Dear fellow enthusiasts,
So I designed, breadboarded and listened to this peculiar amplifier. It is basic in using class AB bias, about 100W into 4 Ohm and achieving linearity through plenty of NFB. However, opposed to most amps the transconductance stage is formed by cascoded complementary JFET's where global feedback is applied to the source pins. The following transimpedance stage is a complimentary Darlington common emitter amplifier with two pole compensation. Finally, is output stage CFP VMOS with folded BJT pre-drivers and BJT drivers. The overall goal was to achieve high open-loop bandwidth. Resulting performance is more than satisfactory.
In the real world however, it needed some help. So I introduced a DC-servo, passive power rail regulation and took some measures to ensure clean clipping. I determined output bias to be thermally sound. Nevertheless, it could hover about a little less.
All was well until...
I fed it a 100kHz square wave to full output power and the MOSFET's did not hesitate to destroy themselves in spectacular fashion. I was feeding the amp from a raw supply trough 10A fast fuses. Also I had some Zener's installed, not enough. Even without load, the power FET's will try to pass enormous currents when fed with high f square waves. Since to me the distortion introduced by normal BJT current limiters are unacceptable (they act too soon and other reasons) I started thinking about the ideal protection circuitry.
I want it to detect overcurrent, overheating and DC offset while being sonically transparent. When a fault is detected a solid-state relay to the speaker should be disconnected and the drive current to the MOSFET's should be removed. After say 100ms it should reset and check for a fault again. In LTspice I conceived a circuit that meets most of these requirements however, the auto-reset function is where I run into trouble. Additionally it is not fast enough to prevent the MOSFETS's from seeing a 5us burst of 45A when trying to drive 0.1 Ohm. If the image below is hard to read please find the pdf attached below. It is taken from LTspice and does not include catching diodes and gate Zeners.
I am simply hoping to pick your masterful brains on how to make the protection circuit act faster and on how to implement an auto-reset function. Any other response is welcome too. If you have questions about any aspect of this amp I'd be happy to elaborate further.
Big Cheers,
Ruben
So I designed, breadboarded and listened to this peculiar amplifier. It is basic in using class AB bias, about 100W into 4 Ohm and achieving linearity through plenty of NFB. However, opposed to most amps the transconductance stage is formed by cascoded complementary JFET's where global feedback is applied to the source pins. The following transimpedance stage is a complimentary Darlington common emitter amplifier with two pole compensation. Finally, is output stage CFP VMOS with folded BJT pre-drivers and BJT drivers. The overall goal was to achieve high open-loop bandwidth. Resulting performance is more than satisfactory.
In the real world however, it needed some help. So I introduced a DC-servo, passive power rail regulation and took some measures to ensure clean clipping. I determined output bias to be thermally sound. Nevertheless, it could hover about a little less.
All was well until...
I fed it a 100kHz square wave to full output power and the MOSFET's did not hesitate to destroy themselves in spectacular fashion. I was feeding the amp from a raw supply trough 10A fast fuses. Also I had some Zener's installed, not enough. Even without load, the power FET's will try to pass enormous currents when fed with high f square waves. Since to me the distortion introduced by normal BJT current limiters are unacceptable (they act too soon and other reasons) I started thinking about the ideal protection circuitry.
I want it to detect overcurrent, overheating and DC offset while being sonically transparent. When a fault is detected a solid-state relay to the speaker should be disconnected and the drive current to the MOSFET's should be removed. After say 100ms it should reset and check for a fault again. In LTspice I conceived a circuit that meets most of these requirements however, the auto-reset function is where I run into trouble. Additionally it is not fast enough to prevent the MOSFETS's from seeing a 5us burst of 45A when trying to drive 0.1 Ohm. If the image below is hard to read please find the pdf attached below. It is taken from LTspice and does not include catching diodes and gate Zeners.
I am simply hoping to pick your masterful brains on how to make the protection circuit act faster and on how to implement an auto-reset function. Any other response is welcome too. If you have questions about any aspect of this amp I'd be happy to elaborate further.
Big Cheers,
Ruben
Last edited:
Not a materful brain here and I can't comment on your circuit right now, but have you identified the root cause for the spectacular destruction of the OPS?
It it maybe cross-conduction of the power MOSFETs?
Or related to the amplification somehow?
It it maybe cross-conduction of the power MOSFETs?
Or related to the amplification somehow?
Hi , i see one possible problem, indirectly related. Outputs mosfets you use are connected not as source followers(fast operation) ,but as common source ( easier to drive, but slow). Also 0.05 ohm source resistors are too small , in such case you need 0,1-0,5 ohm for mosfets current balance and as local feedback , which improves linearity. Also gate resistor use huge 470 ohms , ok for 20khz ,but not for 2x mosfets with few nanofarads input capacitance. Try to use totem pole driver with small idle current for each polarity mosfets , will decrease shoot through. Few years ago have played alot with microcap , tested output mosfet operation in several ways , common source is harder to get work fast and has high distortions. The same apply to bjt variants. However , use of source follower requires bootstrapped power supply , but it can be easily accomplished with resistor, diode ,zener ,and capacitor. But its worth to , reduces mosfet heating and voltage drop ,when close to saturation.
The part of the protection circuit (DC offset, if I'm not mistaken) comprised of Q18, Q19, Q22, Q23, Q24, Q25, will operate faster if Base-Emitter resistors are fitted -- especially in the turn-off.
I haven't looked up your opto-isolators, but they'll need a ton of current gain to function where they are in the circuit. Also, the bigger they are, generally, the slower they are, again especially turning off.
Cheers
I haven't looked up your opto-isolators, but they'll need a ton of current gain to function where they are in the circuit. Also, the bigger they are, generally, the slower they are, again especially turning off.
Cheers
Ooops, sorry, I was wrong 😳 on the ". . ton of current gain . ." bit -- not sure what I was looking at . .
But nevertheless, the 'input/driver-rail' turn-offs would be faster if the Emitter or Collector opto-lead didn't have to slew half the full supply rails. Maybe find a way to turn it off at the rail, instead of from Ground.
Best Regards
But nevertheless, the 'input/driver-rail' turn-offs would be faster if the Emitter or Collector opto-lead didn't have to slew half the full supply rails. Maybe find a way to turn it off at the rail, instead of from Ground.
Best Regards
Maybe have the opto-couplers shunt Vbe of Q16 and Q15. That'll drop the rails quicker than trying to discharge that 0,1uF from 40V to Ground. Also, then those pass transistors won't have to suffer the short-duration, reverse Vbe as the 2uF sub-rail capacitors try to hold up the Emitters.
Dear Rick, you understand what im trying to achieve with the protection circuit.
Indeed im shunting the front end supplies to ground by engaging the opto couplers. Perhaps 'disconnecting ' the front end supply by shutting down more solidstate relays could act faster? More mosfet switches.
Much thanks for thinking with me. The amp is generously linear. I just hope to make it fool proof without compromising performance too much. Instead im allowing more complexity and will design an auxilary protection board.
Cheers! 🍻
Ruben
Indeed im shunting the front end supplies to ground by engaging the opto couplers. Perhaps 'disconnecting ' the front end supply by shutting down more solidstate relays could act faster? More mosfet switches.
Much thanks for thinking with me. The amp is generously linear. I just hope to make it fool proof without compromising performance too much. Instead im allowing more complexity and will design an auxilary protection board.
Cheers! 🍻
Ruben
OK, well maybe at least put a diode, reverse-biased across Q15 and Q16's Base-Emitter, so it doesn't suffer every time the opto-couplers turn on.
Best Regards 🍻
Best Regards 🍻
Haha you're right. Those diodes cant hurt. Shunting Q15 and Q16 Vbe though yields the same result. The front end supplies still need to slew down before the drivers are devoid of current. What I tried before is shutting down the current sources feeding the pre-drivers. This however will cause the front end to go berserk, hence my choice to remove all front end current. Maybe I should combine?
Hi Ximikas,
Could you elaborate what you mean with bootstrapped supplies? Could you show a little schematic?
Ive done alot of breadboard testing and LTspice simulation. Source follower output stage is where I started. It was okay but not spectacular. Common source mode like my design alows for hefty local feedback, far superior linearity actually. Gate resistors are will be 330 ohms. 470 ohms are the 'gate to source ' resistors
Much tanks and cheers,
Ruben
Bootstrapped power supply is when you use Nmos for positive rail and Pmos for negative rail ,and they needs to be about 3-8 volts to fully open ,to have minimal voltage drop .Also to mention , played many hours in microcap with common source design as yours ,but had not satisfied with speed and distortion,also voltage drop is high ,several volts .So more heat .
Bootstrapped supply is not my idea ,i just adjusted it a little : lets say you have +-45V supply ,so Nmos mosfet gate must be able reach +55V when needed ,or simply up to 10 volts more positive than output .The same with Pmos ,for negative output swing.This easily acomplished with resistor and zener typically, but i made it regulated , i don't like low resistance resistors dissipating a lot of heat ,simplicity is good only until some point .
I will try attach screenshot of my design ,if i can call it so .I have this amplifier playing for few years and now,and would say i'm satisfied with simulation results and actual audio quality ,its powerfull .I would like to improve it too , by adding current limiting ,but at moment have no working ideas how to implement it properly .
Bootstrapped supply is not my idea ,i just adjusted it a little : lets say you have +-45V supply ,so Nmos mosfet gate must be able reach +55V when needed ,or simply up to 10 volts more positive than output .The same with Pmos ,for negative output swing.This easily acomplished with resistor and zener typically, but i made it regulated , i don't like low resistance resistors dissipating a lot of heat ,simplicity is good only until some point .
I will try attach screenshot of my design ,if i can call it so .I have this amplifier playing for few years and now,and would say i'm satisfied with simulation results and actual audio quality ,its powerfull .I would like to improve it too , by adding current limiting ,but at moment have no working ideas how to implement it properly .
Dear Lee,
Well im not sure why the mosfets try to cunduct distructive current when handling big squarewaves. When using lateral Mosfets this does not occur. The vertical fets seem to dangerously conduct at the same time.
Cheers!
Ruben
Some explanation : D17 in series with R36 and C19 forms a supply for positive bootstrap .D20 amd R39 adds additional current to it when output is negative ,but i'm unsure if it works as expected . C19 capacity is selected according to be able supply driver at lowest frequencies like few hertz .Why 3 mosfets ? Because in first tests i had single mosfet and no ferrite beads on G pins and had some oscillation at many MHZ ,which made overheat and mosfets were killed ,but i thinked they were fake and not 150w ,but 50w rated ,so i didn't wanted to risk . Q5 Q6 is driver with 1ma idle current ,more than enough to drive easily 3 mosfets . Why so low gate resistor ? Because with resistance increase i lose mosfet benefit - speed , this causes output delay and therefore harder to stabilyse amp , its more prone to oscillation with higher resistor ,also distortion increase at 100khz .This is overkill offcourse ,but ...
L2 L3 L4 are just small ferrite bead with 10ohms resistor going through bead hole .But it stops oscillation . Without ferrites these gate resistors heating up when playing louder ,that was very strange .
Q3 is positive VAS , which is supplied ,like driver ,from bootstrapped supply ,when main supply is not enough ,to increase possible swing .
L2 L3 L4 are just small ferrite bead with 10ohms resistor going through bead hole .But it stops oscillation . Without ferrites these gate resistors heating up when playing louder ,that was very strange .
Q3 is positive VAS , which is supplied ,like driver ,from bootstrapped supply ,when main supply is not enough ,to increase possible swing .
Thanks for your elaboration. Its teaching me a couple new things. How does your creation perform? did you measure it?
For my current design im sticking to common source. I found that with the right implementation of beefier zeners the output current can be hard limited.
Now the mosfets are never seeing more then 9Amps even without shunting down the rails. The zeners add distortion but only 0.00015% at 20Khz 100W.
When the zeners conduct however, the drivers content with just over 200mA and the VAS bjts 400mA of emitter to base current (!). Fixing these problems would make it perfect. But how to tackle those?
Cheers!
Your Q11 with mosfets form something similar to darlington ,so you can limit output current by limiting voltage swing at Q11 base in respect to ground or output . You can try insert transistor in signal path ,which is always saturated ,in example another emitter follower ,base feeded to +VCC by 100k in example , and when you need limit current ,shunt to output .But then you need add common current sense resistor at point where both D comes together .At moment this comes as fast idea . Did you tried that ? Limiting at gate is not always proper way ,you take load to driver ,which must be current limited then ,CCS in example.
Also i don't think that with 470 ohms mosfets are able to withstand sqare wave input .You mentioned 100khz , if is correct ,remember then any DC DC converter ... How mosfets are controlled there ? Resistances are just few ohms ,and discharging is done with pnp ot npn at least.
Try to simulate your output stage with sqare wave input 100khz and resistive load , you will be surprised to see not square at output with such 470 ohm resistor .However audio don't need so high speed , but any oscillation or too fast signal will make mosfets shorting supply between rails , conducting at same time for some time .
Also i don't think that with 470 ohms mosfets are able to withstand sqare wave input .You mentioned 100khz , if is correct ,remember then any DC DC converter ... How mosfets are controlled there ? Resistances are just few ohms ,and discharging is done with pnp ot npn at least.
Try to simulate your output stage with sqare wave input 100khz and resistive load , you will be surprised to see not square at output with such 470 ohm resistor .However audio don't need so high speed , but any oscillation or too fast signal will make mosfets shorting supply between rails , conducting at same time for some time .
Have not done any meaurements yet ,but i'm using this desing for few years, noticed more clean high frequency output by ear .My speakers are old S-90D 8 ohms ,but modded a bit ,also have overload led for each speaker .Older amplifier (from radio magazine) harder get bass overload led to turn on ,this one easier ,supply voltage was similar .In simulation this design also performs more powerful. If interested ,i can post pictures of simulation results only .
Dear Fellows,
Right from the start I felt like the astable multivibrator was gonna be the answer. Finally I got it implemented somewhat how I like. I'm still playing with the timing.
When the protection is tripped the uppermost optocoupler allows current to flow to Q30 which is working with Q33 to generate short pulses of current. These pulses turn off Q31 briefly which disables the latch (Q24, Q25) and resets the circuit.
Additionally I opted for shunting the input stage cascode references opposed to the front end rails. Thanks to those who suggested enforcing shutdown elsewhere. Using the cascodes to disable front end current is faster acting but doesn't prevent all current from flowing. However, it does block the signal which is adequate too. The gate to source zeners at the output mosfets do burden the driver bjt;'s (rated at 200mA) with peaks just over 200mA but do impose an output current hard limit of 8.2A per device. When shutdown mode is activated some microseconds later these peaks are prevented.
The protection circuitry now acknowledges overheating, DC offset and overcurrent. It resets automatically (in all three cases) and is sonically transparent. The amplifier still needs minor inherent protection to survive until shutdown mode is established about 6us after detection. Furthermore, it doesn't tell you what went wrong and like mentioned before it auto-resets in all three cases. Ideally it would light a different LED in each case and only auto-reset when overcurrent has occurred. When overheating or DC offsetting should shutdown mode be engaged as long as those are detected? Maybe DC offsetting should only disengage the speaker relay but not prevent signal current?
Feel free to let me know what you think.
Much cheers and thanks
Ruben
Right from the start I felt like the astable multivibrator was gonna be the answer. Finally I got it implemented somewhat how I like. I'm still playing with the timing.
When the protection is tripped the uppermost optocoupler allows current to flow to Q30 which is working with Q33 to generate short pulses of current. These pulses turn off Q31 briefly which disables the latch (Q24, Q25) and resets the circuit.
Additionally I opted for shunting the input stage cascode references opposed to the front end rails. Thanks to those who suggested enforcing shutdown elsewhere. Using the cascodes to disable front end current is faster acting but doesn't prevent all current from flowing. However, it does block the signal which is adequate too. The gate to source zeners at the output mosfets do burden the driver bjt;'s (rated at 200mA) with peaks just over 200mA but do impose an output current hard limit of 8.2A per device. When shutdown mode is activated some microseconds later these peaks are prevented.
The protection circuitry now acknowledges overheating, DC offset and overcurrent. It resets automatically (in all three cases) and is sonically transparent. The amplifier still needs minor inherent protection to survive until shutdown mode is established about 6us after detection. Furthermore, it doesn't tell you what went wrong and like mentioned before it auto-resets in all three cases. Ideally it would light a different LED in each case and only auto-reset when overcurrent has occurred. When overheating or DC offsetting should shutdown mode be engaged as long as those are detected? Maybe DC offsetting should only disengage the speaker relay but not prevent signal current?
Feel free to let me know what you think.
Much cheers and thanks
Ruben
I see a problem again. Optocouplers are very slow devices. You mentioned short pulses . Look at pc817 or similar datasheet. There you will find info , that with 100ohm load of ooto-transistor 60khz bandwidth is possible. In practice with 1k or even 680ohms load resistance optocoupler operation speed is always adequate . Waveform is looking only similar to triangle, not even close to square wave. I can only guess ,what waveform you can get of reset pulses with 82k load you used. However , there exists special optocouplers with digital output, which able to overcome this limit , like those 10mbit or 50mbit rated , designed for data transmission. I would recommend you to redesign this part or use those special optocouplers. TLP2367 as example.
Looking back this post is right on the money. Indeed 470 ohms cause crosscoduction. I spend a lot of time trying to get similar performance with lower Rgs. I now landed on using an emitter follower inbetween to drive the gates. As soon as the driver runs out of current, the mosfets may crossconduct. Drivers need to much more idle current making the ops slower and requiring more overall compensation.
Also the best way to protect against overcurrent is indeed the idea suggested here. So dear ximikas, cheers to you!