Hi all,
In 2012 I had an idea for a non-switching class-AB amplifier with class-AB control loop (similar to the one in Electronics World February 1996) that also has a class-A mode. With a potmeter, you could continuously vary the bias current. The amplifier would indicate by means of an LED when it goes into class AB at the output current peaks. That way you could use it in class A without wasting lots of power when you listen at low volumes.
Now five years later I still haven't built it (mainly because my old non-switching class-AB amplifier is good enough for me), so I thought I would just post the untested idea in case anyone else is interested. Attached is a simplified schematic, an explanation and the full schematic will follow later.
Best regards,
Marcel
In 2012 I had an idea for a non-switching class-AB amplifier with class-AB control loop (similar to the one in Electronics World February 1996) that also has a class-A mode. With a potmeter, you could continuously vary the bias current. The amplifier would indicate by means of an LED when it goes into class AB at the output current peaks. That way you could use it in class A without wasting lots of power when you listen at low volumes.
Now five years later I still haven't built it (mainly because my old non-switching class-AB amplifier is good enough for me), so I thought I would just post the untested idea in case anyone else is interested. Attached is a simplified schematic, an explanation and the full schematic will follow later.
Best regards,
Marcel
Attachments
Last edited:
The schematic shows an output stage of an amplifier. The earlier stages are to provide two signal currents in anti-phase to the gates of Q1 and Q2, and to draw DC bias currents. For example, the inputs of the schematic can be connected to an NPN differential pair.
OK, first forget about Q10 and replace the LED of OK1 with a short circuit. You then basically have a class-AB control loop of the type that Johan H. Huijsing and Frans Tol invented in 1976, but with a different non-linear network.
When the current through Q1 gets too small, Q5 turns on, and via the current mirror it increases the gate drive of both Q1 and Q2. Similarly, when the current through Q2 gets too small, Q4 turns on and again both Q1 and Q2 get more gate drive. When Q1 and Q2 are at their quiescent point, Q4 and Q5 both conduct a bit. Neglecting the voltage drop that the currents through Q4 and Q5 produce across the 0.4 ohm resistors, the control law is exp(-0.4 ohm*IDQ1*q/(kT)) + exp(-0.4 ohm*IDQ2*q/(kT)) = constant, with kT/q ~= 26 mV. This is all exactly the same as in my Electronics World article from February 1996.
With Q10 added and the potmeter increased to a few hundred ohms, it is Q10 that will control the quiescent current. It senses the sum of the voltages across the 0.4 ohm resistors, changing the control law into something close to IDQ1 + IDQ2 = Vpotmeter/0.8 ohm. That is, as long as Q10 is in control, the output stage works in class A.
When the output signal current gets so large that the output stage can not stay in class A anymore, either Q4 or Q5 will kick in and will ensure that Q2 or Q1, respectively, is not switched off completely. That is, the circuit transitions into non-switching class AB. The LED of optocoupler OK1 turns on when this happens. The other side of the optocoupler drives a monostable multivibrator that drives an LED (not shown on the schematic), which indicates to the user that the circuit has gone into class AB.
OK, first forget about Q10 and replace the LED of OK1 with a short circuit. You then basically have a class-AB control loop of the type that Johan H. Huijsing and Frans Tol invented in 1976, but with a different non-linear network.
When the current through Q1 gets too small, Q5 turns on, and via the current mirror it increases the gate drive of both Q1 and Q2. Similarly, when the current through Q2 gets too small, Q4 turns on and again both Q1 and Q2 get more gate drive. When Q1 and Q2 are at their quiescent point, Q4 and Q5 both conduct a bit. Neglecting the voltage drop that the currents through Q4 and Q5 produce across the 0.4 ohm resistors, the control law is exp(-0.4 ohm*IDQ1*q/(kT)) + exp(-0.4 ohm*IDQ2*q/(kT)) = constant, with kT/q ~= 26 mV. This is all exactly the same as in my Electronics World article from February 1996.
With Q10 added and the potmeter increased to a few hundred ohms, it is Q10 that will control the quiescent current. It senses the sum of the voltages across the 0.4 ohm resistors, changing the control law into something close to IDQ1 + IDQ2 = Vpotmeter/0.8 ohm. That is, as long as Q10 is in control, the output stage works in class A.
When the output signal current gets so large that the output stage can not stay in class A anymore, either Q4 or Q5 will kick in and will ensure that Q2 or Q1, respectively, is not switched off completely. That is, the circuit transitions into non-switching class AB. The LED of optocoupler OK1 turns on when this happens. The other side of the optocoupler drives a monostable multivibrator that drives an LED (not shown on the schematic), which indicates to the user that the circuit has gone into class AB.
Last edited:
Hi, Marcel!
With modern microprocessor-controlled relay-based ladder attenuators and typical line input levels we can simply change idle current based on adjusted volume.
So with common all-day low-level listening we can run moderate idle currents in OPS, and when we want a party levels we can run as high as possible current based on heatsink capability.
Hi Marcel,
In the particular case of BJT OPS there is an unique optimum bias level. Anything over or under this optimum bias will result in high level of distortion
The classB crossover distortion is caused by variation in amplifier output impedance with load current level at crossover point.
The optimal class B biasing for minimal crossover distortion requires that R * gm =1 where R is the total ohmic resistance seen from the emitter towards the base of one output emitter follower (one output power BJT).
This resistance is : Re + re + Rs/(beta +1) where Re is the emitter resistor on which you measure Vq, re is the parasitic internal emitter resistance, Rs is the total base + source resistance. If Re is dominant, then the condition becomes gm * Re = 1 which means Io * Re = Vt = 26mV. This is the usual condition (Self) because gm = Io/Vt where Io is the bias and Vt is the thermal voltage 26 mV at room temp.
If Re is made smaller to avoid losses, then re may not be neglected in front of Re and gmR=1 becomes gm ( Re + re ) =1
This means Io Re + Io re = 26mV and you are measuring the first term.
It is then normal that at optimum IoRe should be lower than 26mV.
In other words, if you want optimum class B bias and also as much as possible Class A, you can not do it by increasing bias over 20 ...
26mV (20mV for 0302/0281 with 0.1ohm Re), voltage measured over one emitter resistor. What other option do we have if we want more A class? - to increase the number of output devices biased at optimum class B current.
In the particular case of BJT OPS there is an unique optimum bias level. Anything over or under this optimum bias will result in high level of distortion
The classB crossover distortion is caused by variation in amplifier output impedance with load current level at crossover point.
The optimal class B biasing for minimal crossover distortion requires that R * gm =1 where R is the total ohmic resistance seen from the emitter towards the base of one output emitter follower (one output power BJT).
This resistance is : Re + re + Rs/(beta +1) where Re is the emitter resistor on which you measure Vq, re is the parasitic internal emitter resistance, Rs is the total base + source resistance. If Re is dominant, then the condition becomes gm * Re = 1 which means Io * Re = Vt = 26mV. This is the usual condition (Self) because gm = Io/Vt where Io is the bias and Vt is the thermal voltage 26 mV at room temp.
If Re is made smaller to avoid losses, then re may not be neglected in front of Re and gmR=1 becomes gm ( Re + re ) =1
This means Io Re + Io re = 26mV and you are measuring the first term.
It is then normal that at optimum IoRe should be lower than 26mV.
In other words, if you want optimum class B bias and also as much as possible Class A, you can not do it by increasing bias over 20 ...
26mV (20mV for 0302/0281 with 0.1ohm Re), voltage measured over one emitter resistor. What other option do we have if we want more A class? - to increase the number of output devices biased at optimum class B current.
For AB class OPS.
Barney Oliver.
http://www.diyaudio.com/forums/soli...et-point-class-ab-power-amps.html#post4320070
Slightly better resolution:
https://www.dropbox.com/s/ugp7y56midblsyg/Barney Oliver distortion in B-class.pdf?dl=0
Just increase idle current to keep OPS in class A and increase feedback depth with its effective bandwidth around OPS.
Well, what will happen when class A idle current will be exceeded by the requested program level and the OPS will abruptly enter in class B, which, by the way, has an idle current heavily over-biased?
That's an interesting thought! A more old-fashioned variant would be to mechanically couple the volume and bias current potentiometers. In either case, you will need to take some safety margin because you usually don't know the line levels very accurately, nor the current that your loudspeakers need at a given signal voltage level.
Just how?
We know industry's standard 2,83 V line output, we know input stepped attenuator setting, than we can estimate and set needed idle current based on load impedance (let's assume current phase doesn't matter).
Of course, being a little bit paranoidy we can provide some margin, say 30-50%...
No, with any bias current at this point all of the current flows through the load, just because other shoulder are closed, there are no way to flow. So we have mostly the simple voltage divider between Re and Rload.
Of course, some power will be dissipated at Re, but we are speaking about A-class amp, so we are ready to be insane in dissipation.
The next is a real signal peak factor, say 3-4. Just because of this for moderate average power of 10 Wt we are forced to be able to provide 100 Wt peaks, and in any conditions OPS active devices dissipation will prevail over the load power to a (peak factor^2).
And, of course, we are speaking about modern amplifiers, so closing ~100 dB feedback loop around OPS having ~-40 dB THD can result in amplifier with THD figures at an order of best available DAC's.
Yup. But don't forget to put some negative tempco device inside the bias loop. Time constant of 5-10 sec would be good enough.
Just second potz or 4-5 way switch for margin setup and adjusting based on the speakers resistance.
Last edited:
Then you get slightly more distortion on occasional signal peaks, while the softer parts of the music are still processed in class A.
In any case, your whole analysis applies to a very specific topology that I'm not using, namely a bipolar complementary emitter follower stage with emitter resistors, with the stage being driven from a lowish impedance and with the way the current is distributed between the transistors set by the transistors themselves. I use a current driven MOSFET output stage with a class-AB bias loop that forces the current distribution law. To the best of my knowledge, my output stage doesn't have such a distortion minimum anyway.
Attached is the main schematic. The right-hand side wasn't scanned in properly; the rightmost diode is actually a 3V9, 400 mW zener and the upper side of the rightmost 22 nF capacitor is meant to be connected to the output relay.
The SN74AHCU04's are used as transistor arrays. I have done noise measurements on various brands of 74HCU04 and 74AHCU04 and found that the Texas Instruments SN74AHCU04 had the lowest level of 1/f noise, especially when the PMOS side is biased at a much larger current than the NMOS side.
The BCM847's on the right are used as translinear one-quadrant multipliers. They act as parts of a safe operating area protection circuit. The 220 kohm resistors sense the voltage across the output MOSFETs, the 8.2 kohm resistors the current. The multipliers then calculate the power by multiplying these. The ((2.2 uF//10 kohm) + 2.7 kohm) // 220 nF networks are a scaled model of the thermal impedance between the IRFP140N MOSFET junctions and the heatsink.
The SN74AHCU04's are used as transistor arrays. I have done noise measurements on various brands of 74HCU04 and 74AHCU04 and found that the Texas Instruments SN74AHCU04 had the lowest level of 1/f noise, especially when the PMOS side is biased at a much larger current than the NMOS side.
The BCM847's on the right are used as translinear one-quadrant multipliers. They act as parts of a safe operating area protection circuit. The 220 kohm resistors sense the voltage across the output MOSFETs, the 8.2 kohm resistors the current. The multipliers then calculate the power by multiplying these. The ((2.2 uF//10 kohm) + 2.7 kohm) // 220 nF networks are a scaled model of the thermal impedance between the IRFP140N MOSFET junctions and the heatsink.
Last edited:
This is the tone control, simple version that only controls the low frequencies. It matches the differences between equal-loudness contours reasonably well. I find a loudness-derived control useful in case you have to listen at very low volumes. It is basically the dual circuit of a tone control published by L. V. Viddeleer in the 1940's and 1950's. The control part (right side of the schematic) is common for the left and right channels.
Power supply and phono preamplifier. The second-order filter on the right is meant to cancel rumble that is in anti-phase between left and right (as in Bob Cordell's VinylTrak), but it actually messes up the channel separation quite severely, so it is probably not a good idea to implement it like this.
Hi Dimitri,
Like I wrote in the first post, I still haven't built it after five years, so chances are I never will.
As I haven't actually built the circuit, it is likely to need some debugging. It doesn't seem fair to me to have a buggy circuit published in an audio magazine. Worst of all, my favourite audio bookazine Linear Audio stopped recently.
Still, I don't want the idea to get wasted, so I just started this thread.
Like I wrote in the first post, I still haven't built it after five years, so chances are I never will.
As I haven't actually built the circuit, it is likely to need some debugging. It doesn't seem fair to me to have a buggy circuit published in an audio magazine. Worst of all, my favourite audio bookazine Linear Audio stopped recently.
Still, I don't want the idea to get wasted, so I just started this thread.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.