I am in the process of sorting out my "hardware archives".
In doing so, I stumbled upon a very unimpressive prototype, dating back from the pre-2000, pre-sim years:
One shouldn't judge it based on its shaggy looks though: I tested it, and it performs surprising well.
Like most of my amplifiers, it has some kind of auto-bias circuit.
Very often, the auto-bias introduces unwanted artifacts, especially in the crossover region.
To evaluate the quality, I tested it under low voltage amplitude, high current, high frequency conditions with a 50kHz triangle wave and a 2 ohm load:
The result is pretty immaculate!
The quiescent current for the total amplifier is ~200mA.
The circuit is stable, clean and tolerant and could serve as a starting base for a real amplifier.
Even better, I managed to reconcile the physical circuit with my paper archives (an outstanding feat!):
The circuit is not ultra-simple, but it isn't ultra complicated either.
In doing so, I stumbled upon a very unimpressive prototype, dating back from the pre-2000, pre-sim years:
One shouldn't judge it based on its shaggy looks though: I tested it, and it performs surprising well.
Like most of my amplifiers, it has some kind of auto-bias circuit.
Very often, the auto-bias introduces unwanted artifacts, especially in the crossover region.
To evaluate the quality, I tested it under low voltage amplitude, high current, high frequency conditions with a 50kHz triangle wave and a 2 ohm load:
The result is pretty immaculate!
The quiescent current for the total amplifier is ~200mA.
The circuit is stable, clean and tolerant and could serve as a starting base for a real amplifier.
Even better, I managed to reconcile the physical circuit with my paper archives (an outstanding feat!):
The circuit is not ultra-simple, but it isn't ultra complicated either.
Ah, point to point prototyping. True do it yourself audio. No serfs making PCB's for you.
Nicely compact too. Two would fit in my ST120 chassis if it had a center tap transformer.
Nicely compact too. Two would fit in my ST120 chassis if it had a center tap transformer.
Very interesting... Will try to sim it. Very simple auto bias.
I guess you got bored during holidays 🙂
I guess you got bored during holidays 🙂
Sim attached.
1) DC offset at the output is over 200mV, I guess this amp, as a CFA, would need a servo.
2) I tried with higher rail voltages, but this would require lowering of the idle current. Not sure what would be the best way of doing this.
R22 doesn't provide enough room.
3) With rails 24V Q4/Q5 can be small transistors, with higher rails they need to be bigger.
4) From the look of the auto-bias, it seems like it could be adapted to any BJT output ?? Elvee, can you confirm?
5) How to add more output pairs? With emitter (and collector) resistors???
1) DC offset at the output is over 200mV, I guess this amp, as a CFA, would need a servo.
2) I tried with higher rail voltages, but this would require lowering of the idle current. Not sure what would be the best way of doing this.
R22 doesn't provide enough room.
3) With rails 24V Q4/Q5 can be small transistors, with higher rails they need to be bigger.
4) From the look of the auto-bias, it seems like it could be adapted to any BJT output ?? Elvee, can you confirm?
5) How to add more output pairs? With emitter (and collector) resistors???
Thanks Mr minek123 for translating schematic to simulation model. Similar symmetrical concept was used by Elektor in Cresendo amplifier in the year 1972 with Hitachi series 2SK135 & 2SJ50 Lateral MOSFET. I too think it would need a servo for stabilisation for offset which elektor used in some of its later series amplifiers. Most Cresendo builders feedback was regarding stability issues and oscillations.
The wave form in the scope seem perfect. I would also try to build it.
The wave form in the scope seem perfect. I would also try to build it.
Last edited:
Thanks
I was probably lucky, my prototype seems to have only a few mV. I didn't actually measure it, but I am going to do it
Note that in my build, D3 and D4 aren't plain 1N4148: they are BA318 or similar, more like a transistor junction than a fast, gold-doped diode. It is also possible to play with the values of R25, 26.
I see that in my build, they are paralleled with 4K7's
Yes, video transistors or tall TO92 like MPSWxy types. This prototype was a proof of concept, it didn't require a high power to demonstrate its feasibility
I think so; some adaptations might be required depending on the actual amplifier
Good question: emitter resistors are certainly needed, the collector resistors could remain common.
The simulated THD figure looks disappointing, compared with the pristine triangle it managed. I am also going to measure it.
I am lucky: with the input shorted, the output offset is 12mV, and 11mV when it is terminated on a 4K7 resistor.
All my small transistors are BC547/557, maybe they have a better symmetry/complementarity than the types used in the sim.
The THD is not very good: at 25Vpp output on 4 ohm, it reaches 1%. It drops very quickly if the amplitude is decreased: it falls under 0.1% at 18Vpp
All my small transistors are BC547/557, maybe they have a better symmetry/complementarity than the types used in the sim.
The THD is not very good: at 25Vpp output on 4 ohm, it reaches 1%. It drops very quickly if the amplitude is decreased: it falls under 0.1% at 18Vpp
With 42V rails the DC offset was over 500mV. This is not surprising - all CFA designs usually require a servo..
A trimmer could be added (adjusting R10/R13) to get rid of the offset, but this will only work for the sim, in real life, this offset will drift further.
Unless we get rid of the dc offset, Thd number will be high...
I'm tempted to try to use/adopt the auto-bias from this amp, with and amp that has VFA input, and BJT output.
Or even better - a VFA with op-amp at the input..
If op-amp is used, I think it's more rational to use it in the input stage, rather then as the CFA servo..
Wonder if the same (or similar) implementation of auto-bias could be used for HexFet output?
A trimmer could be added (adjusting R10/R13) to get rid of the offset, but this will only work for the sim, in real life, this offset will drift further.
Unless we get rid of the dc offset, Thd number will be high...
I'm tempted to try to use/adopt the auto-bias from this amp, with and amp that has VFA input, and BJT output.
Or even better - a VFA with op-amp at the input..
If op-amp is used, I think it's more rational to use it in the input stage, rather then as the CFA servo..
Wonder if the same (or similar) implementation of auto-bias could be used for HexFet output?
Here is updated sim:
1) Diodes replaced by transistors (so the auto-bias looks more like a real current mirror now), with parallel resistors added
These resistors don't seem to change anything..
2) 2 resistors added to the input to correct DC offset, in my sim it's 1mV now
3) Unfortunately, even with 1mV offset, the Thd is still 0.5%
4) I tried replacing transistors with BC547/557 - offset is different, but still there (more on the '+' side), and Thd is even higher.
Judging from numerous CFA designs that we've seen on this forum, a servo is a must for a CFA
5) In this amp VAS (nor feedback) doesn't need any compensation (caps), so this should be a very fast amp
1) Diodes replaced by transistors (so the auto-bias looks more like a real current mirror now), with parallel resistors added
These resistors don't seem to change anything..
2) 2 resistors added to the input to correct DC offset, in my sim it's 1mV now
3) Unfortunately, even with 1mV offset, the Thd is still 0.5%
4) I tried replacing transistors with BC547/557 - offset is different, but still there (more on the '+' side), and Thd is even higher.
Judging from numerous CFA designs that we've seen on this forum, a servo is a must for a CFA
5) In this amp VAS (nor feedback) doesn't need any compensation (caps), so this should be a very fast amp
Last edited:
Here is my attempt to use this auto-bias with a relatively 'normal' VFA. I selected the simplest of the amps from
'unusual amp' thread - Wiederhold 77, and added auto-bias. This is just 1st, rough attempt, nothing was optimized,
or thoroughly tested/simmed yet.
Auto-bias transistors (Q8/Q9) dissipate 1W, seems little high, maybe there is a way to make it better..
Elvee, do you think this solution will be workable?
Original sim for Wiederhold's amp with 'normal' bias, also attached for comparison purpose.
'unusual amp' thread - Wiederhold 77, and added auto-bias. This is just 1st, rough attempt, nothing was optimized,
or thoroughly tested/simmed yet.
Auto-bias transistors (Q8/Q9) dissipate 1W, seems little high, maybe there is a way to make it better..
Elvee, do you think this solution will be workable?
Original sim for Wiederhold's amp with 'normal' bias, also attached for comparison purpose.
Q4/Q5 are not biased correctly in the above sim. They are barely conducting... with very low current..
Yes, probably: the autobias spreader transistors should be wired as Vbe multipliers to match the threshold voltages. Currently, the multiplication factor is ~1
I meant paralleling the 4K7 with R25, 26.
With 4K7, it does not seem to change the quiescent current significantly in the sim, but when the value is halved, so is the quiescent current:
You have connected the auto-bias directly between the emitters of Q6 and Q7. The circuit needs some compliance to be able to operate, but the emitter followers are too stiff.
With my settings of LTspice, the circuit oscillated, and I had to compensate the opamp with a 100pF, but that's another story
In this version I plugged auto-bias right after VAS, in the same position where traditional Vbe multiplier was.
This seems to work ok.
This seems to work ok.
Good that it works, although I find it a bit surprising: at that location, the spreader sees ~4 Vbe, meaning that each Q11, 12 should be wired as a 2x Vbe-mult, with a suitable B-E resistor for instance.
The fact that it manages to cope with a higher voltage without change is a good sign: it shows the robustness of the scheme. The auto-bias transistors simply do more work.
I don't think you could push it to Hexfets without mods though, that would be far-fetched
The fact that it manages to cope with a higher voltage without change is a good sign: it shows the robustness of the scheme. The auto-bias transistors simply do more work.
I don't think you could push it to Hexfets without mods though, that would be far-fetched
Elvee, for Hexfets it was necessary to use Vbe mult, as you said.
Here is an example (sim attached) that seem to work, and also uses multiple output pairs.
The only problem is that auto-bias is adding a lot of distortion.
In the sim, the same amp with traditional Vbe multiplier, was showing 0.000000% Thd at 1kHz.
Now it shows 0.00xx%. I guess this may not matter in real life...
Here is an example (sim attached) that seem to work, and also uses multiple output pairs.
The only problem is that auto-bias is adding a lot of distortion.
In the sim, the same amp with traditional Vbe multiplier, was showing 0.000000% Thd at 1kHz.
Now it shows 0.00xx%. I guess this may not matter in real life...
Last edited:
As I mentioned in the opening post, autobias circuits are often plagued with unwanted artefacts.
This one looked relatively clean, as the harsh triangle test showed, but if you venture in the ppm region and below, everything matters.
The main problem I think, is that the sensing transistors pollute the VAS signal with class B waveforms.
A way to minimize that is to make sure that the control currents are kept very small compared to the VAS current. In my example, the VAS current was ~50mA, but in your latest circuit, it is ~10x times smaller, meaning the artefacts have a relative impact 10 times greater.
A possible remedy is to scale up the resistors to the maximum possible in that part of the circuit (that would probably be significantly more than ten times, especially if you use superbeta transistors for Q19, 20).
Resistors affected would be R34, 35, and 44 to 48.
You could also use small MOSFETs as Q19, 20: it would allow the reduction of the control currents to the µA level, with the additional benefit that ~no multiplication would be required.
Another (non-exclusive) option is to inject the small polluting signal somewhere in the amp, in antiphase to cancel its effect.
Such an error correction could gain you another order of magnitude in the THD reduction
It does matter: if you make the effort of designing a super-amplifier, it makes no sense to ruin it with some silly gadget, even if it makes the circuit easier and more convenient: you want your cake AND eat it too
I have tested the A-B scheme on another of my amplifiers, The Old-Fashioned lite.
It works, and although it adds a bit of distortion, it is insignificant (from 0.08 to 0.093%), and may not actually be caused by the A-B itself (Iq is changed, headroom is reduced due to the sensing resistors, etc.).
The values are adjusted for a quiescent current of 50mA.
Interestingly, the presence of the A-B circuit allows the introduction of a crude error-correction scheme which now causes a small but significant reduction of the THD, to 0.059%. This EC is simply R16 connected to the output
It works, and although it adds a bit of distortion, it is insignificant (from 0.08 to 0.093%), and may not actually be caused by the A-B itself (Iq is changed, headroom is reduced due to the sensing resistors, etc.).
The values are adjusted for a quiescent current of 50mA.
Interestingly, the presence of the A-B circuit allows the introduction of a crude error-correction scheme which now causes a small but significant reduction of the THD, to 0.059%. This EC is simply R16 connected to the output