When it comes to selecting the output stage technology for minimum harmonic distortion, I see things as follows:
BJTs: Gives the least crossover distortion at optimum bias. However, temperature changes will never be tracked perfectly, making the optimum conditions practically unobtainable! Added to that, BJTs suffer from Beta droop.
FETs: Regardless of lateral or vertical, crossover distortion is lower at higher bias currents. The more the better! The only need to govern the bias seems to be in the case of vertical devices - to ensure that they don't just get hotter and hotter. Did I miss something??
So, in conclusion, it seems that the best design uses vertical MOSFETs whose N/P devices are well matched (e.g. 2SK1530 & 2SJ201). Potentially some padding with source resistors to further reduce droop. This assumes suitable driving circuitry and thermal tracking to avoid thermal runaway.
Anyone not agree?!!
BJTs: Gives the least crossover distortion at optimum bias. However, temperature changes will never be tracked perfectly, making the optimum conditions practically unobtainable! Added to that, BJTs suffer from Beta droop.
FETs: Regardless of lateral or vertical, crossover distortion is lower at higher bias currents. The more the better! The only need to govern the bias seems to be in the case of vertical devices - to ensure that they don't just get hotter and hotter. Did I miss something??
So, in conclusion, it seems that the best design uses vertical MOSFETs whose N/P devices are well matched (e.g. 2SK1530 & 2SJ201). Potentially some padding with source resistors to further reduce droop. This assumes suitable driving circuitry and thermal tracking to avoid thermal runaway.
Anyone not agree?!!
Yea, hundreds of things of which I am just barely understanding most. For over generalizations, BJT's are a bit more linear. MOSFETS are harder to stabilize. Vertical are designed to be switches, lateral designed to be linear. Look at the gate Cap. The list goes on and on but as I am a beginner in this, I don't want to put my foot in my mouth.
I only disagree with your conclusion that MOSFETS are better. Neither is better. Pick which set of good things and bad things fits YOUR design goals the best. There is no right answer.
Look into an EF BJT output with L-MOSFET drivers. Advantages on VAS loading, not starving the outputs and only compensating 2Vce. Modeled distortion is higher than BJT, but well within hi-fi if not playing the numbers game.
When testing my DH-120 rebuild with Exicon outputs, more bias was not better. I settled on about 130 mA, a tad leaner than some suggested.
I only disagree with your conclusion that MOSFETS are better. Neither is better. Pick which set of good things and bad things fits YOUR design goals the best. There is no right answer.
Look into an EF BJT output with L-MOSFET drivers. Advantages on VAS loading, not starving the outputs and only compensating 2Vce. Modeled distortion is higher than BJT, but well within hi-fi if not playing the numbers game.
When testing my DH-120 rebuild with Exicon outputs, more bias was not better. I settled on about 130 mA, a tad leaner than some suggested.
In my case, minimum distortion is pretty well at the top of the list of design goals. Class A is tempting but I like the challenge of getting almost there with a more practical AB design! With all this in mind I started looking at BJTs but as I say, things like bias stability and Beta drop make me believe that a MOSFET design is preferable (again, for low distortion). Of the two MOSFET types, it looks to be like the vertical one will give lower crossover distortion (less transconductance droop).
Boosted rails for the IPS and VAS a must unless you want to give up a bit of headroom.
It seems tempting to drive lateral mosfets right from the VAS, but don't. You can't overcome the capacitance. Far worse with V-fets. That was the problem with the B&K ST 140 and how it lacked dynamics. My Hafler DH-120 easily beat it in that category.
There is low distortion, and there is number chasing. I have learned a lot about how loads effect the amp, stability, thermal compensation, and clipping recovery. Basic low THD is not the only game in town by far. I wish I had the time to build iterations of all the ideas to see where the audible dividing line is. I was led astray for years with amps that due to their design, could mask bad tweeters than others. With good tweeters, a completely different set of things can be heard, right and wrong. ( How many times have you seen that $2000 amp plugged into a big-box speaker and the owner going on and on about subtle details. Usually cables are in the mix. )
Actually with vertical you will have a dickens of a time with the capacitance. Same with HEXFET, but go read everything Nelson Pass has to say on it. Might read Ron Elliot ( ESP) too. Laterals were designed to be linear amplifiers. HEXFET and V-FET were designed to be switches. The data sheet is often a hint. Go read the experts.
I have owned many of each:1 V-FET, 5 lateral FET, about 20 BJT of various typologies. Some all BJT, some hybrid, some all FET. No one is better. No one is worse. If I were to put my best guess out there, I would say the best sounding was the most carefully designed. Remember, the amplifier itself is only about half of what goes into an amp.
Right now, the best sounding amp I own is FET input with BJT output. Second best is BJT input with laterals on the output.
Thanks! Many good comments. I agree that lowest distortion is not the only thing but it is an interesting challenge to see what can be achieved (without having to go class A) - which brings me back to the question of bias optimisation when using vertical MOSFETs. Is there an optimal bias point (like for BJTs) or is more bias simply better - as long as no damage occurs?
The input capacitance of Mosfets is a relatively minor annoyance. Just run a driver with sufficient current/power.
One major difference between BJT and Mosfets is what they do when clipped.
If you think you never clip your amp, check again.
Minimum distortion?
Not really a major design goal. A false goal, imo.
Better to be aware of and look at the spectrum of distortion and distortion spectrum vs. power.
There are a number of designs out there and on here that are frighteningly low in distortion. I'd go with one of them if that is your goal. Syn08 has an excellent bipolar design - he's left here - his website documents.
But for ultra low distortion you're into Hawksford and Halcro.
The other interesting thing is that multiple paralleled output devices reduce distortion in the output stage, or so I am told.
_-_-
One major difference between BJT and Mosfets is what they do when clipped.
If you think you never clip your amp, check again.
Minimum distortion?
Not really a major design goal. A false goal, imo.
Better to be aware of and look at the spectrum of distortion and distortion spectrum vs. power.
There are a number of designs out there and on here that are frighteningly low in distortion. I'd go with one of them if that is your goal. Syn08 has an excellent bipolar design - he's left here - his website documents.
But for ultra low distortion you're into Hawksford and Halcro.
The other interesting thing is that multiple paralleled output devices reduce distortion in the output stage, or so I am told.
_-_-
" sufficient" Maybe easier said than done. Again, please consult folks like Nelson Pass on this.
In my model at least, doubling the outputs made as big an improvement as adding the darlington VAS in the distortion numbers. With an EF output, almost a no-brainer. With a CFP, better think twice. I also look in the model at what the voltages and currents look like all around. These can be really scary. I quickly convinced myself to double the outputs instead of I/V limiting. Maybe a false sense of security, but if the internals look good when stressed by impedance, temps, freq etc, then it may just sound good. Many SOP designs do not manage thermal compensation well. Many clip asymmetrically and this gets terrible into low impedance which buy the way, as a reactive load you will see. "Taint no such ting as an 8 Ohm speaker". After being lead down the correct path by fellow members, I found many of the published designs ( and commercially sold amps) starve the output bases under some conditions.
BJT and MOSFET have different things to protect and different ways to do it. Read up.
Chasing numbers is as bear says, not always the best goal. How it behaves is equally as important. Clipping, recovery, thermals, low impedance, beta droop, gm doubling, crossover issues, and on and on and on. Otherwise we would all just build Gain Clones amps. (Which are not bad in this respect by the way)
All that said, any well executed amp should be better than about .02 @ 20K 1-9, full power into 8 Ohms. If it says .0001, I would not bet it sounds better. It may. It may not. I have not heard one at .5% that did sound good. As an example, some find the Blameless fantastic. It is very clean by the numbers. Others find it very non-musical.
ESP has a nifty circuit that tells you when you actually clip, not some vague preset, but by comparing the output to the power supply. I intend to add this to my current build.
I offer this, hoping not to cause a long argument, but really, it IS all about simple harmonic distortion. The problem is, it is not about static conditions you can easily measure or model, but the stresses when putting in real dynamics and ugly (real speakers) as loads. Almost every defect I can think of will manifest itself as harmonic distortion if you know where and when to look. What I may call a low distortion amp is one that remains well behaved under all conditions.
In my model at least, doubling the outputs made as big an improvement as adding the darlington VAS in the distortion numbers. With an EF output, almost a no-brainer. With a CFP, better think twice. I also look in the model at what the voltages and currents look like all around. These can be really scary. I quickly convinced myself to double the outputs instead of I/V limiting. Maybe a false sense of security, but if the internals look good when stressed by impedance, temps, freq etc, then it may just sound good. Many SOP designs do not manage thermal compensation well. Many clip asymmetrically and this gets terrible into low impedance which buy the way, as a reactive load you will see. "Taint no such ting as an 8 Ohm speaker". After being lead down the correct path by fellow members, I found many of the published designs ( and commercially sold amps) starve the output bases under some conditions.
BJT and MOSFET have different things to protect and different ways to do it. Read up.
Chasing numbers is as bear says, not always the best goal. How it behaves is equally as important. Clipping, recovery, thermals, low impedance, beta droop, gm doubling, crossover issues, and on and on and on. Otherwise we would all just build Gain Clones amps. (Which are not bad in this respect by the way)
All that said, any well executed amp should be better than about .02 @ 20K 1-9, full power into 8 Ohms. If it says .0001, I would not bet it sounds better. It may. It may not. I have not heard one at .5% that did sound good. As an example, some find the Blameless fantastic. It is very clean by the numbers. Others find it very non-musical.
ESP has a nifty circuit that tells you when you actually clip, not some vague preset, but by comparing the output to the power supply. I intend to add this to my current build.
I offer this, hoping not to cause a long argument, but really, it IS all about simple harmonic distortion. The problem is, it is not about static conditions you can easily measure or model, but the stresses when putting in real dynamics and ugly (real speakers) as loads. Almost every defect I can think of will manifest itself as harmonic distortion if you know where and when to look. What I may call a low distortion amp is one that remains well behaved under all conditions.
Of this statement, I whole heartily disagree. It depends on how the mosfets are used. If the mosfets are used as if they are BJTs, then this statement is agreeable. However, if the circuit is designed to use mosfets properly then there is no reason one cannot build an excellent quality amplifier using them bias class AB. Proper gate drive and local error correction will make a significant difference in performance.
Mosfets have a significant difference than BJTs when it comes to the bias. The difference is in transconductance. Gm of the mosfets is significantly lower than BJTs at lower current conduction, or class AB bias. You can see in the first picture the difference in Gm of mosfets compared to BJTs. The second photo shows why the bias of mosfets must be higher than that of BJTs. The output impedance is based on the sum of the two devices so if the bias is two low, then the output Z rises at the zero current crossing resulting in a non-linear output Z and thus distortion related to that. There is some more info in this thread. Although lateral mosfets are much less conductive than vertical types, the transfer characteristics of lateral fets tend to mimic the 'square law' while the transfer characteristics of vertical fets are more exponential. This has a large influence in the distortion components generated by the different fets.
Attachments
Last edited:
Regarding the second photo, ideally you want a horizontal line to represent the sum of Gm of the two output devices. This is more difficult for mosfets because of the change in Gm vs Id. It is easy to see why BJTs would be more linear in this respect but there are advantages to mosfets with regard to reactive power handling and secondary breakdown (or rather lack there of). BJTs suffer beta droop and also a reduction in Ft as current rises. Mosfets do not have this problem and Ft actually increases with higher current flow, particularly vertical types. Mosfets have a non-linear capacitance and particularly Cgd becomes significantly larger at Vds saturation. Vertical mosfets are easier than BJT's to keep thermally stable and fluctuations in bias setting is not as critical because the Gm of the fets is so much lower at AB bias. Whichever device you choose, it is important that the driving circuit be designed for that device. Generally, BJTs require less extra components and complexity to use. Lateral fets have internal Zeners and are designed to be used as amplifiers but tend to be more expensive. Verticals can be a bit more on the tricky side.
Last edited:
You hit the nail pretty well on the head there! You mentioned something that I have thought about (but not yet analysed in depth) and that is the difference in transfer function (square law vs. exponential). I haven't focused on that part so much but guess that it is one of the advantages with lateral MOSFETs??
By the way, I read several warnings about the difficulties of coping with vertical FETs. As I see it, this is not insurmountable but mora a case of correct design and careful layout.
By the way, I read several warnings about the difficulties of coping with vertical FETs. As I see it, this is not insurmountable but mora a case of correct design and careful layout.
I am with you on that! How fantastic it is we can share these conversations. Think back before the internet. We were trapped by Popular Electronics and Heathkit.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.