Only recently I came across a still very interesting article about Cascode amp design written by Nelson Pass (see link). I saw that the article dates from 1978, so I am sure many off you know this article for some time (or years).
It’s plain from a technical point of view that the Cascode topology is favorable in many cases. But after searching this forum and reading some of the threads, the general opinion seems to be that Cascodes are liked in VAS stages but are disliked in Output Stages.
Technically, I don’t understand the reason for this. This means apparently that in Output Stages the disadvantages are greater than the advantages (besides the extra components). But what are these disadvantages? Why is the result of a Cascode Output Stage not what one would expect regarding the technical benefits.
I also wonder; did anyone perform a FAIR a-b comparison between a normal Output Stage and a Cascode equivalent? And which one sounded best? With fair I mean that in an amplifier only the Output Stage was modified between non Cascode and Cascode and the rest of the circuit and poer supply was left the same.
Peter
http://www.passlabs.com/pdf/articles/cascode.pdf
It’s plain from a technical point of view that the Cascode topology is favorable in many cases. But after searching this forum and reading some of the threads, the general opinion seems to be that Cascodes are liked in VAS stages but are disliked in Output Stages.
Technically, I don’t understand the reason for this. This means apparently that in Output Stages the disadvantages are greater than the advantages (besides the extra components). But what are these disadvantages? Why is the result of a Cascode Output Stage not what one would expect regarding the technical benefits.
I also wonder; did anyone perform a FAIR a-b comparison between a normal Output Stage and a Cascode equivalent? And which one sounded best? With fair I mean that in an amplifier only the Output Stage was modified between non Cascode and Cascode and the rest of the circuit and poer supply was left the same.
Peter
http://www.passlabs.com/pdf/articles/cascode.pdf
The main reason output cascodes were (are) used is to overcome SOA limitations of the output devices. This was a real concern 30 years ago when th e Leach was developed, or before that the Ampzilla (I think the Ampzilla was one of the earlier commercial implementations).
I really doubt the neccessity for a cascode output now days. There are great output devices with good SOA - much better than 30 years ago. The more common practice now is to parallel output devices to improve power handling.
Cascodes are used in VAS stages to reduce the effect of Miller capacitance- so the rationale in the VAS application is different to the output stage.
Hope this helps.
I really doubt the neccessity for a cascode output now days. There are great output devices with good SOA - much better than 30 years ago. The more common practice now is to parallel output devices to improve power handling.
Cascodes are used in VAS stages to reduce the effect of Miller capacitance- so the rationale in the VAS application is different to the output stage.
Hope this helps.
I would call the Leach output "Totem pole", not "Cascode"
in the Leach the series output devices simply split the V drop between them and both see (1/2 of) the varying V
Cascode operation reduces the V swing across one device by orders of magnitude
Cascode operation also "costs" more output V drop than the saturation V of the 2 devices - you need substantially more operating V than the saturation V for the cascoded transistor to show improved characteristics from cascode operation
in power amps paralleling the same transistors for higher output current/less β droop is often a bigger improvement and doesn't hurt the power output
The β22 is a active project amp that has a Cascode output:
http://www.amb.org/audio/beta22/
http://headwize.com/ubb/showpage.php?fnum=3&tid=6923
in the Leach the series output devices simply split the V drop between them and both see (1/2 of) the varying V
Cascode operation reduces the V swing across one device by orders of magnitude
Cascode operation also "costs" more output V drop than the saturation V of the 2 devices - you need substantially more operating V than the saturation V for the cascoded transistor to show improved characteristics from cascode operation
in power amps paralleling the same transistors for higher output current/less β droop is often a bigger improvement and doesn't hurt the power output
The β22 is a active project amp that has a Cascode output:
http://www.amb.org/audio/beta22/
http://headwize.com/ubb/showpage.php?fnum=3&tid=6923
Thanks for the interesting replies.
An important question remains: which sounds best? Cascode or non cascode output stage.
I never trust a design or a design idea until I’ve heard the results. I’ve build things where people raved about but were disappointed with the results afterwards. When a design (or an idea of myself) is interesting enough, I always start to build a test setup to see if I like it or not. And also to be able to tweak it until it’s performing at its max. Only after that a definite version is made.
Peter
An important question remains: which sounds best? Cascode or non cascode output stage.
I never trust a design or a design idea until I’ve heard the results. I’ve build things where people raved about but were disappointed with the results afterwards. When a design (or an idea of myself) is interesting enough, I always start to build a test setup to see if I like it or not. And also to be able to tweak it until it’s performing at its max. Only after that a definite version is made.
Peter
Cascode outputs not only help with heat dissipation in the output devices, but they reduce the effective capacitance that the front end has to drive. That's no small benefit, particularly if you're using MOSFETs in the output. Yes, if you're using a follower, the Cgs is pretty much out of the picture, but that's not the only capacitance you have to worry about.
As to pitfalls...well, there are cascodes and there are cascodes. If you look at the schematic for one of the old Threshold amps, you'll see that the cascode devices were run from a resistive divider. So what? Doesn't everyone? Well...
If I recall correctly (don't have that schematic in front of me at the moment, so I could be wrong), the non-rail end of that resistor divider was attached to the output, not ground. This makes perfect sense if your goal is to swing as much voltage as possible at the output because the cascode devices "get out of the way" when the output followers come barreling through. But as usual, there's a price to be paid. When the cascodes start swinging along with the followers (think of the bellows on an accordion) the operating characteristics as no longer as fixed as you might wish.
So is it the end for cascoded outputs?
Of course not.
If you're willing to give up some voltage swing, you can fix the resistors to ground instead of the output. If you want to take it a step further you can either bootstrap the Gate/base voltage reference or give that reference a small, dedicated supply at a higher voltage. Giving the circuit plenty of voltage at the rails is a good idea, as you'll have to set the cascodes comparatively far from the outputs in order to leave sufficient voltage swing. Yes, that could get intimidating, but this is DIY and the masthead says something about fanatics. Well, do you want to be a fanatic or don't you?
There are problems. Then there are solutions. Then there are more problems. Then there are more solutions. It's all a question of how badly you want to investigate the topology. It can be done--has been done, in fact--and some variant on the basic concept could very well be the Next Big Thing.
Grey
As to pitfalls...well, there are cascodes and there are cascodes. If you look at the schematic for one of the old Threshold amps, you'll see that the cascode devices were run from a resistive divider. So what? Doesn't everyone? Well...
If I recall correctly (don't have that schematic in front of me at the moment, so I could be wrong), the non-rail end of that resistor divider was attached to the output, not ground. This makes perfect sense if your goal is to swing as much voltage as possible at the output because the cascode devices "get out of the way" when the output followers come barreling through. But as usual, there's a price to be paid. When the cascodes start swinging along with the followers (think of the bellows on an accordion) the operating characteristics as no longer as fixed as you might wish.
So is it the end for cascoded outputs?
Of course not.
If you're willing to give up some voltage swing, you can fix the resistors to ground instead of the output. If you want to take it a step further you can either bootstrap the Gate/base voltage reference or give that reference a small, dedicated supply at a higher voltage. Giving the circuit plenty of voltage at the rails is a good idea, as you'll have to set the cascodes comparatively far from the outputs in order to leave sufficient voltage swing. Yes, that could get intimidating, but this is DIY and the masthead says something about fanatics. Well, do you want to be a fanatic or don't you?
There are problems. Then there are solutions. Then there are more problems. Then there are more solutions. It's all a question of how badly you want to investigate the topology. It can be done--has been done, in fact--and some variant on the basic concept could very well be the Next Big Thing.
Grey
If my memory serves me correctly the data presented by Nelson Pass in his paper referred to many SUBSTANTIAL benefits from cascoded outputs in the quality of sound and dynamics. There was very little discussion about using cascode to increases the rail voltage and hence amplifier output.
I am hoping to experiment with this idea when and if we get the Leach Superamp project worked out. As Grey suggested Dr. Leach did not use the same output design as Nelson Pass but maybe some of the benefits will be apparent on experimentation. Learning this firsthand is one of the unique and fun parts of DIY and having this forums wealth of knowledge to draw upon. Tad
I am hoping to experiment with this idea when and if we get the Leach Superamp project worked out. As Grey suggested Dr. Leach did not use the same output design as Nelson Pass but maybe some of the benefits will be apparent on experimentation. Learning this firsthand is one of the unique and fun parts of DIY and having this forums wealth of knowledge to draw upon. Tad
Grey,
The circuit in the link is an example of what you mention in your post.
http://www.audio-circuit.dk/Schematics/Threshold_Stasis_1.pdf
I was thinking: wouldn’t it be more accurate to drive the upper part of the Cascode from the INPUT instead of the OUTPUT. One of the main goals of this construction to keep the Vce of the output device constant. Because the voltage at the output has added distortion, time lag and a small voltage attenuation, wouldn’t it be better to drive both upper and lower parts of the Cascode from the SAME input voltage? It’s also very important that both parts of the Cascode (upper and lower) are identical. So if you use a darlington or triple darlington for the lower part, use the same darlington or triple darlington for the upper part.
A few years ago I was working on a tube Cascode stage and measured the how constant the voltage of the lower part of the Cascode was held. It was clear then that the best result were obtained when the upper part got it’s signal from the input.
Peter
The circuit in the link is an example of what you mention in your post.
http://www.audio-circuit.dk/Schematics/Threshold_Stasis_1.pdf
I was thinking: wouldn’t it be more accurate to drive the upper part of the Cascode from the INPUT instead of the OUTPUT. One of the main goals of this construction to keep the Vce of the output device constant. Because the voltage at the output has added distortion, time lag and a small voltage attenuation, wouldn’t it be better to drive both upper and lower parts of the Cascode from the SAME input voltage? It’s also very important that both parts of the Cascode (upper and lower) are identical. So if you use a darlington or triple darlington for the lower part, use the same darlington or triple darlington for the upper part.
A few years ago I was working on a tube Cascode stage and measured the how constant the voltage of the lower part of the Cascode was held. It was clear then that the best result were obtained when the upper part got it’s signal from the input.
Peter
I like Cascoded output stages because of the fast transient response i hear and yes i agree with Mr Pass completely that cascoding can make the transistor more Linear because of constant Vce plus the Advantage of substantially lower miller capacitance yes this quality is also impotent in Emitter Follower's. I have found Cascoded output stages to be as prone to instability as the more popular CFP type BTW is not the Stasis circuit come sort of gene splicing of a casscoded and a multi-Transistor CFP output Topology ????? attached is a Diamond Buffer cascoded type of low power line level output stage.
I came up with a similar bootstrapped diamond buffer variant about a year ago. See http://www.battletonphoenix.co.uk/buffer.pdf
PigletsDad said:
Yes you did! How was it's sound compared to conventional Diamond Buffer? I Amuse that with 4 volts on the Emitter's of Q2 & Q6 that the LED's you used were Blue? Have you tried different reference voltages? I found the 2 Blue LED's almost perfect.
BTW Cascoding also substantially reduces the amount of high frequency EMI polution on the AC Mains even better than two cascaded regulation stages.
pietjers said:
depends on which characteristics are most important
often the drive impedance of the BJT or even MOSFET output device is a very nonlinear load to the VAS - cascoding the output Q with bootstrapped Vref from the source/emitter gives higher input Z and less nonlinearity for the VAS to drive - SS outputs are quite different form tubes with their much higher gate Z
It is my understanding that there are 2 types of Totem pole output stage topologies used lot in the early 1970's. With the Dynaco Stereo 400 long with early SAE and Sumo products representing what is often referred to as a Bootstrapped configuration using Resistive voltage dividers along with a pair of Large Value Electrolytic capacitor supplying the AC drive and thus performing the Bootstrap operation, if we then remove the capacitor and replace the Resistive voltage divider with a Current source and Stable Voltage Reference then we have what is referred to as a cascode configuration with the Voltage reference floating along with the Output wave form as is the case with the resistive network however in the second case the stable voltage reference keeps the DC Vce of the lower cascode transistors constant, This type of cascoded output stage topology is what is shown in Mr. Pass's Technical white paper on the subject previously mentioned. Any way that is my understanding of the different configurations.
Peter,
in my view: cascoding originally was meant to extend the bandwidth in RF applications already in the tube era by decreasing the (sonically highly deteriorating) Miller capacitances in voltage amplification stages, but it provides significant improvements in numerous other ways and places, like the output stage. There is no justified reason to desist from using it.
The cascoding device should be applied in grounded base configuration with (by definition) zero signal voltage at its base* and keeping Vds or Vce constant by a stable low impedance reference voltage.
* The cascoding device should be bipolar or (discretely composed) Darlington.
in my view: cascoding originally was meant to extend the bandwidth in RF applications already in the tube era by decreasing the (sonically highly deteriorating) Miller capacitances in voltage amplification stages, but it provides significant improvements in numerous other ways and places, like the output stage. There is no justified reason to desist from using it.
The cascoding device should be applied in grounded base configuration with (by definition) zero signal voltage at its base* and keeping Vds or Vce constant by a stable low impedance reference voltage.
* The cascoding device should be bipolar or (discretely composed) Darlington.
Thanks again for the interesting replies.
It is a bit surprising that with al the technical benefits that cascode output stages have, almost all designs (still) are normal CFP or EF.
Or is the performance on paper better than in real life?
What are the sonic benefits? Has anyone experience in this area?
Peter
It is a bit surprising that with al the technical benefits that cascode output stages have, almost all designs (still) are normal CFP or EF.
Or is the performance on paper better than in real life?
What are the sonic benefits? Has anyone experience in this area?
Peter
Peter,
The most remarkable compound is the CFP, combining these would certainly result in unbeatable linearity in the output stage, where usually high distortion levels occur. Look out for oscillations!
It is not surprising, people prefer stereotype topologies. One may have considerable difficulties seeing why those amplifiers would sound better now than 30 years ago. Of course, they can not.
Briefly, a less distorted sound. It is due to reduced burden of the amplifying devices by limiting the operating range and placing them in their more linear region. Improved power supply ripple rejection, lower thermal related distortions and noise, the provided insulation are additional reasons, there are more. Cascoding often allows use of devices with better properties.
The most remarkable compound is the CFP, combining these would certainly result in unbeatable linearity in the output stage, where usually high distortion levels occur. Look out for oscillations!
Cascoding for the outputstage is not very practical in real life.
If you want best linearity, you use the transistor in a very small voltage range, thus small voltage swing. Consequently you have greatly reduced output power.
It's the same for the cascoded power jfet; how much power can you get from power jfets, cascoded to about 5 volts to keep dissipation down? C'mon!
If you enlarge the outputswing, you get reduced linearity. Next thing is that especially mosfets like high Drain-Source voltages for linearity, so usually it's better to drop the cascode and get the larger voltage swing.
But of course feel free to build and see for yourself!
By the way, read D. Selfs book, the outputstage does not produce so much distortion compared with the VAS or the input. While most here will not agree with his high-feedback attitude, he shows undebatable measurements.
Have fun, Hannes
If you want best linearity, you use the transistor in a very small voltage range, thus small voltage swing. Consequently you have greatly reduced output power.
It's the same for the cascoded power jfet; how much power can you get from power jfets, cascoded to about 5 volts to keep dissipation down? C'mon!
If you enlarge the outputswing, you get reduced linearity. Next thing is that especially mosfets like high Drain-Source voltages for linearity, so usually it's better to drop the cascode and get the larger voltage swing.
But of course feel free to build and see for yourself!
By the way, read D. Selfs book, the outputstage does not produce so much distortion compared with the VAS or the input. While most here will not agree with his high-feedback attitude, he shows undebatable measurements.
Have fun, Hannes
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.