I've been playing around with amp designs for a while now, messing about with various implementations. I thought it would be nice to gather all the common output stages together comparitively in terms of output swing. So here they are:
The same schematic was used for all stages. Output stage quiescent current was adjusted in each case to be about 22mA. Voltage rails were 33.5V and no 'emitter' resistors were used. Load driven was 8 ohms resistive.
These results say to me that there is no major winner in output stage efficiency, despite what some people may claim. Driver 'emitter' resistors were 150R and I was able to reduce swing by increasing this resistance, so this is clearly a factor to consider when designing output stages.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
The same schematic was used for all stages. Output stage quiescent current was adjusted in each case to be about 22mA. Voltage rails were 33.5V and no 'emitter' resistors were used. Load driven was 8 ohms resistive.
These results say to me that there is no major winner in output stage efficiency, despite what some people may claim. Driver 'emitter' resistors were 150R and I was able to reduce swing by increasing this resistance, so this is clearly a factor to consider when designing output stages.
Hi,
I think you have chosen a set of output conditions that do not exaggerate the differences between the devices.
But, I can see some differences in the plots.
Look at the length of the flat top/bottom for each type.
By far the longest is the Nch FET and you did not show Pch.
It would appear from this scant information that the FET would have a lower output than any of the others if you reduced the voltage to just at or below clipping.
If you reduced the load impedance, you may find that the other devices will also show some difference.
But try to find a better way to plot your information and the differences may jump out at you. eg. the two diode drop of the Darlington should stand out.
I think you have chosen a set of output conditions that do not exaggerate the differences between the devices.
But, I can see some differences in the plots.
Look at the length of the flat top/bottom for each type.
By far the longest is the Nch FET and you did not show Pch.
It would appear from this scant information that the FET would have a lower output than any of the others if you reduced the voltage to just at or below clipping.
If you reduced the load impedance, you may find that the other devices will also show some difference.
But try to find a better way to plot your information and the differences may jump out at you. eg. the two diode drop of the Darlington should stand out.
Andrew, I should have been a little clearer about the information on the plots. The clipping points are measured by markers and if you look at the bottom of the plots the cursor points are listed.
I was fully aware that the n-channel vertical MOSFET was the least efficient of the bunch. My point was that there is barely a volt between the best and worst performer, at 8 ohms load.
I did not show a pure p-channel vertical MOSFET output stage because nobody uses them.
I was fully aware that the n-channel vertical MOSFET was the least efficient of the bunch. My point was that there is barely a volt between the best and worst performer, at 8 ohms load.
I did not show a pure p-channel vertical MOSFET output stage because nobody uses them.
Googling I found some CFP a.k.a. "Sziklai pair" figures from Douglas Self site:
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
AndrewT said:
No you have misinterpreted things. The plots show the output node when driving an 8 ohm load, i.e. the available output swing. The output stages are built as follows:
Plot 1: complementary Darlington, i.e. NPN upper and PNP lower
Plot 2: complementary Sziklai (also known as complementary feedback pair, CFP), i.e. PNP upper and NPN lower
Plot 3: pure NPN, i.e. Darlington upper and Sziklai lower
Plot 4: pure N V-MOSFET, i.e. as Plot 3 but NPN replaced with N V-MOSFET.
Hope that clears things up.
My comment on the test is that an output stage circuit should be designed around the chosen audio devices. If so you can have the same efficiency out of almost any type of power semiconductor.
You would probably never connect different devices in a similar circuit in practical use.
It's like saying, which motor is best a car engine or a boat engine, by putting both in a car.
My guess is that the car engine will win, simply because it's doing what it's designed to do.
So the test doesn't really show anything.
You would probably never connect different devices in a similar circuit in practical use.
It's like saying, which motor is best a car engine or a boat engine, by putting both in a car.
My guess is that the car engine will win, simply because it's doing what it's designed to do.
So the test doesn't really show anything.
Depending on V-MOSFET Rds-on, you may get rail-to-rail output with less than 1V drop if the gates are driven in such a way that the output won't be limited by Vgs. Replacing a bipolar device with a MOSFET is not an optimum solution due to the Vgs drop.
Lars Clausen said:
On the contrary, I would argue that the test shows exactly what I intended it to; maximum output swing using different OPS topologies/devices 🙂
Eva said:
The driving circuit used a bootstrapped VAS so that is pretty much giving a maximised drive to the OPS (hence addressing the Vgs issue) in fact, as best it can be without going to fully separate rails for the OPS and other stages. The MOSFET used was IRF140, which has negligible Rds_on at 0.077 ohm.
Sziklai became more & more popular about ten years ago, when these fast japanese ring emitter BJTs flooded the audio world.
Also several Elektor amps use it.
Szklai should be definitely the circuit, which can swing most close to the DC rail. I guess the advantage would be stronger in an implementation with emitter resitors.
But honestly speaking, who cares about the last 3V ?
It's just 0.x db difference in unclipped output level.
I would be much more interested in achievable distorsion levels
at medium and small output power.
Say in the range between 0.1V peak...2V peak.
People say that the Sziklai is fortunate for distorsions, because it offers an very fast single stage feedback.
I never simulated or performed measurement in this regard, but I would definitely be interested in such an evaluation.
Also several Elektor amps use it.
Szklai should be definitely the circuit, which can swing most close to the DC rail. I guess the advantage would be stronger in an implementation with emitter resitors.
But honestly speaking, who cares about the last 3V ?
It's just 0.x db difference in unclipped output level.
I would be much more interested in achievable distorsion levels
at medium and small output power.
Say in the range between 0.1V peak...2V peak.
People say that the Sziklai is fortunate for distorsions, because it offers an very fast single stage feedback.
I never simulated or performed measurement in this regard, but I would definitely be interested in such an evaluation.
In fact my first The End amplifiers from 1989 used this output circuit. It has very good merits of sound quality.
Fast and snappy bass, and also a solid and smooth top.
BUT the backside is, that the output transistors are hard to turn off quickly enough. In case the amplifier goes into any kind of oscillation, or a high frequency is attempted, you risk cross conduction and burnout. That's why i stopped using the Sziklai circuit in 1992. A normal darlington circuit does not necessarily have the same drawbacks.
Fast and snappy bass, and also a solid and smooth top.
BUT the backside is, that the output transistors are hard to turn off quickly enough. In case the amplifier goes into any kind of oscillation, or a high frequency is attempted, you risk cross conduction and burnout. That's why i stopped using the Sziklai circuit in 1992. A normal darlington circuit does not necessarily have the same drawbacks.
Eva said:
ahhh, yes. but are such drive circuits common?
Most VAS cannot go higher than 1V....5V below the supply rail.
And the Sziklai in such an application will have an output swing that is just 0.7V less than the VAS output.
Richie, which results do you get if we assume the same rails as you did before, but VAS output of 30V peak?
Lars Clausen said:
Sound quite reasonable. Which BJTs did you use?
From my perception the Sziklai is crying for these fast 20MHz...30MHz beasts. Then you should have a good chance to fight oscillations without slowing it down to unacceptable levels.
ChocoHolic said:
You can build a sziklai output with some gain, let's say 20%, then you get rail to rail output (minus the Vce-sat of outputdevices). You just need to be careful with clipping behaviour. This is especially interesting when using mosfets in low supply voltage amps.
Mike
The difference between CFP/Sziklai and darlington is mostly cosmetic. That's because the global current gain is the same in both circuits, thus the VAS current required to drive them is also the same. Well, actually the CFP has a nasty current gain discontinuity around zero-crossing that the darilington does not have. The advantages of the CFP are bias setting without thermal feedback and less "voltage distortion" (at the expense of more "current distortion").
I'm using CFP with MOSFETs because I have two gain stages before the gates and the magnitude of the drive current discontinuity seen by the VAS is less than 20uA, but I wouldn't use it with bipolars.
I'm using CFP with MOSFETs because I have two gain stages before the gates and the magnitude of the drive current discontinuity seen by the VAS is less than 20uA, but I wouldn't use it with bipolars.
Eva said:
...hm, you mean that it will depend on the output impedance of the used VAS, if CFP or Darligton will show less distorsion?
Or on the speaker load? Or on both?
... I hate when things are getting complicated...
😉
Yes, the behaviour is drive-impedance dependent 😀
Load impedance is not as relevant because it does more or less the same nasty things to both topologies. In the darlington there are two Vbe junctions showing non-linear voltage drop as Ic is increased and Vce is reduced, and in the CFP there is only one, but behaving exactly in the same non-linear way. Also, current gain non-linearities as load current increases are the same in both topologies.
Load impedance is not as relevant because it does more or less the same nasty things to both topologies. In the darlington there are two Vbe junctions showing non-linear voltage drop as Ic is increased and Vce is reduced, and in the CFP there is only one, but behaving exactly in the same non-linear way. Also, current gain non-linearities as load current increases are the same in both topologies.
- Status
- Not open for further replies.