In benchmarks on single transistors and in datasheets FETs
are faster and less powerconsuming.(eeeh .. generally for power transistors)
This may be only ignorance from my part, but I´ve read what
Douglas Self wrote about it in "Audio Amplifier Design Handbook"
and dont really agree.
Atleast my simulations tell me otherwise.
Otherwise in the real world?
So why do people not use FET output stages?, risk of self-oscillation perhaps?
/Dave
are faster and less powerconsuming.(eeeh .. generally for power transistors)
This may be only ignorance from my part, but I´ve read what
Douglas Self wrote about it in "Audio Amplifier Design Handbook"
and dont really agree.
Atleast my simulations tell me otherwise.
Otherwise in the real world?
So why do people not use FET output stages?, risk of self-oscillation perhaps?
/Dave
It's in the specific transistor, and how you use it...
So who says folks don't use FET's in output stages? Better not say that to NP, he'll wonder where you've been all these years... 😉
Self's simulations are corrects, as far as they go. Remember, in most cases he's been comparing a single set of bipolars vs a single set of FETs, typically the lateral FETs such as Hitachi's which were popular for audio, due to their ease of implementation.
Within any class of output device there's a lot of variation in performance and capability. This is true in bipolars, as well as FETs. You'll get much different results using sustained beta Toshiba or Sanken bipolars versus the older parts from Mot/On Semi - the extended bandwidth (10X Ft, typically 30-50 MHz) and extended gain linearity have a big impact, even before you start paralleling devices. These transistors also have relatively high HFE compared with older "audio" bipolars or conventional bipolar transistors for switching applications- typically Hfe over 100 or more; minimum of 60-80, depending on type.
With FETs, there's only really two choices- VDMOS, such as made by IR, Fairchild, and others, which are really designed for switching applications, have high gate threshold, and have a thermal bias stability point which is close to the rated forward current. Pass and others use these, it takes a bit of attention to the thermal compensation in the biasing network to have a stable amp, as the device tendency would be towards bias runaway even quicker than a bipolar.
The other choice is the lateral devices such as those now made by Semlab in England, which are built on the technological approach Hitachi pioneered. These have lower transconductance (hence the curves Self generates in his simulations, because emitter follower/source follower behavior is largely dependent on Gm), a relatively low threshold (doesn't take as much enhancement voltage), and a thermal bias stability operating point usually around 200 mA. They're designed by audio, and with the right implementation (lots of devices in parallel, for example), can work fairly nicely. An example of a commerical design using these parts is the Ayre V1x. I wouldn't doubt there are many others, I just don't know who.
While the choice of output transistor will have a decided impact on the driver and biasing arrangements required, it's my opinion very high levels of performance can be reached with either type. Perhaps you'll want to make your intial choices based on what you feel more comfortable with. Regarding the self oscillation issue, any wideband power device can be prone to this; even older Motorola power bipolars can exhibit this behavior if the right design practices aren't followed.
Having done all MOSFET, all bipolar, and mixed MOSFET with bipolar output stages, I'd say there's no reason to be dogmatic one way or the other...
😀
Just have fun, and listen to the music... let your ears be your guide.
~Jon
So who says folks don't use FET's in output stages? Better not say that to NP, he'll wonder where you've been all these years... 😉
Self's simulations are corrects, as far as they go. Remember, in most cases he's been comparing a single set of bipolars vs a single set of FETs, typically the lateral FETs such as Hitachi's which were popular for audio, due to their ease of implementation.
Within any class of output device there's a lot of variation in performance and capability. This is true in bipolars, as well as FETs. You'll get much different results using sustained beta Toshiba or Sanken bipolars versus the older parts from Mot/On Semi - the extended bandwidth (10X Ft, typically 30-50 MHz) and extended gain linearity have a big impact, even before you start paralleling devices. These transistors also have relatively high HFE compared with older "audio" bipolars or conventional bipolar transistors for switching applications- typically Hfe over 100 or more; minimum of 60-80, depending on type.
With FETs, there's only really two choices- VDMOS, such as made by IR, Fairchild, and others, which are really designed for switching applications, have high gate threshold, and have a thermal bias stability point which is close to the rated forward current. Pass and others use these, it takes a bit of attention to the thermal compensation in the biasing network to have a stable amp, as the device tendency would be towards bias runaway even quicker than a bipolar.
The other choice is the lateral devices such as those now made by Semlab in England, which are built on the technological approach Hitachi pioneered. These have lower transconductance (hence the curves Self generates in his simulations, because emitter follower/source follower behavior is largely dependent on Gm), a relatively low threshold (doesn't take as much enhancement voltage), and a thermal bias stability operating point usually around 200 mA. They're designed by audio, and with the right implementation (lots of devices in parallel, for example), can work fairly nicely. An example of a commerical design using these parts is the Ayre V1x. I wouldn't doubt there are many others, I just don't know who.
While the choice of output transistor will have a decided impact on the driver and biasing arrangements required, it's my opinion very high levels of performance can be reached with either type. Perhaps you'll want to make your intial choices based on what you feel more comfortable with. Regarding the self oscillation issue, any wideband power device can be prone to this; even older Motorola power bipolars can exhibit this behavior if the right design practices aren't followed.
Having done all MOSFET, all bipolar, and mixed MOSFET with bipolar output stages, I'd say there's no reason to be dogmatic one way or the other...
😀
Just have fun, and listen to the music... let your ears be your guide.
~Jon
Hello Jon,
I just wanted to make a very minor correction to your very thorough post. One time I did some measurements of the zero tempco operating point of the lateral MOFET output devices. I put them on a curve tracer and hit them with both a heat gun and some freeze spray and watched what happened.
The Hitachi parts weren't well matched. For the big plastic devices (2SJ160/2SK1058 family), the N-channel had a zero tempco around 100 mA, while the P-channel was around 75 mA. The smaller Hitachis (2SJ77/2SK213 family) had a similar mismatch, with the N-channel parts around 20 mA and the P-channels around 15 mA. The Semelab parts (BUZ900/BUZ905 family) were matched much better, with both polarities having a zero tempco around 120 mA.
Cheers,
Charles Hansen
I just wanted to make a very minor correction to your very thorough post. One time I did some measurements of the zero tempco operating point of the lateral MOFET output devices. I put them on a curve tracer and hit them with both a heat gun and some freeze spray and watched what happened.
The Hitachi parts weren't well matched. For the big plastic devices (2SJ160/2SK1058 family), the N-channel had a zero tempco around 100 mA, while the P-channel was around 75 mA. The smaller Hitachis (2SJ77/2SK213 family) had a similar mismatch, with the N-channel parts around 20 mA and the P-channels around 15 mA. The Semelab parts (BUZ900/BUZ905 family) were matched much better, with both polarities having a zero tempco around 120 mA.
Cheers,
Charles Hansen
Doug Self doesn't seem to like Mosfets. I speculate that he is
enamored of the work he has done getting high performance
at low bias currents with BJTs. Myself, I like Mosfets because
I can build simpler circuits with them.
As Charles points out, the vertical Mosfets have more thermal
drift at lower currents, and more distortion than lateral, but
they have other advantages, most specifically much higher
transconductance. I solve the problem by running them at high
bias currents and using large heat sinks. I do not thermally
compensate the bias in any of my products, that is to say I don't
mount bias transistors on the heat sinks or employ other tracking
devices.
If you want to build an amp that runs cool, the BJTs are a good
choice. If you want a Class A amp, vertical Mosfets are a good
choice. If you want something in between, you might try the
lateral Mosfets.
😎
Almost any choice of device will intoduce problems to be solved. Different types of devices, different types of problems. Another way of looking at it is that each problem/solution pair also results in a compromise. Different people have different things they are willing to compromise on.
In the case of D.Self, if you look at the words along the edge of the the distortion plots in his book you will see the words "Audio Precision" this means he owns or has access thousands of dollars of test equipment. He can measure tings and make adjustments at a much finer level than most of us who haven't much more than a DMM and some old stuff from ebay. For instance his favorite (appearently) BJT topology is CFB which will generally get you the lowest THD+N figures of any output topology *if* you can get the bias just right. FETS, on the otherhand may not get you the lowest measurement but your bias adjustment can be much further away from optimum and still have nearly minumum THD figures. When/if the BJT CFB bias drifts a little, the THD numbers may soon be than better than the FET.
In general, suposse its true that BJT ouput sections are "better" than FET sections. However, effective implementaion of a FET output stage may be be easier for DIYers with limited recources, with the result that *as implemented* by a typical DIYer the FET gives better results.
In the case of D.Self, if you look at the words along the edge of the the distortion plots in his book you will see the words "Audio Precision" this means he owns or has access thousands of dollars of test equipment. He can measure tings and make adjustments at a much finer level than most of us who haven't much more than a DMM and some old stuff from ebay. For instance his favorite (appearently) BJT topology is CFB which will generally get you the lowest THD+N figures of any output topology *if* you can get the bias just right. FETS, on the otherhand may not get you the lowest measurement but your bias adjustment can be much further away from optimum and still have nearly minumum THD figures. When/if the BJT CFB bias drifts a little, the THD numbers may soon be than better than the FET.
In general, suposse its true that BJT ouput sections are "better" than FET sections. However, effective implementaion of a FET output stage may be be easier for DIYers with limited recources, with the result that *as implemented* by a typical DIYer the FET gives better results.
- Status
- Not open for further replies.