I've been trying to design a Class-AB power amplifier using two Sziklai Pairs in the output stage, using BD140+TIP35 (the PNP pair) and BD139+TIP36 (the NPN pair).
The design is in the image I attached.
It has some a considerable amount of distortion in the whole spectrum, plus clipping when I try to amplify a larger bass signal, a reasonably treble and hard clipping much before the clipping should actually occur.
And the thing is that I think I don't know what I'm doing anymore. In the beginning, I tested the design in MultiSIM and apparently it worked, but in reality it had a ridiculously large RF instability that caused the current drawn by U2 (I used a multimeter in the same place of the one in the design) to increase by 10 fold (from 20 to 200 mA) when I simply connected two wires to the output, the wires being connected to nothing, just because of their capacitance. And when it did this there was a 100 R resistance limiting the total current in my source itself, could be much larger without it (I only removed it after making the circuit stable).
It took me long to figure out how to fix this, then I found this design that used a capacitor from the output to the stage before the output stage (C5).
Connecting a capacitor similarly (also C5 in mine), the instability went away, but I'm not sure if it makes sense in my configuration, even why here I had to keep its value small, otherwise there was much more distortion. With 100 uF there was distortion at low volumes even, with 1 uF it worked a lot better.
But still, no matter how finely I adjust the bias in the rubber diode, there's still noticeable harmonic and non-harmonic distortion, the worst being a hard clipping at larger volumes, in bass in particular.
But it's happening really before it should.
Q7 and Q8 transistors in my diagram should be limiters, activated only when the voltage drop on the 0.15 R resistances were around 1.4 V or so, so a current of 9A or more, which I doubt I'm hitting before this clipping I'm hearing.
Should I test removing them just for a test, or perhaps this is coming from another source?
Any way, I seem to never get a really good quality audio no matter how much I try adjusting the biasing rubber diode, the clipping isn't the only issue.
So I'm asking to check what flaws I have in my circuit that I should address.
The design is in the image I attached.
It has some a considerable amount of distortion in the whole spectrum, plus clipping when I try to amplify a larger bass signal, a reasonably treble and hard clipping much before the clipping should actually occur.
And the thing is that I think I don't know what I'm doing anymore. In the beginning, I tested the design in MultiSIM and apparently it worked, but in reality it had a ridiculously large RF instability that caused the current drawn by U2 (I used a multimeter in the same place of the one in the design) to increase by 10 fold (from 20 to 200 mA) when I simply connected two wires to the output, the wires being connected to nothing, just because of their capacitance. And when it did this there was a 100 R resistance limiting the total current in my source itself, could be much larger without it (I only removed it after making the circuit stable).
It took me long to figure out how to fix this, then I found this design that used a capacitor from the output to the stage before the output stage (C5).
Connecting a capacitor similarly (also C5 in mine), the instability went away, but I'm not sure if it makes sense in my configuration, even why here I had to keep its value small, otherwise there was much more distortion. With 100 uF there was distortion at low volumes even, with 1 uF it worked a lot better.
But still, no matter how finely I adjust the bias in the rubber diode, there's still noticeable harmonic and non-harmonic distortion, the worst being a hard clipping at larger volumes, in bass in particular.
But it's happening really before it should.
Q7 and Q8 transistors in my diagram should be limiters, activated only when the voltage drop on the 0.15 R resistances were around 1.4 V or so, so a current of 9A or more, which I doubt I'm hitting before this clipping I'm hearing.
Should I test removing them just for a test, or perhaps this is coming from another source?
Any way, I seem to never get a really good quality audio no matter how much I try adjusting the biasing rubber diode, the clipping isn't the only issue.
So I'm asking to check what flaws I have in my circuit that I should address.
Attachments
Last edited:
What you mean "useless"? Without it in my circuit there's an absurd RF oscillation drawing a ton of current, not to mention a lot of high-pitched, audible noise if I try to play anything without it. Only with it it became anywhere stable. I didn't had it in the beginning.
Plus, the DC point of the output of the opamp itself is around 9V, isn't this enough room from GND? Because it should vary very little around 9V, only +- 0.5 V around 9V until the output stage swing is at maximum, according to the simulation. It doesn't swing close to GND, the base of Q1 however is close to GND.
Plus, the DC point of the output of the opamp itself is around 9V, isn't this enough room from GND? Because it should vary very little around 9V, only +- 0.5 V around 9V until the output stage swing is at maximum, according to the simulation. It doesn't swing close to GND, the base of Q1 however is close to GND.
Last edited:
I'll post the .msm file, just let me figure out how to send the models for the TIP35C and TIP36C I'm using, cause I had to spend days making them. It didn't had them natively and I even found some online, but they were nonsense, had a much larger gain and other largely different parameters from those in the datasheet (which fitted the ones I had in hands much better as well).
I had to reverse-engineer graphs from its datasheet and derive the adequate parameters so it would behave as it should, but I don't know where are the model files I made.
I had to reverse-engineer graphs from its datasheet and derive the adequate parameters so it would behave as it should, but I don't know where are the model files I made.
The proper way to fix it is to add the usual level shifting stage in between the op amp and Q1, such that the output of the op amp wants to sit between its supply and GND. Then on to the stability issues. You will probably find out quickly that CFPs are a liability you can’t afford, say the hell with them, and switch to EFs. In addition to reworking the entire compensation scheme. Values are all wrong.
C5 is a very bad way to stabilize your CFPs. The proven method is a Zobel network. Using an op-amp front end is another recipe for oscillations because the OP-amp has its own "dominant pole" which competes with the VAS stage. Successful solutions usually choose the VAS and take measures to neutralize the op-amp compensation. If you need more than 100pF on the VAS for compensation then you will get terrible slew limiting, so back off the loop gain, add some phase lead (1nF is too large). Q5+Q6 CCS is also probably unstable and needs either a capacitor across Q6 or a base resistor. The problem with your feedback is that an open input changes the feedback and again leads to oscillations. etc... You need to do some phase margin simulation.
PS: The current limit needs resistance in the base circuits so that the first overload doesn't blow them out. I suggest a 2:1 resistor divider instead of the diodes.
PS: The current limit needs resistance in the base circuits so that the first overload doesn't blow them out. I suggest a 2:1 resistor divider instead of the diodes.
The first place I would try is around R3, but it has to be carefully tuned to match the second pole or correct the feedback phase at the critical loop gain zero crossing. I would discourage any beginners from using an op-amp as a front end because it is so difficult to stabilize, unless the output is just an EF, but that limits you to the op-amp supply voltage limits. There is a long list of other topologies for op-amp front ends to amuse those with experience, including taking the output from the op-amp rails and applying feedback to the op-amp output.
Hi guferr,
How does your power supply circuit look like? Since the inverting input of TL072 is driven by the power supply rail (divided by the pot), independent of the signal, every fluctuation of the power supply will appear at the output. You will want a low pass filter to reduce it. This is something simulations may miss.
What kind of preamp do you use to drive the input? Is it capable of driving something less than 500 ohm with 47uF?
The output transistor pairs (BD139/TIP36, BD140/TIP35) are running under extremely different voltages. Using +/- rails will eliminate that problem together with others. If you can supply two power supply rails, then modify them to be +/- 15~18V. From the beginning, you are not getting less than +/-10V swing from TL072, so +/- 15V rail should give you simpler and better circuit with similar output swing.
Just my thoughts. Good luck and enjoy!
Best,
Satoru
How does your power supply circuit look like? Since the inverting input of TL072 is driven by the power supply rail (divided by the pot), independent of the signal, every fluctuation of the power supply will appear at the output. You will want a low pass filter to reduce it. This is something simulations may miss.
What kind of preamp do you use to drive the input? Is it capable of driving something less than 500 ohm with 47uF?
The output transistor pairs (BD139/TIP36, BD140/TIP35) are running under extremely different voltages. Using +/- rails will eliminate that problem together with others. If you can supply two power supply rails, then modify them to be +/- 15~18V. From the beginning, you are not getting less than +/-10V swing from TL072, so +/- 15V rail should give you simpler and better circuit with similar output swing.
Just my thoughts. Good luck and enjoy!
Best,
Satoru
Last edited:
Hmmm, did I kill the thread??
Well, one thing I forgot to check: This amplifier is an inverted topology amp and the non-inverting and inverting inputs of TL072 in the schematic are flipped, right?
Due to the issue of virtual ground for the non-inverted input, inverted amp is not a good choice for single supply amplifiers. If you want to stick to inverted topology, then design with +/- dual power supply. If you want to stick to single power supply, then use non-inverting topology. As I mentioned before, you have an issue of low impedance with huge capacitance at the input. Moving to non-inverting topology will eliminate that issue.
Sziklai pair is nice when you want to drive the output with less swing, but has an issue with stability (anyone can confirm this?) compared to the regular Darlington topology. Since the circuit has ample voltage to spare here and there, using regular Darlington would be advisable.
Best,
Satoru
Well, one thing I forgot to check: This amplifier is an inverted topology amp and the non-inverting and inverting inputs of TL072 in the schematic are flipped, right?
Due to the issue of virtual ground for the non-inverted input, inverted amp is not a good choice for single supply amplifiers. If you want to stick to inverted topology, then design with +/- dual power supply. If you want to stick to single power supply, then use non-inverting topology. As I mentioned before, you have an issue of low impedance with huge capacitance at the input. Moving to non-inverting topology will eliminate that issue.
Sziklai pair is nice when you want to drive the output with less swing, but has an issue with stability (anyone can confirm this?) compared to the regular Darlington topology. Since the circuit has ample voltage to spare here and there, using regular Darlington would be advisable.
Best,
Satoru
Last edited:
Hi all, sorry for the delay. I managed to make this stable and non-distorted without C5.
As steveu recomended, I've added a small cap around R3 to give a bit of phase lead, and reduced the phase-lead compensation in the feedback loop.
Turns out that, despite MultiSIM doesn't account for parasitic capacitances, just checking the phase response of each part (and for the whole) and trying to make them smoother (reducing the number of peaks to a minimum) made it stable. I had to also use a resistance in series with the capacitance of the phase-lead compensation to make its phase-response smoother.
Plus, as satoru mentioned, I forgot to add a capacitor in the voltage divider at the inverting input of the TL072. This eliminated most distortion. It was supposed to exist, but I think I eliminated it due to simulation issues and forgot to re-add in my real circuit.
As for the clipping, I was just being dumb. It was actually hitting maximum voltage swing and hence hard-clipping. I just thought it wasn't because it was too little output for that, but then I noticed I used a capacitor too small in the output coupling, at clipping with a 30 Hz sinusoidal waveform it had 17 VAC RMS (24 V peak) before the output capacitor but only a mere 6 VAC RMS after the capacitor.
This is my new design (attached).
As far as it goes it's looking good and stable against oscillations, but I need to make a more stressful test to see how thermally stable this got.
As steveu recomended, I've added a small cap around R3 to give a bit of phase lead, and reduced the phase-lead compensation in the feedback loop.
Turns out that, despite MultiSIM doesn't account for parasitic capacitances, just checking the phase response of each part (and for the whole) and trying to make them smoother (reducing the number of peaks to a minimum) made it stable. I had to also use a resistance in series with the capacitance of the phase-lead compensation to make its phase-response smoother.
Plus, as satoru mentioned, I forgot to add a capacitor in the voltage divider at the inverting input of the TL072. This eliminated most distortion. It was supposed to exist, but I think I eliminated it due to simulation issues and forgot to re-add in my real circuit.
As for the clipping, I was just being dumb. It was actually hitting maximum voltage swing and hence hard-clipping. I just thought it wasn't because it was too little output for that, but then I noticed I used a capacitor too small in the output coupling, at clipping with a 30 Hz sinusoidal waveform it had 17 VAC RMS (24 V peak) before the output capacitor but only a mere 6 VAC RMS after the capacitor.
This is my new design (attached).
As far as it goes it's looking good and stable against oscillations, but I need to make a more stressful test to see how thermally stable this got.
22k or so resistor is needed between the pot and the op amp input pin. Probably should be on the inside of such a large (47 uF) cap. Add 300 pf right at the pin. One good thing about TL072/LF353 is that if you drive them outside common mode range they don’t just fry instantly. They just clamp it to the op amp’s rail. But the resulting input current needs to be limited to 4 mA. Phase reversals may occur, so speaker DC protection is necessary. In this case an output cap. When DC does go to the speaker due to an output fault, be sure the feedback resistor is high enough to ensure that full rail voltage applied to the input pin through it is under 4 mA.
It changes the DC point of the output but not so much the current from what I've seen, but this is going to be fixed anyway. It's there just for fine-tuning during assembly, as well as R14.
Yes yes, it's just because I rotated it wrong when rearranging to make this version to put here, but I think I'm going to use just two resistors in the future, this seems to be limiting before it should, according to the simulation.