On this forum, my ideas seldom catch the wind ... but I drop it in anyway.
I was about to make Sahlsten's complemetarysymmetrical mosfet ampilifier.
At first, the idea was to make the old circuit purely, but I started simulating it wiht LTspice ...
I realized that Sahlsten's performance in terms of distortion values does not meet my expectations.
As a result, I looked at several schemas ... and more schemas, simulated them and studied their best aspects.
There is a lot of good stuff, and also plenty of incorrect information.
It is worth filtering what you read and, in addition to simulation, studying it with practical examples.
I saw, for example, an amplifier circuit where two resistors of the same value were used in parallel but in different directions.
The claim was that the inductances of the resistor material turned on the coil would cancel each other out.
I had to take two coil springs on the table.
Even if you turn one of them around, the threads are still parallel.
All the nonsense is convincingly written about it.
Few potential shemas were introduced in Bob Cordell's book.
"Designing Audio Power Amplifiers" presented a fast correction circuit near the power fets.
The distortion values improve considerably when using the connection.
The disadvantage is that all the components inside Cordell's correction circuit affect the thermal stability of the quiescent current.
Thermal compensation of the quiescent current can be done, but the connection becomes quite complicated.
There is also a possibility of heat-oscillation/run if heat is not transferring fast enough from component to another.
I decided to go alternate way.
I found a promising schema on another forum, RH PA 0.9.
In this schema the goal is to to make the amplifier so fast that the feedback of a long loop can correct the errors.
With the author's permission I modified it to meet my needs.
The original semiconductors have been changed faster and more voltage-resistant, with higher output power and lower distortion in mind.
1. The input stage made with MPSA42 / MPSA92 is a very fast cascaded complementarysymmetrical differential stage.
The "upper" stages are lifted (+-)15 volts from gnd-level so the input transistors could be some low-voltage transistors or small-signal fets.
As MPSA42/92 are very fast, there is no real need to replace them with another types.
2. The next voltage amplification stage MPSA92 MJE15033 - and its complement are implemented a little differently.
Instead of a pure Darlington connection, the collector of the driving transistor is connected to GND via a resistor.
This implementation is because, according to the simulation, distortion increases with a pure Darlington connection.
The collector resistor is only necessary in a possible overdrive situation, when the base current of the power transistors is limited to an allowable level.
3. The emitterfollower stage controlling the FETs operates in class A within audiofrequencyrange at all powers.
4. The output stage is of a traditional style made with Exicon lateral FETs ECW20N20 and ECW20P20.
The quiescent current adjustment and thermal compensation are done here with a few parts and the adjustment keeps its set value in the range of 20 to 80 degrees centigrade.
The following values have been simulated with ideal voltage sources into an 8 ohm load.
The simulation has used values recommended in Cordell's book, e.g. max timestep which depends on the test frequency.
1kHz .maxstep=0.48831106u ... 20kHz .maxstep=0.02441555u
Frequency response 1.2Hz - 75kHz -1dB ... with 1uH coil.
Distortion levels simulated with optional gain A=37.4
Distortion 1kHz 1W/8ohm ..2nd -122dB ..3rd -130dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.000000%
Distortion 1kHz 4W/8ohm ..2nd -117dB ..3rd -120dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.000000%
Distortion 1kHz 16W/8ohm ..2nd -114dB ..3rd -113dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.000000%
Distortion 1kHz 64W/8ohm ..2nd-110dB ..3rd -109dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.000000%
Distortion 20kHz 1W/8ohm ..2nd -104dB ..3rd -110dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.000697%
Distortion 20kHz 4W/8ohm ..2nd -99dB ..3rd -96dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.002052%
Distortion 20kHz 16W/8ohm ..2nd -95dB ..3rd -88dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.004842%
Distortion 20kHz 64W/8ohm ..2nd -91dB ..3rd -86dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.006681%
At about 200W / 8ohm power (1kHz) the third harmonic starts to raise its head, which means going worse side of the -100dB distortion level.
Noise analysis over the audio range 20-20k from the speaker output gives 65uV.
I was about to make Sahlsten's complemetarysymmetrical mosfet ampilifier.
At first, the idea was to make the old circuit purely, but I started simulating it wiht LTspice ...
I realized that Sahlsten's performance in terms of distortion values does not meet my expectations.
As a result, I looked at several schemas ... and more schemas, simulated them and studied their best aspects.
There is a lot of good stuff, and also plenty of incorrect information.
It is worth filtering what you read and, in addition to simulation, studying it with practical examples.
I saw, for example, an amplifier circuit where two resistors of the same value were used in parallel but in different directions.
The claim was that the inductances of the resistor material turned on the coil would cancel each other out.
I had to take two coil springs on the table.
Even if you turn one of them around, the threads are still parallel.
All the nonsense is convincingly written about it.
Few potential shemas were introduced in Bob Cordell's book.
"Designing Audio Power Amplifiers" presented a fast correction circuit near the power fets.
The distortion values improve considerably when using the connection.
The disadvantage is that all the components inside Cordell's correction circuit affect the thermal stability of the quiescent current.
Thermal compensation of the quiescent current can be done, but the connection becomes quite complicated.
There is also a possibility of heat-oscillation/run if heat is not transferring fast enough from component to another.
I decided to go alternate way.
I found a promising schema on another forum, RH PA 0.9.
In this schema the goal is to to make the amplifier so fast that the feedback of a long loop can correct the errors.
With the author's permission I modified it to meet my needs.
The original semiconductors have been changed faster and more voltage-resistant, with higher output power and lower distortion in mind.
1. The input stage made with MPSA42 / MPSA92 is a very fast cascaded complementarysymmetrical differential stage.
The "upper" stages are lifted (+-)15 volts from gnd-level so the input transistors could be some low-voltage transistors or small-signal fets.
As MPSA42/92 are very fast, there is no real need to replace them with another types.
2. The next voltage amplification stage MPSA92 MJE15033 - and its complement are implemented a little differently.
Instead of a pure Darlington connection, the collector of the driving transistor is connected to GND via a resistor.
This implementation is because, according to the simulation, distortion increases with a pure Darlington connection.
The collector resistor is only necessary in a possible overdrive situation, when the base current of the power transistors is limited to an allowable level.
3. The emitterfollower stage controlling the FETs operates in class A within audiofrequencyrange at all powers.
4. The output stage is of a traditional style made with Exicon lateral FETs ECW20N20 and ECW20P20.
The quiescent current adjustment and thermal compensation are done here with a few parts and the adjustment keeps its set value in the range of 20 to 80 degrees centigrade.
The following values have been simulated with ideal voltage sources into an 8 ohm load.
The simulation has used values recommended in Cordell's book, e.g. max timestep which depends on the test frequency.
1kHz .maxstep=0.48831106u ... 20kHz .maxstep=0.02441555u
Frequency response 1.2Hz - 75kHz -1dB ... with 1uH coil.
Distortion levels simulated with optional gain A=37.4
Distortion 1kHz 1W/8ohm ..2nd -122dB ..3rd -130dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.000000%
Distortion 1kHz 4W/8ohm ..2nd -117dB ..3rd -120dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.000000%
Distortion 1kHz 16W/8ohm ..2nd -114dB ..3rd -113dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.000000%
Distortion 1kHz 64W/8ohm ..2nd-110dB ..3rd -109dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.000000%
Distortion 20kHz 1W/8ohm ..2nd -104dB ..3rd -110dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.000697%
Distortion 20kHz 4W/8ohm ..2nd -99dB ..3rd -96dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.002052%
Distortion 20kHz 16W/8ohm ..2nd -95dB ..3rd -88dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.004842%
Distortion 20kHz 64W/8ohm ..2nd -91dB ..3rd -86dB at quiescent current 187mA/fet ... Total Harmonic Distortion: 0.006681%
At about 200W / 8ohm power (1kHz) the third harmonic starts to raise its head, which means going worse side of the -100dB distortion level.
Noise analysis over the audio range 20-20k from the speaker output gives 65uV.
I like the hafler style lateral mosfet amps.
In the later hafler version, they used jfet instead of bjt as the input stage.
I would omit DC servo. Anything less than 20mV should not be a concern for a speaker.
In the later hafler version, they used jfet instead of bjt as the input stage.
I would omit DC servo. Anything less than 20mV should not be a concern for a speaker.
I suspect that the VAS pre driver will emit their magic smoke after first heavy overdrive occurence.
Not nonsense as such. If you parallel two inductances in such away that the two fields are in opposite direction they will cancel, and the measured induction will cancel, normally only partly.
But it probably is nonsense in the case you describe, for audio, with the level of inductances involved.
Jan
Last time I checked, Bob's circuits had excellent thermal stability. Maybe you overlooked something?
Jan
In a practical audio circuit, the loop length has no effect on the feedback working. What may happen with long loops is that it picks up parasitic capacitance and/or inductance, or that it couples to other wiring in the amp. That can effect amplifier stability, so for that it is a good idea to keep the loop tight.
BTW I like the way you connected the servo.
Jan
Coupling of the feedback to other wiring also introduces DISTORTION. The inverting input cannot distinguish between the desired signal through the feedback resistor and a signal out of the air. That signal out of the air can contain 30% distortion, if it comes from the power supply wiring. -60 dB of coupling is ridiculously easy to achieve. And you don’t want it. Any distortion or noise picked up here cannot be corrected, even with 80 dB of feedback. The main reason I just roll eyes when someone goes bragging about a simulation showing -120 dB distortion products. Or 90 degree phase margins, for that matter. Implementation matters, more than the actual circuit design, given today’s technology.
Feedback output point is to be carefully selected, symmetrically between Re resistors and identically to output speaker point. All the bypass capacitor returns must not pollute the signal ground, they have to have separate tracks. Then, excellent PSR and no mains hum components may be achieved. The length of the feedback track itself is unimportant.
Yes. Never forget that feedback corrects the signal at the point where it is connected.
This may or may not be the speaker output point. Here is where really good PCB layouts shine.
Jan
This may or may not be the speaker output point. Here is where really good PCB layouts shine.
Jan
I'm sure Bob's circuits have good thermal stability ... what I meant is there are many more components to keep in same temperature to keep stability.
Amplifier has been assembled few months ago and has been in everyday use since.
New pcb's were mounted in broken PS-audio delta 200 enclosure.
All the capacitors (1994) were replaced with new and higher capacitance ones.
(In this picture are the rc-filtered driverstage pwrsupplies which I replaced with FET-filtered supplies later)
On top of main transformer is 12V controlled soft-start circuit.
New pcb's were mounted in broken PS-audio delta 200 enclosure.
All the capacitors (1994) were replaced with new and higher capacitance ones.
(In this picture are the rc-filtered driverstage pwrsupplies which I replaced with FET-filtered supplies later)
On top of main transformer is 12V controlled soft-start circuit.
Not often one see Vbe multiplier after the VAS drivers Q3, Q15, imo, generally it should preferably be placed before the drivers, although lateral FET's are much more forgiving and usually don't need any particular Vgs and temp-co biasing circuit.
Potentially even more problematic due to use of dual LTP input stages which may, due to slightest mismatch between them, cause "fighting" currents between upper and lower side, which could put even more strain and demand on the following unusual driver-Vbe multiplier combo.
The circuit seem to be working so all is good, but from theoretically logic point of view a bit questionable choice.
Potentially even more problematic due to use of dual LTP input stages which may, due to slightest mismatch between them, cause "fighting" currents between upper and lower side, which could put even more strain and demand on the following unusual driver-Vbe multiplier combo.
The circuit seem to be working so all is good, but from theoretically logic point of view a bit questionable choice.
If one considers Q3 and Q15 as usual emitter follower stage biased so that it works in class A trough audiofrequencyrange it is not so wild choise.