Hi
Forgive me the indulgence but I've posted another modified version of
your design, I know you said you don't like
This version incorporates all the cascoding and will be horribly inefficient
but uses the dual voltage supply to give a larger linear output swing.
My back of the envelope calculations give about 160W dissipation
for I think a 4W output. This assumes that differentially we have +/- 1A
peak current or +/- 8v peak differentially into 8 ohms.
If we want to hit 10W peak output this would imply a 1.6amp peak current
and an overall dissipation of roughly 250W. Is this correct ?
The schematics indicate that the lower cascodes are optional. However
how optional they are will depend on a couple of things.
Firstly my feeling is they make sense from a symmetry perspective: the +ve and
-ve signal swings should behave in roughly the same way as the output signal level
is increased.
Secondly their use will depend on the current you want to run the amp at.
I've tried to keep in mind your 20W power dissipation requirement for each
transistor.
This posted version is supposed to run at a tail current of 2 amps.
Hence the cascode transistor M21 on M12 which will share the power dissipation
may not be strictly neccessary but should help with PSRR and give a "stiffer"
current soure. At this current level the power dissipation of
M12 is of the order of 8W. M21 will dissipate about 16W but will depend on
the degeneration resistance you select, I've dropped your 6 volt value here to
3, increasing our gain 6dB. You could double all the currents in the amp to get
back to the 6 volt value while maintaining the gain improvement, but you will
have to be very careful about the power dissipation of all the transitors.
The input transistors M7 and M8 are running at 1 amps but now with nearly
20v Vds, if the input gets hard driven we could potentially have 40W here,
in your original design the Vds of these devices reduces with increasing current
no longer the case here where the Vds is essentially fixed.
We could use 2 devices (or more !) in parallel for each input thereby dividing the
dissipation in half (or more). Actually using 4 (or more) input devices
here will open the door to all sorts of interesting linearising experiments.
M10 and M11 dissipate about 10W, they should have 2 amps through them.
M13, M15, M19 and M20 will have input signal related power dissipation. With the
amp hard switched one way or the other their max current is 2 amps but as in the
original design more current comes with reduced Vds and vice versa, statically the
devices dissipate 15W (since the common mode feedback is set to Ov at the output)
with a maximum dissipation of 20W.
M17 and M18 have a static dissipation of 10W.
I've used zeners to bias the cascodes but have some doubts as to their noise
performance hence the decoupling caps. In theory noise on the gates should
be rejected differentially hence they may be unnecessary, again this is the
thinking behind using a common bias for the cascode devices.
PSRR should be reasonably good except for the tail current source which
looks suspect to me from signal pick up off the negative supply, hence again
the decoupling cap.
There are much better ways of achieving the common mode feedback which will
ensure better thermal stability and supply rejection but at the expense of some
more complexity and components.
All the errors are mine and mine alone
I suppose I ought to build this thing, but I don't own speakers that would
benefit from it, maybe a trip to eBay is in order
herisson
I almost forgot to post the circuit again
Forgive me the indulgence but I've posted another modified version of
your design, I know you said you don't like
so apologies in advance.
This version incorporates all the cascoding and will be horribly inefficient
but uses the dual voltage supply to give a larger linear output swing.
My back of the envelope calculations give about 160W dissipation
for I think a 4W output. This assumes that differentially we have +/- 1A
peak current or +/- 8v peak differentially into 8 ohms.
If we want to hit 10W peak output this would imply a 1.6amp peak current
and an overall dissipation of roughly 250W. Is this correct ?
The schematics indicate that the lower cascodes are optional. However
how optional they are will depend on a couple of things.
Firstly my feeling is they make sense from a symmetry perspective: the +ve and
-ve signal swings should behave in roughly the same way as the output signal level
is increased.
Secondly their use will depend on the current you want to run the amp at.
I've tried to keep in mind your 20W power dissipation requirement for each
transistor.
This posted version is supposed to run at a tail current of 2 amps.
Hence the cascode transistor M21 on M12 which will share the power dissipation
may not be strictly neccessary but should help with PSRR and give a "stiffer"
current soure. At this current level the power dissipation of
M12 is of the order of 8W. M21 will dissipate about 16W but will depend on
the degeneration resistance you select, I've dropped your 6 volt value here to
3, increasing our gain 6dB. You could double all the currents in the amp to get
back to the 6 volt value while maintaining the gain improvement, but you will
have to be very careful about the power dissipation of all the transitors.
The input transistors M7 and M8 are running at 1 amps but now with nearly
20v Vds, if the input gets hard driven we could potentially have 40W here,
in your original design the Vds of these devices reduces with increasing current
no longer the case here where the Vds is essentially fixed.
We could use 2 devices (or more !) in parallel for each input thereby dividing the
dissipation in half (or more). Actually using 4 (or more) input devices
here will open the door to all sorts of interesting linearising experiments.
M10 and M11 dissipate about 10W, they should have 2 amps through them.
M13, M15, M19 and M20 will have input signal related power dissipation. With the
amp hard switched one way or the other their max current is 2 amps but as in the
original design more current comes with reduced Vds and vice versa, statically the
devices dissipate 15W (since the common mode feedback is set to Ov at the output)
with a maximum dissipation of 20W.
M17 and M18 have a static dissipation of 10W.
I've used zeners to bias the cascodes but have some doubts as to their noise
performance hence the decoupling caps. In theory noise on the gates should
be rejected differentially hence they may be unnecessary, again this is the
thinking behind using a common bias for the cascode devices.
PSRR should be reasonably good except for the tail current source which
looks suspect to me from signal pick up off the negative supply, hence again
the decoupling cap.
There are much better ways of achieving the common mode feedback which will
ensure better thermal stability and supply rejection but at the expense of some
more complexity and components.
All the errors are mine and mine alone
I suppose I ought to build this thing, but I don't own speakers that would
benefit from it, maybe a trip to eBay is in order
herisson
I almost forgot to post the circuit again
though i've now
Hi herisson,
Thank you for the updated schematic. I did not mean to criticize you by my remarks. I was only trying to give some insight into my own personal point of view.
Regarding your schematic, I don't know where to start. I have only looked at it briefly, but I believe it won't work. I can't figure out whether you meant this to be a single stage SOZ with cascodes on all the transistors or a two stage design of some type so I can't really comment further. I would have two recommendations:
1) Search this site for examples of Zen amps and cascoding.
2) As you say - try building a prototype and try it out. If you can't do that then try your design on a simulator. SIMetrix seems to be the popular one around here.
Regards,
Graeme
Thank you for the updated schematic. I did not mean to criticize you by my remarks. I was only trying to give some insight into my own personal point of view.
Regarding your schematic, I don't know where to start. I have only looked at it briefly, but I believe it won't work. I can't figure out whether you meant this to be a single stage SOZ with cascodes on all the transistors or a two stage design of some type so I can't really comment further. I would have two recommendations:
1) Search this site for examples of Zen amps and cascoding.
2) As you say - try building a prototype and try it out. If you can't do that then try your design on a simulator. SIMetrix seems to be the popular one around here.
Regards,
Graeme
Very cool indeed. I certainly wasn't looking for that. As I stated I wasn't really wasn't sure what herisson was trying to do. In his comments it appeared he was simply trying to add cascodes to most or all of the transistors in an F1 like design. He doesn't give any indication in the text that he was creating an embodiment of #5,376,899. Forests and trees.
Friend herisson, it appears that you have produced something very cool. You now have no choice. Start building.
Friend herisson, it appears that you have produced something very cool. You now have no choice. Start building.
Hi Graeme, Mr. Pass,
This is all I've done Well not quite, I started off
doing this but felt the cascodes ate up too much output
signal swing. The folded cascode seemed a more logical
extension since your design already has the dual power
supplies. At the cost of efficiency of course.
The is still you guy's design, a transconductance amp.
for driving loudspeakers, it's just a dc coupled alternative.
I've tried to keep as much as possible Graeme's building
blocks, re-used the common mode feedback from the zen 7 and
the cascode biasing using the zeners since I figured people
on the thread would be more familiar with these circuits.
the voltage feedback. I'm not claiming to have invented
anything here, just trying to see how far this transconductance
amp idea thing can go.
I've no idea what the X600 front end looks like so I can't comment,
though maybe my comments above may explain the resemblance if any.
I clearly have no other alternative other than to ruin myself
financially buying exotic speakers and destroy my marriage
by spending even later nights with a hot soldering iron trying
to figure out where that bizzare 10MHz oscillation is coming
from.
And it's all my own fault
herisson
This is all I've done
Well not quite, I started offdoing this but felt the cascodes ate up too much output
signal swing. The folded cascode seemed a more logical
extension since your design already has the dual power
supplies. At the cost of efficiency of course.
The is still you guy's design, a transconductance amp.
for driving loudspeakers, it's just a dc coupled alternative.
I've tried to keep as much as possible Graeme's building
blocks, re-used the common mode feedback from the zen 7 and
the cascode biasing using the zeners since I figured people
on the thread would be more familiar with these circuits.
I agree the gl5 topology does resemble that circuit minus
the voltage feedback. I'm not claiming to have invented
anything here, just trying to see how far this transconductance
amp idea thing can go.
I've no idea what the X600 front end looks like so I can't comment,
though maybe my comments above may explain the resemblance if any.
I clearly have no other alternative other than to ruin myself
financially buying exotic speakers and destroy my marriage
by spending even later nights with a hot soldering iron trying
to figure out where that bizzare 10MHz oscillation is coming
from.
And it's all my own fault
herisson
Hi herisson,
Well regardless of what you intended to do you've still done something very significant.
Your design accomplishes overall feedback by using SuSy as per #5,376,899 and like the SOZ it retains an arbitrarily high input impedance and as you state, it remains DC coupled. There is no direct connection from the output node to the input to implement the feedback. I have wrestled with this last issue. You have made me expand my thinking and I thank you.
I don't have the experience to help you with the potential stability problem. Hopefully some other members will step in to help.
The Fostex FE166E drivers cost $125.00US a pair here - plus the cost to build a pair of basic enclosures. Your wife shouldn't object too much to that.
Graeme
Well regardless of what you intended to do you've still done something very significant.
Your design accomplishes overall feedback by using SuSy as per #5,376,899 and like the SOZ it retains an arbitrarily high input impedance and as you state, it remains DC coupled. There is no direct connection from the output node to the input to implement the feedback. I have wrestled with this last issue. You have made me expand my thinking and I thank you.
I don't have the experience to help you with the potential stability problem. Hopefully some other members will step in to help.
The Fostex FE166E drivers cost $125.00US a pair here - plus the cost to build a pair of basic enclosures. Your wife shouldn't object too much to that.
Graeme
Hi Graeme
I was sitting here trying to work out the heatsinking requirements
for this and how to expain to my wife that
we'll see
Well I'm not Mr. Pass but I'll try my best to explain.
In the 5,376,899 Fig. 1, the resistors R37, R23 and R36, R22 form
2 inverting voltage feedback networks. They essentially set the
voltage gain. Let me back-up a bit here.
Actually almost all solid-state "amplifying" devices
can be considered as transconductors, voltage in current out
(Ok, so a BJT has base current but this can be considered as
a product of it's essentially transconductance action and yes you
can drive them with a current source ...) and they are all modelled
as such. Voltage gain is developed by driving the current through some
impedance, in this case a complex one, a loudspeaker coil. In the case
of the gl5 and the F1 and you original circuit the output current will
be a copy of the input voltage multiplied by the circuits transconductance
gm. Please forgive me if this is already obvious, I'm not trying to be
condescending in any way. The output voltage however for a transconductance
amp will do what it's told to do by the load impedance (like me except it's
my wife telling me what to do).
Now, if we wrap some form of shunt feedback
(essentially the output voltage fed back to the input)
around the transconductance amp the amp is no longer a transconductor.
The voltage has been freed from it's impedance dependance, but it now
must follow the input faithfully I'm getting carried away with the
wife analogies here I'm sorry. Our transconductance amp is no longer
but has become a plain voltage amp. And that is the difference between
gl5 and Mr. Pass's circuit 5,376,899. (Oh, Before anyone misunderstands
me I'm not implying the 5,376,899 is a "plain voltage amp")
Because of the symmetrical nature of things we can also take a voltage
amp and apply series feedback (current feedback BTW Mr. Pass was refering
to a current sensing resistor technique which achieves this in an earlier post)
we can turn our voltage amp back into a transconductance amp. Now the output
current is driven by the input voltage. Notice I used the word back since
any voltage amp was originally a transconductance device forced by us to
produce a voltage proportional to its voltage input.
It certainly doesn't explain why the current feedback doesn't "sound" good,
but this double transconductance -> voltage -> transconductance transformation
lacks the elegance of using a transistor to do what it does best:
kick out a current for a given voltage input.
Again I apologise in advance if this was already obvious.
Actually until I started looking at this my thinking was that the
folded cascode was a bit of a "luxury item" for an audio amp, generally
we use them for circuits with really low supply voltages but which
require high open loop gain, in your circuit however it seems to make sense.
The other nice thing about this implementation is that we fix the common
mode voltage at the output to ground, this is a great reference voltage
to use, it doesn't drift with temperature and in places shouldn't be too
noisy either.
I guess while we're talking about this stuff
If the Amp had some form of global negative feedback around it, which end of the series resistance was it connected to ?
herisson
I was sitting here trying to work out the heatsinking requirements
for this and how to expain to my wife that
we'll see
Well I'm not Mr. Pass but I'll try my best to explain.
In the 5,376,899 Fig. 1, the resistors R37, R23 and R36, R22 form
2 inverting voltage feedback networks. They essentially set the
voltage gain. Let me back-up a bit here.
Actually almost all solid-state "amplifying" devices
can be considered as transconductors, voltage in current out
(Ok, so a BJT has base current but this can be considered as
a product of it's essentially transconductance action and yes you
can drive them with a current source ...) and they are all modelled
as such. Voltage gain is developed by driving the current through some
impedance, in this case a complex one, a loudspeaker coil. In the case
of the gl5 and the F1 and you original circuit the output current will
be a copy of the input voltage multiplied by the circuits transconductance
gm. Please forgive me if this is already obvious, I'm not trying to be
condescending in any way. The output voltage however for a transconductance
amp will do what it's told to do by the load impedance (like me except it's
my wife telling me what to do).
Now, if we wrap some form of shunt feedback
(essentially the output voltage fed back to the input)
around the transconductance amp the amp is no longer a transconductor.
The voltage has been freed from it's impedance dependance, but it now
must follow the input faithfully
I'm getting carried away with thewife analogies here I'm sorry. Our transconductance amp is no longer
but has become a plain voltage amp. And that is the difference between
gl5 and Mr. Pass's circuit 5,376,899. (Oh, Before anyone misunderstands
me I'm not implying the 5,376,899 is a "plain voltage amp")
Because of the symmetrical nature of things we can also take a voltage
amp and apply series feedback (current feedback BTW Mr. Pass was refering
to a current sensing resistor technique which achieves this in an earlier post)
we can turn our voltage amp back into a transconductance amp. Now the output
current is driven by the input voltage. Notice I used the word back since
any voltage amp was originally a transconductance device forced by us to
produce a voltage proportional to its voltage input.
It certainly doesn't explain why the current feedback doesn't "sound" good,
but this double transconductance -> voltage -> transconductance transformation
lacks the elegance of using a transistor to do what it does best:
kick out a current for a given voltage input.
Again I apologise in advance if this was already obvious.
Actually until I started looking at this my thinking was that the
folded cascode was a bit of a "luxury item" for an audio amp, generally
we use them for circuits with really low supply voltages but which
require high open loop gain, in your circuit however it seems to make sense.
The other nice thing about this implementation is that we fix the common
mode voltage at the output to ground, this is a great reference voltage
to use, it doesn't drift with temperature and in places shouldn't be too
noisy either.
I guess while we're talking about this stuff
If the Amp had some form of global negative feedback around it, which end of the series resistance was it connected to ?
herisson
Hi herisson,
Thank you for your discussion on transconductance in devices and amplifiers. I think I get your meaning.
I am still not certain what topology you wish to implement in your gl5 circuit. Do you want it to be Super Symmetrical or are you wanting to build a balanced folded cascode with or without global feedback?
I am thinking that it would be interesting to build the circuit in Fig. 1 of patent #5,376,899 as shown but scale it up to the current levels we are talking about. It sort of looks a like a transconductance amp already with transistors 30 and 31 each operating into a CCS. The question in my mind is do resistors 36 and 37 turn the whole circuit back into a voltage amplifier. I think so. What about connecting resistor 36 to the drain of transistor 20 instead of 30 and similarly on the other side. Would this be a voltage amp with SuSy feedback (like the Zen V7) driving a voltage to current convertor? Items 26, 27 40, 42, and 43 could change to look more like the V7 circuit. And there are probably some similar input biasing changes. It is interesting to consider that this would probably still qualify as a balanced single stage amplifier. I need to think about this some more.
Regards,
Graeme
Thank you for your discussion on transconductance in devices and amplifiers. I think I get your meaning.
I am still not certain what topology you wish to implement in your gl5 circuit. Do you want it to be Super Symmetrical or are you wanting to build a balanced folded cascode with or without global feedback?
I am thinking that it would be interesting to build the circuit in Fig. 1 of patent #5,376,899 as shown but scale it up to the current levels we are talking about. It sort of looks a like a transconductance amp already with transistors 30 and 31 each operating into a CCS. The question in my mind is do resistors 36 and 37 turn the whole circuit back into a voltage amplifier. I think so. What about connecting resistor 36 to the drain of transistor 20 instead of 30 and similarly on the other side. Would this be a voltage amp with SuSy feedback (like the Zen V7) driving a voltage to current convertor? Items 26, 27 40, 42, and 43 could change to look more like the V7 circuit. And there are probably some similar input biasing changes. It is interesting to consider that this would probably still qualify as a balanced single stage amplifier. I need to think about this some more.
Regards,
Graeme
Well ,if you want to put a label on it the gl5 circuit
is a differential (or balanced) transconductance amplifier
using folded cascodes. The circuit has NO global voltage
feedback. If it did it would no longer be a transconductance
amplifier but a voltage amp.
Yes.
No, not quite, because the cascode devices 30 and 31 force a
fixed dc voltage on nodes 28 and 29. The modification you suggest
will modify the dc operating point of the circuit. The dc voltage
on the gates of the the input devices will be fixed by the ratio
of resistors 37 23 and 36 22 and the dc voltage on nodes 28 and 29.
You will now need some form of common mode feedback scheme to fix
the output dc voltage as per the gl5.
The voltage on 28 and 29 is not quite dc only, there will be some
signal related waveform due to the signal current modulating the
Vgs of 30 and 31, this will be non-linear and fedback to your inputs.
The signal is small and generally the feedback ratio high but will
still degrade the overall linearity of the circuit.
Yes.
herisson
Hi Herisson,
You're right. I am not as familiar with folded cascodes as I am with the straight type. Again I was hasty and didn't think long enough.
I re-read Hawksfords paper on transconductance amplifier design over the weekend. I think that the concept of a voltage amplifier to provide and set gain followed by an open loop unity current buffer is attractive. However, the complexity of his design is overwhelming.
Let me try another idea on you for a transconductance X amplifier.
Using Fig. 1 of #5,376,899 as a starting point, add a new transistor in parallel with 30 connecting its gate to 32 and its source to 28. Its drain would become "-OUT Current" and would have a new, additional CCS to V-. I'm assuming that we would have 1A of current through 20, 30 and the new transitor. 24 would have to increase to 3A. The same would be done to the other side. I'm assuming appropriate matching of components. We would now have a functional SuSy voltage amplifier that would provide and set the voltage gain. The two new output nodes would provide a mirrored current output. The level shifting common gate transistors have unity current gain and their non-linearities would be cancelled by the SuSy action on 30 and 31. Again we would retain DC coupled inputs and a reasonable input impedance.
And yes, the efficiency in terms of power consumption and number of transistors used is terrible. But we're talking about a hobby project not a commercial product.
Mr. Pass, thank you for the new article.
Regards,
Graeme
You're right. I am not as familiar with folded cascodes as I am with the straight type. Again I was hasty and didn't think long enough.
I re-read Hawksfords paper on transconductance amplifier design over the weekend. I think that the concept of a voltage amplifier to provide and set gain followed by an open loop unity current buffer is attractive. However, the complexity of his design is overwhelming.
Let me try another idea on you for a transconductance X amplifier.
Using Fig. 1 of #5,376,899 as a starting point, add a new transistor in parallel with 30 connecting its gate to 32 and its source to 28. Its drain would become "-OUT Current" and would have a new, additional CCS to V-. I'm assuming that we would have 1A of current through 20, 30 and the new transitor. 24 would have to increase to 3A. The same would be done to the other side. I'm assuming appropriate matching of components. We would now have a functional SuSy voltage amplifier that would provide and set the voltage gain. The two new output nodes would provide a mirrored current output. The level shifting common gate transistors have unity current gain and their non-linearities would be cancelled by the SuSy action on 30 and 31. Again we would retain DC coupled inputs and a reasonable input impedance.
And yes, the efficiency in terms of power consumption and number of transistors used is terrible. But we're talking about a hobby project not a commercial product.
Mr. Pass, thank you for the new article.
Regards,
Graeme
Hi Graeme,
I like your thinking here, the idea is interesting, and I can see how you are
attempting to implement Dr. Hawksford's idea. Actually your idea seems to resemble a
current feedback op-amp architecture (as does the circuit in Dr. Hawksfords paper)
Actually, I probably shouldn't have posted, I need to think about this a bit
more.
I was thinking that this topology doesn't seem to lend itself well to applying
current feedback. (which from an efficiency point of view would seem to be
the case).
Nelson Pass's article has probably saved my marriage, since I can change all
the crossover networks and rewire my 'speakers over a weekend and no-one will
be any the wiser.
Great stuff.
herisson
Hi herisson,
I have done some further thinking and I realize that you probably need to double the current in the input diff pair as well because 24 is a CCS and there is no current gain. The current supplied by 24, 25, 26, and 27 would also need to increase. And the effiency would continue to drop!
Perhaps a better idea would would be to use Fig. 2 of #5,376,899 as a starting point. At least there you have complementary symmetry as well and would regain a significant amount of efficiency.
Regards,
Graeme
I have done some further thinking and I realize that you probably need to double the current in the input diff pair as well because 24 is a CCS and there is no current gain. The current supplied by 24, 25, 26, and 27 would also need to increase. And the effiency would continue to drop!
Perhaps a better idea would would be to use Fig. 2 of #5,376,899 as a starting point. At least there you have complementary symmetry as well and would regain a significant amount of efficiency.
Regards,
Graeme
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.