BJTs vs Mosfets
Hello guys,
We have built amplifiers with both types of output devices. My opinion is that BJTs sound a lot nicer than Mosfets and that the fact that some Mosfets have massive SOA ratings this does not tell the complete story.
Mosfets need lots of idling current to lower their THD to an acceptable level and they also need thermal compensation.
Parallel BJTs out perform any Mosfet circuit.
Class G/H is nice for improved efficiency but do not sound very good. Why? Well when the inner devices are saturated and the outer devices are dumping the higher supply, it is more difficult to bring the saturated devices out of this state. In addition the commutation diodes make a noise no matter how fast they are even with subbers.
QSC use class H in all their higher powered amplifiers, and this is OK for PA work, they sound awful in a HiFi situation. Their output stage is a common emitter type and this has serious common mode issues. Their open loop gain is massive (They use an NE5532 as the differential front end) and so applying massive global feedback certainly lowers static THD, the amplifiers sound cglassy and tin like. Hell but who can argue against $100 million a year sales.
I agree with Mr. Curl, multiple paralell BJTs for high power amps works the best. Beta droop is minimized and very high powered amplifiersa can be made with relatively low cost plastic devices. I confess that I do not like TO-3 devices due to their high cost, costly mounting and typically the dies they put in these cans are not the best - typically 1-4MHz devices.
Stephen Mantz
Zed Audio Corp.
Hello guys,
We have built amplifiers with both types of output devices. My opinion is that BJTs sound a lot nicer than Mosfets and that the fact that some Mosfets have massive SOA ratings this does not tell the complete story.
Mosfets need lots of idling current to lower their THD to an acceptable level and they also need thermal compensation.
Parallel BJTs out perform any Mosfet circuit.
Class G/H is nice for improved efficiency but do not sound very good. Why? Well when the inner devices are saturated and the outer devices are dumping the higher supply, it is more difficult to bring the saturated devices out of this state. In addition the commutation diodes make a noise no matter how fast they are even with subbers.
QSC use class H in all their higher powered amplifiers, and this is OK for PA work, they sound awful in a HiFi situation. Their output stage is a common emitter type and this has serious common mode issues. Their open loop gain is massive (They use an NE5532 as the differential front end) and so applying massive global feedback certainly lowers static THD, the amplifiers sound cglassy and tin like. Hell but who can argue against $100 million a year sales.
I agree with Mr. Curl, multiple paralell BJTs for high power amps works the best. Beta droop is minimized and very high powered amplifiersa can be made with relatively low cost plastic devices. I confess that I do not like TO-3 devices due to their high cost, costly mounting and typically the dies they put in these cans are not the best - typically 1-4MHz devices.
Stephen Mantz
Zed Audio Corp.
Re: BJTs vs Mosfets
Define the sound of a BJT and MOSFET in precise terms.MOER said:
I'd love to see your class AB BJT amp that significantly outperforms Bob's MOSFET amp with old 80s devices, i.e. significantly better than Bob's <10 parts per million THD over the full audio band and full power.
Re: BJTs vs Mosfets
Hi Mantz,
You are correct to some extent, but the story is quite different...
In Bob's Error Correction Mosfet amp based on Hawksford EC technique....I think the idle current is indeed low[<50mA a guess]....and THD figures were impressive even with HF signals....
We drive the mosfets in different way, each mosfet has its own driver configured as voltage controlled current source , it senses the variation of Isource and acts acordingly and therefore it needs no thermal compensation at all, the bias is extremely stable whether there is a variation in temperature [no thermal tracking is employed from heatsink]or supply rail voltage and the Mosfets share the current 99.99% exactly....much greater sharing than bipolars....
Class-G/H is a different field...its main area is increased efficiency and SOA reliability..In Class-H amps the output devices never go into saturation as well, because the rails are stepped in an instant...
If you consider Class-TD [Tracking rails] there performance is very well because there is no rail stepping and no commutation diode artifacts present in the output siganl.....there is no distraction of rails present....
In Class-AB amps large die mosfets are costly than paralleling several low cost BJT's...inspite of this normal die mosfets could very well replace the Bjts also if paralleled, But in high power amps using Class-H/TD high power [1KW upwards]large die mosfets were the perfect choice....
Cheers,
Kanwar
MOER said:
Hi Mantz,
You are correct to some extent, but the story is quite different...
In Bob's Error Correction Mosfet amp based on Hawksford EC technique....I think the idle current is indeed low[<50mA a guess]....and THD figures were impressive even with HF signals....
We drive the mosfets in different way, each mosfet has its own driver configured as voltage controlled current source , it senses the variation of Isource and acts acordingly and therefore it needs no thermal compensation at all, the bias is extremely stable whether there is a variation in temperature [no thermal tracking is employed from heatsink]or supply rail voltage and the Mosfets share the current 99.99% exactly....much greater sharing than bipolars....
Class-G/H is a different field...its main area is increased efficiency and SOA reliability..In Class-H amps the output devices never go into saturation as well, because the rails are stepped in an instant...
If you consider Class-TD [Tracking rails] there performance is very well because there is no rail stepping and no commutation diode artifacts present in the output siganl.....there is no distraction of rails present....
In Class-AB amps large die mosfets are costly than paralleling several low cost BJT's...inspite of this normal die mosfets could very well replace the Bjts also if paralleled, But in high power amps using Class-H/TD high power [1KW upwards]large die mosfets were the perfect choice....
Cheers,
Kanwar
Gentlemen,
Allow me to compliment you on a most excellent thread which was remarkably void from the usual plethora of posts by ill-informed soap-box gurus that seems to frequent this forum and which judge the 'sound' of an amplifier by its transistor count. It took me the better part of my sunday afternoon to read through all of it, but it has been a worthwhile read, if only to get the perspective of other designers.
I've just recently posted a DIY amplifier article on my website that offers Halcro type of distortion figures. I can't take credit for the design as Bruno Putzeys did the bulk of it, but we both worked on this for about 6-months where I tackled all of the prototyping and most of the (LTspice) simulations. The design was published in the Dutch Elektuur Audio Special magazine, for which I also happened to be the editor in chief for the solid state and loudspeaker sections.
http://www.hardwareanalysis.com/content/article/1842/
What we inspired to do is design an amplifier which would work as a reference and allow us to compare other designs to, both subjectively, in terms of sound quality, but also in terms of performance figures. The design is very similar to a balanced opamp in every respect and hence uses two bridged output stages and a balanced frontend with a common mode reference to ground.
During the design process we tried a few different approaches, such as a CFP bipolar output stage as well as a EF FET output stage. The latter was only simulated, but performed worse than the EF bipolar using ring-emitter Sanken 2SC2922/2SA1216. However experience shows that FETs do perform well when used in EF stages and biased quote heavily, in this case however (with ~2A per MT200 device) the bipolars seemed to be favorable. FYI, we compared to IRFP240 and IRFP9140 as those are a better complimentary match than the often used 240/9240.
I'd be interested in your comments regarding our design.
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
Allow me to compliment you on a most excellent thread which was remarkably void from the usual plethora of posts by ill-informed soap-box gurus that seems to frequent this forum and which judge the 'sound' of an amplifier by its transistor count. It took me the better part of my sunday afternoon to read through all of it, but it has been a worthwhile read, if only to get the perspective of other designers.
I've just recently posted a DIY amplifier article on my website that offers Halcro type of distortion figures. I can't take credit for the design as Bruno Putzeys did the bulk of it, but we both worked on this for about 6-months where I tackled all of the prototyping and most of the (LTspice) simulations. The design was published in the Dutch Elektuur Audio Special magazine, for which I also happened to be the editor in chief for the solid state and loudspeaker sections.
http://www.hardwareanalysis.com/content/article/1842/
What we inspired to do is design an amplifier which would work as a reference and allow us to compare other designs to, both subjectively, in terms of sound quality, but also in terms of performance figures. The design is very similar to a balanced opamp in every respect and hence uses two bridged output stages and a balanced frontend with a common mode reference to ground.
During the design process we tried a few different approaches, such as a CFP bipolar output stage as well as a EF FET output stage. The latter was only simulated, but performed worse than the EF bipolar using ring-emitter Sanken 2SC2922/2SA1216. However experience shows that FETs do perform well when used in EF stages and biased quote heavily, in this case however (with ~2A per MT200 device) the bipolars seemed to be favorable. FYI, we compared to IRFP240 and IRFP9140 as those are a better complimentary match than the often used 240/9240.
I'd be interested in your comments regarding our design.
Best regards,
Sander Sassen
http://www.hardwareanalysis.com
Re: BJTs vs Mosfets
Stephen,
Thanks for your observations and opinions. I agree with much of what you have said, but I think some of your generalizations are not justified.
There are plenty of extraordinarily good-sounding amplifiers that have been built with MOSFETs and others with bipolars. Similarly, there are many of each type that are poorly executed or don't sound good for some reason. The best designers know how to best use their chosen device, be it MOSFET or bipolar. It may be true that your BJT amplifeirs sound better than your MOSFET amplifiers for some reason.
I agree that SOA is only one small piece of the puzzle.
It is true that conventional MOSFET amplifiers need higher idle current than conventional Class-AB BJT amplifiers, but if error correction is used, at the expense of about six small-signal transistors, this issue also goes away. My MOSFET amplifier that achieved 0.001% THD-20 was biased at less than 150 mA. The higher bias current difference also tends to disappear when comparing Class AAB amplifiers of each type.
It is also true that HEXFET-based MOSFET amplifiers need bias temperature compensation, but they need less of it and are more temperature-transient stable than typical BJT designs. See my MOSFET power amplifier paper at www.cordellaudio.com for real experimental data that illustrates that.
Your statement that "parallel BJTs outperform any MOSFET circuit" is a completely unjustified generalization. You don't even say in what way. High frequency distortion? Does one of your BJT designs out-perform the Halcro MOSFET design (Stereophile's best-sounding amplifier)?
Can you supply us with some specs on your BJT amplifiers that outperform MOSFET amplifiers?
John Curl and others have shown that exceptional BJT amplifiers can be made by knowledgable designers that know how to get the best out of BJTs, no question. But your generalization about bipolars outperforming MOSFETs is planly wrong.
Cheers,
Bob
MOER said:
Stephen,
Thanks for your observations and opinions. I agree with much of what you have said, but I think some of your generalizations are not justified.
There are plenty of extraordinarily good-sounding amplifiers that have been built with MOSFETs and others with bipolars. Similarly, there are many of each type that are poorly executed or don't sound good for some reason. The best designers know how to best use their chosen device, be it MOSFET or bipolar. It may be true that your BJT amplifeirs sound better than your MOSFET amplifiers for some reason.
I agree that SOA is only one small piece of the puzzle.
It is true that conventional MOSFET amplifiers need higher idle current than conventional Class-AB BJT amplifiers, but if error correction is used, at the expense of about six small-signal transistors, this issue also goes away. My MOSFET amplifier that achieved 0.001% THD-20 was biased at less than 150 mA. The higher bias current difference also tends to disappear when comparing Class AAB amplifiers of each type.
It is also true that HEXFET-based MOSFET amplifiers need bias temperature compensation, but they need less of it and are more temperature-transient stable than typical BJT designs. See my MOSFET power amplifier paper at www.cordellaudio.com for real experimental data that illustrates that.
Your statement that "parallel BJTs outperform any MOSFET circuit" is a completely unjustified generalization. You don't even say in what way. High frequency distortion? Does one of your BJT designs out-perform the Halcro MOSFET design (Stereophile's best-sounding amplifier)?
Can you supply us with some specs on your BJT amplifiers that outperform MOSFET amplifiers?
John Curl and others have shown that exceptional BJT amplifiers can be made by knowledgable designers that know how to get the best out of BJTs, no question. But your generalization about bipolars outperforming MOSFETs is planly wrong.
Cheers,
Bob
Re: Re: BJTs vs Mosfets
My mosfet amp doesnot need thermal compensation of any type from the Heatsink..at all....No thermal runaway even with high bais with devices mounted on small heatsink......
What you have to say about it.....Bob...
One thing is for sure the BJT designers are uncomfortable with Mosfets, I think...
Cheers'
Kanwar
Bob Cordell said:
My mosfet amp doesnot need thermal compensation of any type from the Heatsink..at all....No thermal runaway even with high bais with devices mounted on small heatsink......
What you have to say about it.....Bob...
One thing is for sure the BJT designers are uncomfortable with Mosfets, I think...
Cheers'
Kanwar
john_ellis said:
Its not a Style , its a Technique based on Concrete information obtained from real world practical experiments, not just reading a book would solve your real world issue....
Have you done any Short Circuit testing on a Mosfet amp so far...
Bob Cordell said:
We came across many situations dealing with high current conditions in mosfets, when they got damaged, just because there wasnot enough matching in between them, source resistors were there but they do nothing...
If you took mosfets from same lot[batch] or tube, then the criteria of matching is very much good, but you must have to match them tightly in order to guarantee safe and sound reliability in all terms...
But again there's a pitfall,
Lets take an example...
In a bunch of tightly matched paralleled mosfets in an amplifier, if one of the mosfet fails for some reason, what should be done in order to replace the damaged part with newer one, wouldnot the matching could be difficult at that time because a large mismatch of batch and also if some one wants to repair that amp, then again its not feasible for him to get the newer part matched tightly with older ones....
But we have a good solution for this, as we use VCCS stage which ensures same bias current setting and same performance from the replaced device, no matter what order of mismatch is between the newer one and the older parts.... This solution keeps the biasing stable, no thermal compensation, ensures near 100% current sharing under all conditions, thereby maximizes the reliability...
Cheers,
Kanwar
Re: Re: Re: BJTs vs Mosfets
Kanwar,
Understood; in your approach you put each output MOSFET in a feedback loop with an op amp of high gain. This forces the source to rigidly follow the input to the cell. This also makes the output impedance of that block approximately equal to that of the source resistor for that block. This does have some nice features. As you pointed out, it eliminates the issue of Vds differences, since they are essentially servo'd out by the enclosed op amp. For the same reason, you effectively increase the transconductance of the enclosed device to a very high value, maybe even higher than that of a bipolar by itself.
Here is what I wonder about in your approach:
You are enclosing the output transistor in a very high-gain NFB loop, maybe using up all of your stability margin there, and maybe not leaving much for the global NFB.
Similarly, the output combined device may have significant excess phase delay, as the signal must pass through both the op amp, with its 90 degree phase shift, and the output transistor. You seem to be giving up the inherent very high speed of a source follower.
What happens to the op amp output when that side of the stage turns off in the Class AB transition? Does the op amp rail?
Since top and bottom have relatively fixed impedances essentially like that of each Rs, don't you get rather significant gm doubling effects?
In some respects, it would seem that the behavior of this stage might not be unlike that of a CFP MOSFET output stage, with its usual set of pros and cons.
Cheers,
Bob
Workhorse said:
My mosfet amp doesnot need thermal compensation of any type from the Heatsink..at all....No thermal runaway even with high bais with devices mounted on small heatsink......
What you have to say about it.....Bob...
One thing is for sure the BJT designers are uncomfortable with Mosfets, I think...
Cheers'
Kanwar
Kanwar,
Understood; in your approach you put each output MOSFET in a feedback loop with an op amp of high gain. This forces the source to rigidly follow the input to the cell. This also makes the output impedance of that block approximately equal to that of the source resistor for that block. This does have some nice features. As you pointed out, it eliminates the issue of Vds differences, since they are essentially servo'd out by the enclosed op amp. For the same reason, you effectively increase the transconductance of the enclosed device to a very high value, maybe even higher than that of a bipolar by itself.
Here is what I wonder about in your approach:
You are enclosing the output transistor in a very high-gain NFB loop, maybe using up all of your stability margin there, and maybe not leaving much for the global NFB.
Similarly, the output combined device may have significant excess phase delay, as the signal must pass through both the op amp, with its 90 degree phase shift, and the output transistor. You seem to be giving up the inherent very high speed of a source follower.
What happens to the op amp output when that side of the stage turns off in the Class AB transition? Does the op amp rail?
Since top and bottom have relatively fixed impedances essentially like that of each Rs, don't you get rather significant gm doubling effects?
In some respects, it would seem that the behavior of this stage might not be unlike that of a CFP MOSFET output stage, with its usual set of pros and cons.
Cheers,
Bob
Re: Re: Re: Re: BJTs vs Mosfets
Bob,
In our amp, we donot use very heavy global negative feedback, because all the correction is already done at the local feedback loops formed by the opamp+FET... which give excellent THD figures and reliability as well..
The Slewrate we have obtained with such configuration is around 75V/uS with normal opamps[220V/uS is obtained with LM6172 opamp]which i think is enough for justifying the Hi-End performance...
In case of Class-AB transistion, the subsequent opamp output is then automatically stays at 2.9V just to allow the bias of around 25mA to flow even when that side isn't in condcting mode..Its never turns off....but that doesnot means its cross-conducting and causing damage.....In fact the Transition from one polarity to another is very smooth...
The value of RSource is used is 0.05Ohms typically....I dont think gm doubling is a issue here...
Its not a regular CFP stage which is prone to oscillations, Cross-Conduction, poor HF response, and higher output impedance....aslo CFP require thermal feedback from heatsink to get bias stability
No rail sticking even if you drive the input of amp with+11dB signal level, only clean clipping with fast recovery...
The output damping is around 2000 at 1KHZ
FR=DC to 100KHZ
No therrmal tracking from heatsink
Again its a matter of choice as well taste..of every individual designer to do the best
Kanwar
Bob Cordell said:
Bob,
In our amp, we donot use very heavy global negative feedback, because all the correction is already done at the local feedback loops formed by the opamp+FET... which give excellent THD figures and reliability as well..
The Slewrate we have obtained with such configuration is around 75V/uS with normal opamps[220V/uS is obtained with LM6172 opamp]which i think is enough for justifying the Hi-End performance...
In case of Class-AB transistion, the subsequent opamp output is then automatically stays at 2.9V just to allow the bias of around 25mA to flow even when that side isn't in condcting mode..Its never turns off....but that doesnot means its cross-conducting and causing damage.....In fact the Transition from one polarity to another is very smooth...
The value of RSource is used is 0.05Ohms typically....I dont think gm doubling is a issue here...
Its not a regular CFP stage which is prone to oscillations, Cross-Conduction, poor HF response, and higher output impedance....aslo CFP require thermal feedback from heatsink to get bias stability
No rail sticking even if you drive the input of amp with+11dB signal level, only clean clipping with fast recovery...
The output damping is around 2000 at 1KHZ
FR=DC to 100KHZ
No therrmal tracking from heatsink
Again its a matter of choice as well taste..of every individual designer to do the best
Kanwar
Correction
Edit:Correction
Gm doubling could occur...Due to low value of Source Resistor 0.05 OHMS, there is a peak in Tansconductance, but again Local NFB applied to opamp corrects this and the output always tends to get linear......the voltage across the Rs is directly proportional to Vin at the input of cell....because the Cell is configured for unity voltage gain...w.r.t to Vin[opamp input]=>VS[across Source]
Kanwar
Bob Cordell said:
Edit:Correction
Gm doubling could occur...Due to low value of Source Resistor 0.05 OHMS, there is a peak in Tansconductance, but again Local NFB applied to opamp corrects this and the output always tends to get linear......the voltage across the Rs is directly proportional to Vin at the input of cell....because the Cell is configured for unity voltage gain...w.r.t to Vin[opamp input]=>VS[across Source]
Kanwar
Re: Re: Re: Re: Re: BJTs vs Mosfets
Thanks Kanwar,
This all sounds pretty reasonable. What value of full-power THD-20 do you achieve, or alternatively, how far down are the sprectral values of a full-power 19 kHz + 20 kHz twin tone test at 1kHz, 2 kHz, 3 kHz, 18 kHz, 17 kHz and 16 kHz? (e.g., second through 7th order).
Cheers,
Bob
Workhorse said:
Thanks Kanwar,
This all sounds pretty reasonable. What value of full-power THD-20 do you achieve, or alternatively, how far down are the sprectral values of a full-power 19 kHz + 20 kHz twin tone test at 1kHz, 2 kHz, 3 kHz, 18 kHz, 17 kHz and 16 kHz? (e.g., second through 7th order).
Cheers,
Bob
To B. Cordell
I have read your papers on distortion. Great stuf.
If you woud have to start from scatch developing/making top end amplifiers, what kind of test tools would you buy/make based on existing sound card and software to tune your design. This is of course for diy and not for specifying a product.
I am considering Praxis for acoustic and audio and making some low distortion audio oscillators and notch filters for thd.
Thanks for comments
Jean-Pierre Vanderreydt
I have read your papers on distortion. Great stuf.
If you woud have to start from scatch developing/making top end amplifiers, what kind of test tools would you buy/make based on existing sound card and software to tune your design. This is of course for diy and not for specifying a product.
I am considering Praxis for acoustic and audio and making some low distortion audio oscillators and notch filters for thd.
Thanks for comments
Jean-Pierre Vanderreydt
JPV said:
Thanks, Jean-Pierre,
Yes, things have changed a lot from when I first started out, and even from when I did the MOSFET amplifier. Back then, we did not have PC's, and our test equipment mainly had to be analog. I HAD to have a THD analyzer, but they were unaffordable, so I built my own and published the construction article in Audio. It was a challenge to build, but the article was very popular because so many people were in the same boat as I was. We sold over 200 sets of boards.
Two things have changed now. First, the PC with great soundcards and sophisticated software. Secondly, the used equipment market is overflowing with great HP and TEK analog test equipment that I used to have access to only at work and lusted to have for my own lab. The combination of these two makes life a lot better now. Add to that the ability to make purpose-built test equipment like my Distortion Magnifier, that greatly improves the usable dynamic range of a PC-based analyzer, and you are off to the races.
So, first, I'd puruse Ebay and pick up some good HP and TEK gear (I picked up a lot of TM-500 stuff that was well-built and fairly easy to repair - manuals are also available on Ebay). Secondly, I'd get some decent hardware-software for a PC-based system. For speaker design, I use and really like the Clio system. Third, I'd build special-purpose test gadgets to augment my other capabilities. The fourth thing, which I have not done yet, but want to do, is build some PIC-processor-based audio test equipment.
Hope this helps,
Bob
Hello Bob,
Always good to see a knowledgeable amp designer on the board.
A couple of questions for you regarding using power MOSFET outputs. I recently read your JAES article on the error correction amplifer and you say something like
The input capacitance of the MOSFET is effectively bootstrapped by the follower configuration and reduced by nearly an order or magnitude. Think you mention 170pf for a 100V/us slew rate. These were IRF fets.
My question is does a source follower have as low an input capacitance as being used in common source mode. ?
Do you have an idea how much capacitance is being driving in common source mode?.
I want to experiment with mosfets in common source mode for a regulator and smaller ones for a VAS but not sure on sizes of capacitance figures being driven.
In your design, you mention that the mosfets can supply something greater than 20A into 19 or 20khz at very low load impedances. If the output stage was driven by a series regulator would this current be reduced ?
Finally, if you look at this John Linsley Hood design at the top of the page, he uses the input and driver stage at lower voltage (+-50v dc) than the output power mosfet rail. (+-55v dc). He does this with lateral's and the the design is fully regulated. The opposite of what people normally do. The design has phase lead and the lag is generated over the base/gate connection of the VAS. Here is the link.
http://www.diyaudio.com/forums/showthread.php?s=&threadid=69527
I was wondering what you think about this arrangement? and are there disadvantages with clipping etc?. I guess he tried to save power wastage and cost for diyers building the kit originally.
Best Regards
Kevin
Always good to see a knowledgeable amp designer on the board.
A couple of questions for you regarding using power MOSFET outputs. I recently read your JAES article on the error correction amplifer and you say something like
The input capacitance of the MOSFET is effectively bootstrapped by the follower configuration and reduced by nearly an order or magnitude. Think you mention 170pf for a 100V/us slew rate. These were IRF fets.
My question is does a source follower have as low an input capacitance as being used in common source mode. ?
Do you have an idea how much capacitance is being driving in common source mode?.
I want to experiment with mosfets in common source mode for a regulator and smaller ones for a VAS but not sure on sizes of capacitance figures being driven.
In your design, you mention that the mosfets can supply something greater than 20A into 19 or 20khz at very low load impedances. If the output stage was driven by a series regulator would this current be reduced ?
Finally, if you look at this John Linsley Hood design at the top of the page, he uses the input and driver stage at lower voltage (+-50v dc) than the output power mosfet rail. (+-55v dc). He does this with lateral's and the the design is fully regulated. The opposite of what people normally do. The design has phase lead and the lag is generated over the base/gate connection of the VAS. Here is the link.
http://www.diyaudio.com/forums/showthread.php?s=&threadid=69527
I was wondering what you think about this arrangement? and are there disadvantages with clipping etc?. I guess he tried to save power wastage and cost for diyers building the kit originally.
Best Regards
Kevin
Here's one for people who know compensation well
Hello,
I just wanted to add something about regulating the output supplies. It seems to me (from what I have read) that having too large a capacitor on the output of a typical regulator effects the phase margin and can cause instability.
A regulated output can give solid output voltage, over 100db PSRR, very low impedance, noise etc at high voltage into high currents. I will post a "bootstrapped" high voltage high current regulator that I _think_ can do this in the PSU design shortly.
From what I can make out, the problem with output stage regulators is that they limit the transient (impulse) performance of the power supply at low impedance in spite of having several advantages.
I personally prefer the amp to be an ideal voltage source down to a low impedance, so as to not compress dynamics playing bass drums at hign level. This is a good test with moving coil speakers. I bought a Alan Parsons test CD for this - Soundcheck.
So what we need is pole splitting, permitting large C values on the output that have lower impedance than the reg. (delibrately) to supply the impulse performance of a convention PSU with all the advantages of the regulated design.
The compensation is the tricky part though.
Analog Devices recently released a range of LDO reg's that are called AnyCap and I have read some posts on compensation methods that may permit this with the Walt Jung superreg's.
These are definately not good examples of a reg to do this approach with as there impedance is far too low
Some thoughts of mine for the time being - at work.
Be back later.
Kevin
Hello,
I just wanted to add something about regulating the output supplies. It seems to me (from what I have read) that having too large a capacitor on the output of a typical regulator effects the phase margin and can cause instability.
A regulated output can give solid output voltage, over 100db PSRR, very low impedance, noise etc at high voltage into high currents. I will post a "bootstrapped" high voltage high current regulator that I _think_ can do this in the PSU design shortly.
From what I can make out, the problem with output stage regulators is that they limit the transient (impulse) performance of the power supply at low impedance in spite of having several advantages.
I personally prefer the amp to be an ideal voltage source down to a low impedance, so as to not compress dynamics playing bass drums at hign level. This is a good test with moving coil speakers. I bought a Alan Parsons test CD for this - Soundcheck.
So what we need is pole splitting, permitting large C values on the output that have lower impedance than the reg. (delibrately) to supply the impulse performance of a convention PSU with all the advantages of the regulated design.
The compensation is the tricky part though.
Analog Devices recently released a range of LDO reg's that are called AnyCap and I have read some posts on compensation methods that may permit this with the Walt Jung superreg's.
These are definately not good examples of a reg to do this approach with as there impedance is far too low
Some thoughts of mine for the time being - at work.
Be back later.
Kevin
