Current dumping allows a lower switching frequency and smaller inductors! with good efficiency at low average output power like we need with typical music😉. The best of both worlds for audio.
Maybe it's been done - I just haven't looked hard enough at the Class-D world? Meanwhile, I'm happy to continue developing my Class-AB.
And thanks for chiming in.
I categorize class-d as chip amp. I tend to buy them off the shelf. Modern class-d chips are very good. TPA3255 specs 0.006% thd into 1W. As long as the chip is genuine, the performance should be good. I won’t bother to diy class-d.
“Current Dumping” topology is relatively easy to follow. It can be applied to class B, or class AB as well.
I don’t like the name “Current Dumping”. The key component is the compensation “Bridge”, not the “Dump”(a zero biased output stage). I believe that’s why most people overlook this.
Hi All,
I have revisited Post 202 which was a floating power supply autobias with a CFP output stage that showed significant improvement over the standard Class-AB CFP since it is then made non-switching with the autobias loop. The pesky problem with the standard CFP in Class-AB is when one power transistor turns off and this delays its turning on at higher frequencies, and you get nasty ringing when clipping to the rails. The EF Class-AB stage can overcome this turn on delay with the Locanthi resistor between the upper and lower power transistors. And adding a capacitor to the CFP between the bases of the power transistors does not work very well, and introduces power rail noise into the output stage. But the autobias CFP in Post 202 works very well (at least so far in a sim).
So I have revisited the CFP as a voltage follower. It uses a standard power supply, not a floating power supply and no auxiliary +/-9V supply for the drivers. I then use a high voltage opamp for voltage feedback. The trade-off is higher voltage driver and CCS transistors and the high voltage opamp (HV opamps are not common and none are as fast as we would like for audio). I have simmed the LT6090-5 and OPA455.
Unfortunately opamp distortion is not modelled so I have no idea what distortion we will (or can) actually get and how stable THD is with temperature changes. Whatever it turns out to be I'd expect the distortion to be so low as to be a non-problem.
The output stage alone gives 0.003% THD at 1W. The file is attached, but needs editing for LT-XVII or later -- change level.3a to level_3a subcircuit call then add .include UniversalOpamps3.sub.
Two refinements over Post 202 are 1) the capacitors across the spreaders (to damp ringing when leaving clipping) and 2) the 10 Ohms plus 47nF across R13 to keep the power transistors on at the top end like 100kHz. So no R+C used (or needed) between the bases of the power transistors. Autobias operation is maintained up to 100kHz which is my usual top end design rolloff frequency.
The power transistor base-emitter turn off resistors are 33 ohms (were 22 ohms) which are sufficient for 100kHz which reduces the driver dissipation to a minimum.
Thermal stability is determined by thermal linkage from the power diodes to the spreaders as well as the CCS's stability. And the power transistors and drivers do not need any thermal compensation linkage to the spreaders.
Current limiting (for SOA) is crude - the CCS's are set depending on the betas of the driver plus power transistor combination. For DIY we can measure betas and choose suitable resistors for the CCS's. I avoided adding current limit transistors here because of ringing at clip recovery and other issues, so keeping it simple gives a more stable and robust autobias amp -- which is the main aim of this thread.
I have revisited Post 202 which was a floating power supply autobias with a CFP output stage that showed significant improvement over the standard Class-AB CFP since it is then made non-switching with the autobias loop. The pesky problem with the standard CFP in Class-AB is when one power transistor turns off and this delays its turning on at higher frequencies, and you get nasty ringing when clipping to the rails. The EF Class-AB stage can overcome this turn on delay with the Locanthi resistor between the upper and lower power transistors. And adding a capacitor to the CFP between the bases of the power transistors does not work very well, and introduces power rail noise into the output stage. But the autobias CFP in Post 202 works very well (at least so far in a sim).
So I have revisited the CFP as a voltage follower. It uses a standard power supply, not a floating power supply and no auxiliary +/-9V supply for the drivers. I then use a high voltage opamp for voltage feedback. The trade-off is higher voltage driver and CCS transistors and the high voltage opamp (HV opamps are not common and none are as fast as we would like for audio). I have simmed the LT6090-5 and OPA455.
Unfortunately opamp distortion is not modelled so I have no idea what distortion we will (or can) actually get and how stable THD is with temperature changes. Whatever it turns out to be I'd expect the distortion to be so low as to be a non-problem.
The output stage alone gives 0.003% THD at 1W. The file is attached, but needs editing for LT-XVII or later -- change level.3a to level_3a subcircuit call then add .include UniversalOpamps3.sub.
Two refinements over Post 202 are 1) the capacitors across the spreaders (to damp ringing when leaving clipping) and 2) the 10 Ohms plus 47nF across R13 to keep the power transistors on at the top end like 100kHz. So no R+C used (or needed) between the bases of the power transistors. Autobias operation is maintained up to 100kHz which is my usual top end design rolloff frequency.
The power transistor base-emitter turn off resistors are 33 ohms (were 22 ohms) which are sufficient for 100kHz which reduces the driver dissipation to a minimum.
Thermal stability is determined by thermal linkage from the power diodes to the spreaders as well as the CCS's stability. And the power transistors and drivers do not need any thermal compensation linkage to the spreaders.
Current limiting (for SOA) is crude - the CCS's are set depending on the betas of the driver plus power transistor combination. For DIY we can measure betas and choose suitable resistors for the CCS's. I avoided adding current limit transistors here because of ringing at clip recovery and other issues, so keeping it simple gives a more stable and robust autobias amp -- which is the main aim of this thread.
Last edited:
Recently, I found Nelson Pass had a paper/presentation regarding Square Law devices. https://www.firstwatt.com/wp-content/uploads/2023/12/art_the_square_law.pdf
It explains the mechanism how push pull topology cancels out the distortion and the effects of the degeneration resistors in the push pull.
One of the examples is his F7 amplifier. https://www.firstwatt.com/wp-content/uploads/2023/08/prod_f7_man.pdf
It gets 0.05% distortion. (note: it is Class A amp.) You could wrap a GNFB with an OPAMP to lower it more.
I did simple simulation with a simple EF push pull with BJT, without the emitter resistors. 10mA bias, 10Vp into 100 Ohm (I call it Class AB). It gets 0.1% THD without any NFB.
Thus, the emitter resistors play a big role on the distortion. The voltage drop on the resistors make the output transistors switch off deeper than that without the resistors.
The practical way to avoid that is to use Lateral MOSFET as the output devices that its internal resistance raises with the temperature, so that you don't need source/emitter resistors any more to sense the current.
It explains the mechanism how push pull topology cancels out the distortion and the effects of the degeneration resistors in the push pull.
One of the examples is his F7 amplifier. https://www.firstwatt.com/wp-content/uploads/2023/08/prod_f7_man.pdf
It gets 0.05% distortion. (note: it is Class A amp.) You could wrap a GNFB with an OPAMP to lower it more.
I did simple simulation with a simple EF push pull with BJT, without the emitter resistors. 10mA bias, 10Vp into 100 Ohm (I call it Class AB). It gets 0.1% THD without any NFB.
Thus, the emitter resistors play a big role on the distortion. The voltage drop on the resistors make the output transistors switch off deeper than that without the resistors.
The practical way to avoid that is to use Lateral MOSFET as the output devices that its internal resistance raises with the temperature, so that you don't need source/emitter resistors any more to sense the current.
Last edited:
Sense the collector current. Icollector = alpha * Iemitter . . . . . . . and alpha ~=~ 0.99
If you're feeling frisky, do this with optoisolators, in a way similar to Nelson Pass's optical bias patent 4,752,745 . Or else high side current monitoring ICs, or "over the top" opamps, etc.
If you're feeling frisky, do this with optoisolators, in a way similar to Nelson Pass's optical bias patent 4,752,745 . Or else high side current monitoring ICs, or "over the top" opamps, etc.
Yes, but it is tricky to get everything right.
One example is the Musical Fidelity A1. https://www.diyaudio.com/community/...-to-my-youth-babelfishing.419086/post-7824641
With MF-A1, everything is perfect within Class A region. However, once clipping into class B, everything goes to hell.
Thanks for the likes and comments.
Converting the voltage follower opamp version Post 287 to current drive can be done the usual way -- add a resistor in the cold side of the load for current sense feedback to the opamps inverting input node (see below):
Distortion at 1W into 8 ohms and 1kHz is shown above.
Converting the voltage follower opamp version Post 287 to current drive can be done the usual way -- add a resistor in the cold side of the load for current sense feedback to the opamps inverting input node (see below):
Distortion at 1W into 8 ohms and 1kHz is shown above.
And here's an unusual way to do a bridge -- using cross-coupling of the opamps inverting inputs (explained below):
LHS opamp U2 inverting input resistor R18 was connected to common but is now connected to RHS opamp U1 inverting input and its noninverting input is grounded The LHS closed loop voltage gain is about 11 (being 1+R19/R18 since U1 inverting input is a virtual ground). The RHS closed loop voltage gain is made the same by adding R28 to R38 (gain of (R28+R38)/R18 )
BTW R37 does not enter this equation but is retained since without it SPICE finds it difficult to reach convergence. The circuit above is only suitable for 16 ohms with +/-42V rails for SOA reasons, giving 100W into 16 ohms.
The attached file has a parallel output stage and bridged for 8 ohms giving 200W. An FFT is shown below at 4W with 8 ohms.
LHS opamp U2 inverting input resistor R18 was connected to common but is now connected to RHS opamp U1 inverting input and its noninverting input is grounded The LHS closed loop voltage gain is about 11 (being 1+R19/R18 since U1 inverting input is a virtual ground). The RHS closed loop voltage gain is made the same by adding R28 to R38 (gain of (R28+R38)/R18 )
BTW R37 does not enter this equation but is retained since without it SPICE finds it difficult to reach convergence. The circuit above is only suitable for 16 ohms with +/-42V rails for SOA reasons, giving 100W into 16 ohms.
The attached file has a parallel output stage and bridged for 8 ohms giving 200W. An FFT is shown below at 4W with 8 ohms.
Last edited:
The 2N5550/5401 are dissipating 840mW, I wonder if they could survive with a heatsink despite operating over spec.
Hi keantoken,
Oops. Well spotted. I just looked them up. They are only about 400mW devices maximum with Ta of say 70C. The FZT751/851 is SOT-223 looks better for 0.8W on 1in sq pcb. For now I have parallelled two 2N5550/5401 for 400mW each. I'm parallelling because I prefer high FT drivers with CFP's. If they still run too hot then I'll look at 47 ohm base-emitter resistors. And the beta's of the power transistors will need to be selected for high ones otherwise use 3 drivers in parallel.
Oops. Well spotted. I just looked them up. They are only about 400mW devices maximum with Ta of say 70C. The FZT751/851 is SOT-223 looks better for 0.8W on 1in sq pcb. For now I have parallelled two 2N5550/5401 for 400mW each. I'm parallelling because I prefer high FT drivers with CFP's. If they still run too hot then I'll look at 47 ohm base-emitter resistors. And the beta's of the power transistors will need to be selected for high ones otherwise use 3 drivers in parallel.
I have got a current drive bridge running 😵 using a grounded output bridge so the usual current sensing on the cold side can be used (like Post 291).
Notice it uses a floating power supply, but that's a fair enough trade for a grounded bridge IMO.
The above circuit is for voltage drive and for 16 ohms. Linking the "cfb" node to the opamp U2 inverting input will give current drive.
The attached file has the sims for 8 Ohms and 400W with and without current drive. The trick to get it stable without a dual supply (no power common) is what I'd been using previously in this thread with a floating power supply AND to make the grounded half of the bridge slower than the LHS follower half - notice C9 across U1 is 27pF to slow the RHS relative to the LHS.
Notice it uses a floating power supply, but that's a fair enough trade for a grounded bridge IMO.
The above circuit is for voltage drive and for 16 ohms. Linking the "cfb" node to the opamp U2 inverting input will give current drive.
The attached file has the sims for 8 Ohms and 400W with and without current drive. The trick to get it stable without a dual supply (no power common) is what I'd been using previously in this thread with a floating power supply AND to make the grounded half of the bridge slower than the LHS follower half - notice C9 across U1 is 27pF to slow the RHS relative to the LHS.
On TO-92 that run hot you can always add a small heatsink.
https://www.digikey.com/en/products...=N4IgTCBcDaIC4HsCcYAEALApgQzgZwEsA7AaxAF0BfIA
https://www.digikey.com/en/products...=N4IgTCBcDaIC4HsCcYAEALApgQzgZwEsA7AaxAF0BfIA
Hi Keantoken,
How much does it increase Pd with say 70C ambient (my likely Ta on a board in an amp case)? The specs say the TO-92 clip-on has 60C/W.
The bare TO92 200C/W.
Without the HS the derating curve shows 400mW with Ta of 70C. So with the clip-on would that be doubled to 800mW?
If it does then I'd still parallel them with HS's to be safe.
How much does it increase Pd with say 70C ambient (my likely Ta on a board in an amp case)? The specs say the TO-92 clip-on has 60C/W.
The bare TO92 200C/W.
Without the HS the derating curve shows 400mW with Ta of 70C. So with the clip-on would that be doubled to 800mW?
If it does then I'd still parallel them with HS's to be safe.
TO92 junction to ambient is 200C/W, but junction to case is 83C/W. Adding the heatsink to that, 83+60 = 143C/W. At 25C ambient, 125C rise allows 125/143=874mW max dissipation, 1.4x stock. I found some larger TO92 heatsinks once but don't remember where I got them.
Use thermal cement to glue a pair of US 25-cent pieces (nickel plated copper) to the TO-92, front and back. I haven't measured thetaJA in this configuration but I suspect it's low, and I'm certain it's inexpensive.
The LT1223 is a 100MHz CFA is used to drive an autobias stage like Posts 287 (rather than the HV opamp LT0690 or OPA454). The LT1223 is available in DIP-8 for DIY. My circuit is shown below:
When the feedback is taken from "Op1" instead of "Out" there is no global feedback so the amp is less sensitive to reactive loads (like here) but THD is higher (probably unnoticed with audio):
. . . . 1kHz 1W . . 20kHz . . . BW
GFB . 0.3ppm . . 0.6ppm . . 4.6MHz
NGFB . 2 ppm . . 6 ppm . . 2.3MHz
When the feedback is taken from "Op1" instead of "Out" there is no global feedback so the amp is less sensitive to reactive loads (like here) but THD is higher (probably unnoticed with audio):
. . . . 1kHz 1W . . 20kHz . . . BW
GFB . 0.3ppm . . 0.6ppm . . 4.6MHz
NGFB . 2 ppm . . 6 ppm . . 2.3MHz