First, a pointer to (what used to be Hitachi) Lateral power MOSFETs data sheets:
http://www.renesas.com/avs/servlet/request
I'll guess the output devices of interest are the usual 2SK1058 and 2SJ162.
They have a zero tempco bias point of ~100mA drain current with ~0.5V of the correct polarity between gate and source.
For playing around, if you bias at the zero tempco point, you can dispense with the servo'd bias schemes. A resistor is handy for this and simple but brings other issues to be concerned about as some have already pointed out. A couple of rectifiers with a cap across the pair works well also.
With no source degeneration resistors, that's about 1V across the bias resistor. I've seen some folks put the bias pot across a couple of rectifiers in series to avoid a catastrophe if the bias pot opens.
I used to be one of those folks that looked at bootstrapping and poo poo'ed it as old tech that was thankfully passe. Thanks to kind indirect slaps of encouragement from Nelson, Fred, Hugh and others, I've learned my lesson.
edit: OK, OK. Bad links. Go here:
http://www.renesas.com/eng/products/discrete/transistor/index.html
Then, click on link to power MOSFETs for general amplifier.
Sorry.
mlloyd1
(who still enjoys reminiscing while reading through his very first Hitachi power MOSFET databook from years ago and looking at the super simple power amp circuits)
http://www.renesas.com/avs/servlet/request
I'll guess the output devices of interest are the usual 2SK1058 and 2SJ162.
They have a zero tempco bias point of ~100mA drain current with ~0.5V of the correct polarity between gate and source.
For playing around, if you bias at the zero tempco point, you can dispense with the servo'd bias schemes. A resistor is handy for this and simple but brings other issues to be concerned about as some have already pointed out. A couple of rectifiers with a cap across the pair works well also.
With no source degeneration resistors, that's about 1V across the bias resistor. I've seen some folks put the bias pot across a couple of rectifiers in series to avoid a catastrophe if the bias pot opens.
I used to be one of those folks that looked at bootstrapping and poo poo'ed it as old tech that was thankfully passe. Thanks to kind indirect slaps of encouragement from Nelson, Fred, Hugh and others, I've learned my lesson.
edit: OK, OK. Bad links. Go here:
http://www.renesas.com/eng/products/discrete/transistor/index.html
Then, click on link to power MOSFETs for general amplifier.
Sorry.
mlloyd1
(who still enjoys reminiscing while reading through his very first Hitachi power MOSFET databook from years ago and looking at the super simple power amp circuits)
Christer said:
If you create a bias voltage for the output Mosfets with a
resistor only, then the bias voltage will reflect the amount of
current fed through it in the second stage. If this current is
determined by the supply voltage, then any fluctuation in
supply voltage longer than the RC constant of the bootstrap
network will alter the bias of the output stage. Since there are
no Source resistors on the output stage, it will be particularly
sensitive to this. If the supply goes up a hair, the bias can go
up a lot, and if the supply comes in a bit low, then the bias can
shut down. This is why you would need good DC regulation on
the supply, or use a constant current source to feed that portion
of the circuit.
As Fred points out, neither is there temperature compensation.
The positive temperature coefficient at the lower bias currents
of the Mosfets will cause the bias to wander up with
temperature, and a resistor for a bias network will not help
compensate for that. Depending on the choice of output
devices and heat sinking and so on, you could conceivably see
some thermal runaway, often thought to be unlikely with Mosfets.
Personally I would have used a Vgs multiplier as seen in the
A75 and similar amps, as it gives more constant voltage with
varying current, and also some temperature compensation.
But then that's me. Not having built the amp, I am only
speculating, and using lateral Mosfets will be less sensitive in
this regard over vertical, as they have less transconductance.
Thanks Nelson, but I see you still refer to the bootstrapping,
and my question was if you considered the problem to be
related to the bootstrapping or presnet anyway? I do agree
drain resistors would seem sensible and are usually present
also in lateral MOSET designs.
I suspect that even without the bootstrapping we may get
some slow oscillation in bias current, that may affect the sound?
I like the simplicitly of not having to use a Vbe multiplier for
these devices, but maybe it is sensible to put one in anyway?
Edit: Hadn't seen your follow-up post when posting the above.
and my question was if you considered the problem to be
related to the bootstrapping or presnet anyway? I do agree
drain resistors would seem sensible and are usually present
also in lateral MOSET designs.
I suspect that even without the bootstrapping we may get
some slow oscillation in bias current, that may affect the sound?
I like the simplicitly of not having to use a Vbe multiplier for
these devices, but maybe it is sensible to put one in anyway?
Edit: Hadn't seen your follow-up post when posting the above.
Christer said:
if you are going to "waste" one transistor as Vbe multiplier, you might use it as a CCS for the VAS stage and keep the resistor for biasing. You can also add another transistor at the tails of the input differential pair. That way, you have an amp that is (almost) rail voltage independent!
About six month's ago, made some spice simulations of some some variations on L-Mosfet amps. Nothing orginal on my part and the goal was educational rather than preliminary to actually building anything. In software at least using a resistor to bias an SF output seemed to work about as well as a Vbe multiplier. But more interesting was that the resistor value could be surprizing way off from optimal without distortion being affected much.
It left me wondering what would be the range of optimal values it you build the same amp 10 times from randomly purchased components. Perhaps you could then take the median value resistor and just plug it in and though you were certain to be suboptimal, the difference between that and the optimal value would not be audible. If this is a poassability someone must have tried it commercially as it it saves on final assembly ant test labor.
It left me wondering what would be the range of optimal values it you build the same amp 10 times from randomly purchased components. Perhaps you could then take the median value resistor and just plug it in and though you were certain to be suboptimal, the difference between that and the optimal value would not be audible. If this is a poassability someone must have tried it commercially as it it saves on final assembly ant test labor.
'Give me the wisdom to be sweet and pleasant lest I have to eat my words later.....'
In this spirit I offer some words about Rod's new mosfet design.
First, layout is critical, particularly with mosfet amps. Rod is NOT being commercial, and I'm disappointed some should think that after all the incredible project and theory input he's given DIYers all over the world with his excellent, no-nonsense web site.
I do agree with others, however, that the output devices really would be better off with source resistors, if only to render the quiescent relatively immune to supply voltage variation. I agree the the bias control should be a Vbe multiplier (or even better, a Vgs multiplier).
On diff pair emitter degeneration. This amp does not have huge open loop gain largely because the outputs have much lower transconductance than a bipolar stage. Thus, to keep feedback factor high, necessary to ensure good damping factor, low distortion and wide frequency response (remember that the source impedance of these mosfets is nowhere near as low as bipolars), diff pair degeneration is ill-advised.
I suspect that emitter degeneration has been popularized because of paranoia about feedback. I would stress this is a type of political correctness, and it really is ******** (no disrespect intended to anyone here present, BTW). Degeneration at the diff input stage does ameliorate device matching problems, it is true, and if stage current is ramped up the transconductance (Iout over Vin) of the stage can be increased. If the problem of phase shift is solved with careful layout and very fast VAS and output devices, then global feedback might be increased, rather than backed off, for better performance. If the amplifier were to be used only for offset control, not amplification, a lower GNFB regime would be fine. But it is used as an amplifier, and if phase margin is well controlled the NFB is a strength, not a weakness. My own AKSA uses lashes of feedback; I have calculated 64dB overall.
Bootstraps are interesting beasts. They suffer roughly +/-10% variation in VAS stage current across the audio range, FAR worse than a CCS, but they deliver a high impedance load to the collector which facilitates high OLG, necessary for good feedback factors. They also drop off in effectiveness beyond about 300KHz, which loads up the collector and very nicely brings the gain down to meet the Bode/Nyquist criteria by the pole frequency. This is a huge advantage as it permits less draconian methods for lag compensation, improving sonics.
However, a current source is not a particularly linear beast either, as the variation in Vce at the CCS active device considerably and dynamically alters it's Vbe. This changes the current output in time with the voltage fluctuations at the collector. Only cascoding prevents this unhappy phenomenon, and cascoding costs rail efficiency. This flaw adds a tiny but objectionable mix of harmonics to any signal, tending both towards the higher orders and the creation of intermodulation products, which frankly does not sound as good as a simple bootstrap. Elegant, no, effective, yes. You'll notice that nowhere have I mentioned distortion measurements; this is deliberate - I concentrate solely on how the circuit sounds. I can promise that if it sounds good, distortion will be very low anyway, but the reverse is rarely true.
The problem with audio amps, and specifically solid state, is that throwing complexity at the problem works only at one level. We reduce gross effects only to spawn other, more subtle problems. I've tried the lot, and came back after exhaustive listening tests to the very simple toplogies. I'd love to offer something of elegant complexity - it really is most appealing intellectually, and shows the world how frightfully clever one is - but generally such designs, SOTA paragons though they be, just don't seem to cut it sonically. To my ears, the ultimate sonics come from single ended amps (either thermionic or solid state!), but I'd better not go there........
As a general observation I'd opine that very complex designs often replace largish low order distortions with almost unmeasurable high order distortions. The human tragedy in all this is that vanishingly low levels of high order distortion are actually discernible in one subtle way or another by the human ear.......
Cheers,
Hugh
In this spirit I offer some words about Rod's new mosfet design.
First, layout is critical, particularly with mosfet amps. Rod is NOT being commercial, and I'm disappointed some should think that after all the incredible project and theory input he's given DIYers all over the world with his excellent, no-nonsense web site.
I do agree with others, however, that the output devices really would be better off with source resistors, if only to render the quiescent relatively immune to supply voltage variation. I agree the the bias control should be a Vbe multiplier (or even better, a Vgs multiplier).
On diff pair emitter degeneration. This amp does not have huge open loop gain largely because the outputs have much lower transconductance than a bipolar stage. Thus, to keep feedback factor high, necessary to ensure good damping factor, low distortion and wide frequency response (remember that the source impedance of these mosfets is nowhere near as low as bipolars), diff pair degeneration is ill-advised.
I suspect that emitter degeneration has been popularized because of paranoia about feedback. I would stress this is a type of political correctness, and it really is ******** (no disrespect intended to anyone here present, BTW). Degeneration at the diff input stage does ameliorate device matching problems, it is true, and if stage current is ramped up the transconductance (Iout over Vin) of the stage can be increased. If the problem of phase shift is solved with careful layout and very fast VAS and output devices, then global feedback might be increased, rather than backed off, for better performance. If the amplifier were to be used only for offset control, not amplification, a lower GNFB regime would be fine. But it is used as an amplifier, and if phase margin is well controlled the NFB is a strength, not a weakness. My own AKSA uses lashes of feedback; I have calculated 64dB overall.
Bootstraps are interesting beasts. They suffer roughly +/-10% variation in VAS stage current across the audio range, FAR worse than a CCS, but they deliver a high impedance load to the collector which facilitates high OLG, necessary for good feedback factors. They also drop off in effectiveness beyond about 300KHz, which loads up the collector and very nicely brings the gain down to meet the Bode/Nyquist criteria by the pole frequency. This is a huge advantage as it permits less draconian methods for lag compensation, improving sonics.
However, a current source is not a particularly linear beast either, as the variation in Vce at the CCS active device considerably and dynamically alters it's Vbe. This changes the current output in time with the voltage fluctuations at the collector. Only cascoding prevents this unhappy phenomenon, and cascoding costs rail efficiency. This flaw adds a tiny but objectionable mix of harmonics to any signal, tending both towards the higher orders and the creation of intermodulation products, which frankly does not sound as good as a simple bootstrap. Elegant, no, effective, yes. You'll notice that nowhere have I mentioned distortion measurements; this is deliberate - I concentrate solely on how the circuit sounds. I can promise that if it sounds good, distortion will be very low anyway, but the reverse is rarely true.
The problem with audio amps, and specifically solid state, is that throwing complexity at the problem works only at one level. We reduce gross effects only to spawn other, more subtle problems. I've tried the lot, and came back after exhaustive listening tests to the very simple toplogies. I'd love to offer something of elegant complexity - it really is most appealing intellectually, and shows the world how frightfully clever one is - but generally such designs, SOTA paragons though they be, just don't seem to cut it sonically. To my ears, the ultimate sonics come from single ended amps (either thermionic or solid state!), but I'd better not go there........
As a general observation I'd opine that very complex designs often replace largish low order distortions with almost unmeasurable high order distortions. The human tragedy in all this is that vanishingly low levels of high order distortion are actually discernible in one subtle way or another by the human ear.......
Cheers,
Hugh
Emitter degeneration makes input BJT dif. pair considerably less affected by HF interference and D/A residuals, both results in better sonic result compared to non-degenerated design. The designer should try to achieve the best linearity before the NFB is applied, this is the only way how to get superior sonic results, the direct compare is the only judge.
And indeed this is the case. There is variation, but it is not to the degree where it becomes even a slight problem. I have built several of these amps now, and all have performed with almost identical characteristics. Once the bias is set (even in the low power version with no source resistors), normal mains variations cause no major upset to the quiescent current.
There has been a lot of comment about using a bootstrapped VAS, and I too had serious thoughts about going active. However, after I tested the prototype I was reluctant to change, since for a MOSFET amp, it saves an additional diode and capacitor (to allow the drive rail to hold up under load - as used in the +ve supply rail of the circuit now). While these are not expensive at all, they will still not allow the negative swing that the bootstrap circuit does.
The main reason for not changing though, was simply that the performance (both measured and by listening test) was very good (not just my listening - others agreed), and I figured that it wasn't broken, and therefore required no fixing
As Hugh pointed out, degeneration in the input pair reduces gain, and a MOSFET amp really does need more gain than a bipolar. Again, after making several amps and seeing almost identical figures (without matching i/p transistors - I wanted to see "worst case"), that was not broken either.
The vast majority of my visitors are hobbyists, and a great many are newbies - highly complex amps are not the way to introduce people into audio, and just discourage those who wish to experiment. Some very highly regarded amps (such as the AKSA) are also very simple, but there is no criticism of them - what have I done wrong? Just about every design I publish gets rubbished here for some reason
Basically, the P101 amp is a very nice sounding unit, it's simple to make and performs well. It is not perfect, the design was never intended to be the ultimate, and there are a few things that could be "improved" - none of which would make it simpler, cheaper or sound substantially better.
As always, my designs are intended to produce good results without too much chance of ending up with a whole bunch of expensive dead components. That has been my goal all along, and despite the reservations here, it has been achieved.
Interesting discussion though
Cheers, Rod
Gee this thread sure took off quickly...
Well, I hope you'll excuse my criticism of the bootstrap somewhat, in that I wasn't taking into account Rod's design goal of simplicity. Actually, I don't really have much issue with bootstraps in class-A designs, nor have I ever considered bootstrapping an outdated concept. That said, the first thing that has always come to mind for me when I see a bootstrapped VAS on a class-B or AB design is coupling of output stage distortions back into the VAS... and correct me if I'm wrong here, but doesn't output loading bring down the bootstrap voltage, thus reducing OLG, to which a mosfet design like this would be particularly susceptible? Perhaps this is another reason Rod left out the emitter resistors. I also recall reading somewhere about undesireable behaviour during clipping, but in all honesty it's been too long for me to remember the details. Perhaps someone can comment further about the overload behaviour of bootstrapped VAS's? Clipping behaviour is something I've been interested in more and more lately, so any input on the matter would be most welcome...
In any case, Hugh's comments wrt OLG and lag compensation have provided some fresh insight for me, so perhaps it's time to re-evaluate my stance on bootstrapped VAS's.
Well, I hope you'll excuse my criticism of the bootstrap somewhat, in that I wasn't taking into account Rod's design goal of simplicity. Actually, I don't really have much issue with bootstraps in class-A designs, nor have I ever considered bootstrapping an outdated concept. That said, the first thing that has always come to mind for me when I see a bootstrapped VAS on a class-B or AB design is coupling of output stage distortions back into the VAS... and correct me if I'm wrong here, but doesn't output loading bring down the bootstrap voltage, thus reducing OLG, to which a mosfet design like this would be particularly susceptible? Perhaps this is another reason Rod left out the emitter resistors. I also recall reading somewhere about undesireable behaviour during clipping, but in all honesty it's been too long for me to remember the details. Perhaps someone can comment further about the overload behaviour of bootstrapped VAS's? Clipping behaviour is something I've been interested in more and more lately, so any input on the matter would be most welcome...
In any case, Hugh's comments wrt OLG and lag compensation have provided some fresh insight for me, so perhaps it's time to re-evaluate my stance on bootstrapped VAS's.
Regarding diff pair emitter degeneration, I think D. Self has argued sufficiently (to me at least) that gentle to moderate degeneration achieves a more linear diff pair at the expense of some transconductance. I refer you to his article in Sept '93 EW, specifically his graphs on page 733 (fig 5). If anything, Doug Self is certainly not paranoid about global feedback! I think it's a worthwhile tradeoff, if one can spare the OLG.
millwood said:
Yes, but I wasn't thinking specifically of Rods amp here, but
L-MOSFET amps in general with resistor bias.
Sam9,
I have also done a bunch of such simulation on some of Slones
amps and the Zenquito/Mosquito. I haven't kept the simulation
results but also seem to remember a surprisingly small variation
in distorsion when changing the bias resistor. However, one
should keep in mind that it is only simulation and also that
transistor models for Spice usually don't correlate as well with
the datasheets as one would like, especially outside in the
operating region that matters for cross-over distorsion.
Rod,
Nice to see you have made it to this forum. Welcome.
hifiZen said:
And heavy degeneration makes it even more linear. You could
do the math on this, but the formula you get may not
be that intuitive, so it is better to just simulate and plot what
happens.
A diff pair with no degenration is very linear for a difference
voltage up to about +/-60mV (it depends a bit on what you
want to accept as linear) then it starts to behave very
non-linear instead. That is, as long as you stay in the linear
region it has very good performance, but as soon as you go
outside it you get very poor performance. Degeneration
increases the linear region, but at the expense of
transconductance. One may note, for instance, that the Leach
amp uses very heavy degeneration, 300 Ohms if I remember
correctly, the reason being to make sure that a step at the
input cannot overload the input stage into the non-linear
region. (However, that is based on the old TIM theory which I
get an impression is nowadays considered to have been a
bit of a red herring due to not understanding current starving
problems?) An LP filter at the input and high slew rate of the
amp will reduce the need for this, I suppose.
This topic is actually something I have been wanting to hear
peoples opinion on for some time anyway, so since it has
been brought up here, what are peoples opinion on degeneration
and for instance the Leach approach of using very heavy
degeneration. I am not primarily thinking of its effect to help
against mismatching of transistors, but the effect of expanding
the linear region and whether this is useful or not.
I have been a rabbid fan of bootstrap from day one because of their reasonable performance and simplicity.
my experience exactly.
don't be discouraged, Rod. We have a lot of arm-chair designers here (me included) who haven't had any success remotely close to yours but think of themselves too highly.
I am sure it will be an interesting amp to try. It is on my list to do now.
rode said:
my experience exactly.
rode said:Just about every design I publish gets rubbished here for some reason
don't be discouraged, Rod. We have a lot of arm-chair designers here (me included) who haven't had any success remotely close to yours but think of themselves too highly.
rode said:
I am sure it will be an interesting amp to try. It is on my list to do now.
Overload (clipping) recovery is something that I'm almost paraniod about, and if I see "rail sticking" (where the signal sticks to the maximum +ve or -ve output voltage for many microseconds, then suddenly returns to normal) then IMO there is something wrong with the design.
A bootstrapped VAS normally has no effect on whether an amp will have that problem or not. All of my designs are well behaved in that respect - they'd never make to the site if they were otherwise.
I normally drive the input with up to 10V RMS, then drop back to see the recovery time (should be virtually instantenous). It used to be easier until my tone-burst generator died, but I haven't had a chance to fix it. Clipping behaviour is always thoroughly tested though, as it makes a big difference to how an amp sounds when (over)driven hard. Of course, this should never happen, but we all know it will
Some power opamps suffer from extreme rail-sticking, especially if they are driven to the limits. The sound is unbearable when it happens, as the harmonic content becomes extreme.
Regarding the queries of bootstrap linearity, it is not a problem over the normal operating range of an amp. At very low frequencies (where the bootstrap cap has significant reactance WRT the resistors), distortion does increase as the linearity of the bootstrap current source is reduced.
The -3dB frequency of the bootstrap circuit used is about 0.2Hz, so there is no significant degradation even at the lowest frequencies of interest.
Cheers, Rod
Yeah, bootstrapping gets an undeserved bad rap. Properly
done, it works fine, and in the case of an amp like this, also
provides a little extra voltage swing.
The comment about constant current sources suffering from
non-linearity is equally untrue. Evaluated by themselves, a
decent CCS has very low distortion.
done, it works fine, and in the case of an amp like this, also
provides a little extra voltage swing.
The comment about constant current sources suffering from
non-linearity is equally untrue. Evaluated by themselves, a
decent CCS has very low distortion.
Christer post 30......as promised a reply....this thread has roared faster than the tubes one....
Hang on Christer.....you mentioned neg coef.......have you get the definition right ? NPN/PNP Transistors yes, go neg.....Lateral structured mosfets like most other enhancement types have a ++ coef. Otherwise parallelling devices wouldn’t be reliable. Graphically, the rising resistance curve with 2 parallel devices on the same heatsink is a stable balance... .
The previous generation mosfet types (with ident designation as modern types) had inherent internal construct/dissipation problems, which eventually destructed on sustained load. Modern mosfets are vertical/lateral types, supposed to have better constructed internals for heat disspiation. But failures still occur...why ?
Most of the amps I repair often use mosfets paralled. . In my example the partners IRF9540 and 530 are virtually found in most complemetary based amps. The positive thermal coef shouldn’t be a problem, but often the SOA at high temps v.s supply volts often clip the RH derating curves. The temptation for louspeaker bridging and reckless load applications forces too much on standard designs, merely for companies to push up sales.
The constant current sourced driver circuit is the most neglected problem area. A true class A driver for mosfet o/p stages will dissipate considerable heat to maintain dv/dt to overcome output stage Ciss, Cgs & Miller effect for low THD.
The throughput CS quies o/p stage current for single pair in class AB config is often around 80-100mA to avoid conduction overlap, with drivers linked to total supply implying 50+/-50 =100V + rated devices. Failings: poor heatsinking of drivers and cyclic heat with time leading to fatigue.
Do the dissipation equation and see if the current sources for a specified full supply voltage aren’t going to stew the devices...The data and de-rating regarding all types of mosfets should be understood by every equipment designer...at 50°C case temp, alot of the parameters have to be derated by nearly half for proper reliability. The trick is to throw in a fan, okay for MI, but that can ruin Hi Fi aspects.
Evidence with simpler circuit configurations which don’t “thump and plop” with switch-on transients are more reliable. The very early Hitachi 100W (early 1972) complementary version using double diff input/drivers + 2SK133 & 2SJ48 still rules the day without knocking the ornaments off the walls..each time the unit is switched on......
rich
Hang on Christer.....you mentioned neg coef.......have you get the definition right ? NPN/PNP Transistors yes, go neg.....Lateral structured mosfets like most other enhancement types have a ++ coef. Otherwise parallelling devices wouldn’t be reliable. Graphically, the rising resistance curve with 2 parallel devices on the same heatsink is a stable balance... .
The previous generation mosfet types (with ident designation as modern types) had inherent internal construct/dissipation problems, which eventually destructed on sustained load. Modern mosfets are vertical/lateral types, supposed to have better constructed internals for heat disspiation. But failures still occur...why ?
Most of the amps I repair often use mosfets paralled. . In my example the partners IRF9540 and 530 are virtually found in most complemetary based amps. The positive thermal coef shouldn’t be a problem, but often the SOA at high temps v.s supply volts often clip the RH derating curves. The temptation for louspeaker bridging and reckless load applications forces too much on standard designs, merely for companies to push up sales.
The constant current sourced driver circuit is the most neglected problem area. A true class A driver for mosfet o/p stages will dissipate considerable heat to maintain dv/dt to overcome output stage Ciss, Cgs & Miller effect for low THD.
The throughput CS quies o/p stage current for single pair in class AB config is often around 80-100mA to avoid conduction overlap, with drivers linked to total supply implying 50+/-50 =100V + rated devices. Failings: poor heatsinking of drivers and cyclic heat with time leading to fatigue.
Do the dissipation equation and see if the current sources for a specified full supply voltage aren’t going to stew the devices...The data and de-rating regarding all types of mosfets should be understood by every equipment designer...at 50°C case temp, alot of the parameters have to be derated by nearly half for proper reliability. The trick is to throw in a fan, okay for MI, but that can ruin Hi Fi aspects.
Evidence with simpler circuit configurations which don’t “thump and plop” with switch-on transients are more reliable. The very early Hitachi 100W (early 1972) complementary version using double diff input/drivers + 2SK133 & 2SJ48 still rules the day without knocking the ornaments off the walls..each time the unit is switched on......
rich- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.