MOSFET vs. BJT vs. JFET in amplifiers
G. Hugh Song
years although big popular names such as Krell, Threshold, and Mark-
Levinson are all using BJTs.
There must be pro and cons for each transistor type in addition to
the most obvious one about even/odd harmonic structure.
But I have always thought BJTs are more linear in %-wise and less
noisy. What about speed?
If you are to build your own dream power amplifier, which would you choose
as the driver stage device? For what reason?
How about in the preamp or input stage of a power-amp? Some use JFETs.
Others use BJTs.
What would you choose in your own dream amplifier? For what reason?
Thanks a lot.
Hugh
PS: I left out tubes intentinally because the use of tube has been
discussed too often without much tangible information.
JOHN I LIPP
MOSFETS and JFETS are very similar in design and principles of operation. In
general these devices are square-law devices and produce primarily second-harmonic
distortion products. When used in a push-pull design which significantly reduces
even order harmonic distortion, you can get a very low distortion circuit using
less feedback than you could with a BJT design.
MOSFETS are also capabable of higher frequency operation, but we are talking about
the mega Hz range of operation; this is well out of the 20-20k band that audio
is interested in.
I would theorize that the main motivation for using MOSFETS in amplifier power
stages has nothing to do with this though. It has to do with easy of design. A
BJT that is designed for high-current has a very low current gain ("beta"), often
around 10. Thus, if you make an amplifier that can deliver 50 amps, the stage
that drives the output devices must be able to supply 5 amps! Thus you have to
have a modest power amplifier to drive the output stage of your real amplifier!!!
MOSFETS only require enough current to charge and discharge their input capacit-`
ances. The current level to do this is on the order of 10 mA to 100 mA, dependent
on device geometry. In addition, power sharing for MOSFETS works better than
BJT's and thus designing an output stage with several devices in parrallel is
trivial (it only requires one resistor in addition to the power transistor).
Another motivation that is taughted as a MOSFET advantage is the lack of thermal
run-away and no region of secondary breakdown (the two are related). This is only
true when a MOSFET is biased at relatively high current levels, 1 amp or higher
(this is device dependent and can be as high as 10 amps for large devices, and
as low as 250 mA for a 20W device). This corresponds to constant power dissap-
ation of at least 80 watts for a 100 W amplifier channel!! I don't know of many
class-AB amplifiers that fit this description (but almost all class-A amplifiers
do).
The only advantages that I know of for a BJT are cost and assured high-frequency
stability (MOSFETS can oscillate if the circuit layout/design is poor). Also,
BJT op-amp circuits have lower offset voltage as a rule, but I don't think this
applies to discrete designs. Their could be a noise or other issues that I am
unaware of.
/***************************************************************************\
* John Lipp * Loudspeaker design pseudo-expert and software developer *
* li...@mtu.edu * (Working on amplifier and surround-sound psuedo-expertise) *
\***************************************************************************/
Randall Bradley
amp?
This is a complex question which cannot be delt with in a
simple or direct way in the context of a forum, especially
the "lowly" rec.audio.
The issues one faces are of real world compromises, not
selection of idealized dream devices. EVERYTHING in an amp
design represents some sort of COMPROMISE or another.
Therefore there is no "right" choices, or "perfect" amps. There
are only groupings of compromises that err in certain ways
more or less than others, and so represent a "good" choice.
The entire circuit, plus its REAL WORLD implementation must
work together. Witness the rather marvelous sound that can be
culled from modified (updated parts mostly) "old" tube amps. On
paper they cannot vaguely compete spec wise with even the most
horrific ratty sounding cheezo solid state receiver. BUT, they
sound (or can sound) super.
Great amps are more than a circuit design, and more than a
set of expensive parts. There's still much to the ART of
TECHNOLOGY. At least when it comes to audio.
(this is not to imply that I am a rabid tube-o-phile, I'm not.
I merely wish to use tube gear as an example)
regards from audio Nirvanna,
-_-randy bradley-_-
(BEAR Labs)
replies via: ra...@rdrc.rpi.edu (please)
BYRNES,Graham
> Some information on different transistor types:
>
> MOSFETS and JFETS are very similar in design and principles of operation. In
> general these devices are square-law devices and produce primarily second-harmonic
> distortion products. When used in a push-pull design which significantly reduces
> even order harmonic distortion, you can get a very low distortion circuit using
> less feedback than you could with a BJT design.
>
> MOSFETS are also capabable of higher frequency operation, but we are talking about
> the mega Hz range of operation; this is well out of the 20-20k band that audio
> is interested in.
It *is* important: it allows the dominant pole of the amp to be placed higher,
allowing more feedback at the high end of the audio band.
>
> I would theorize that the main motivation for using MOSFETS in amplifier power
> stages has nothing to do with this though.
I disagree :-)
> It has to do with easy of design. A
> BJT that is designed for high-current has a very low current gain ("beta"), often
> around 10. Thus, if you make an amplifier that can deliver 50 amps, the stage
> that drives the output devices must be able to supply 5 amps! Thus you have to
> have a modest power amplifier to drive the output stage of your real amplifier!!!
So you need a compound output stage which is slower still, see comment above.
> MOSFETS only require enough current to charge and discharge their input capacit-`
> ances. The current level to do this is on the order of 10 mA to 100 mA, dependent
> on device geometry. In addition, power sharing for MOSFETS works better than
> BJT's and thus designing an output stage with several devices in parrallel is
> trivial (it only requires one resistor in addition to the power transistor).
>
> Another motivation that is taughted as a MOSFET advantage is the lack of thermal
> run-away and no region of secondary breakdown (the two are related). This is only
> true when a MOSFET is biased at relatively high current levels, 1 amp or
^^^^^^^^
This is wrong. For the Hitachi 2sj49 (7Amp 100W) series, neg thermal feedback
occurs beyond 100mA with normal sized heatsinks.
higher
> (this is device dependent and can be as high as 10 amps for large devices, and
> as low as 250 mA for a 20W device). This corresponds to constant power dissap-
> ation of at least 80 watts for a 100 W amplifier channel!! I don't know of many
> class-AB amplifiers that fit this description (but almost all class-A amplifiers
> do).
>
Actually, a class A would dissipate at least 200W, and considerably more
for MOSFETs, see below.
> The only advantages that I know of for a BJT are cost and assured high-frequency
> stability (MOSFETS can oscillate if the circuit layout/design is poor). Also,
> BJT op-amp circuits have lower offset voltage as a rule, but I don't think this
> applies to discrete designs. Their could be a noise or other issues that I am
> unaware of.
They also have much lower series (drain-source) resistance: a BJT can swing to
within a volt of the rail, a MOSFET will probably not get closer than 6V.
So you need higher rail voltages for the same power rating, hence fets are
less efficient.
>
Graham B
Scott Rowin
is the Parasound series... the HCA1200 uses a bunch of them, reading over
the description on MOSFETs all seems to apply to the HCA1200.. it runs
fairly hot all the time - idle it draws 100W! Course I talked to a tech.
at Parasound who said they Class-A biased the heck out of it, though it still
falls in the Class-AB range. The current specs on it have current levels to 57
amps. Its a very clean sounding amplifier - its one of those amplifiers that
should cost twice as much as it does, but they're a reasonable company
(like Adcom) in pricing..
--> Amigo
G. Hugh Song
: One of the better amplifiers in the affordable class that uses MOSFETs
: is the Parasound series... the HCA1200 uses a bunch of them, reading over
: the description on MOSFETs all seems to apply to the HCA1200.
I am sorry. I do not know about HCA1200. But, from Stereophile review,
their top of the line HCA2200 does not use MOSFETs in the output stage.
They use them in the drive stage. In the output stage, they use BJTs.
: it runs
: fairly hot all the time - idle it draws 100W! Course I talked to a tech
: at Parasound who said they Class-A biased the heck out of it, though it still
: falls in the Class-AB range.
Some class-AB bias BJT amplifiers with comparable wattage rating from
other manufacturers boast that their idling power consumption is 350 W.
: The current specs on it have current levels to 57
: amps. Its a very clean sounding amplifier - its one of those amplifiers that
: should cost twice as much as it does, but they're a reasonable company
: (like Adcom) in pricing..
I don't know what John Curl did. But HCA2200 contains an IC buffer stage
which should be passed through no matter whether you switch between
ballanced and unballanced inputs according to Stereophile.
To me, HCA2200 seems to be the best buy for people who consider the
bigger the better.
Hugh
Bob Neidorff
In article <1993Feb25.1...@mtu.edu>, li...@mtu.edu (JOHN I LIPP) writes:
|> ...When used in a push-pull design which significantly reduces
|> even order harmonic distortion, you can get a very low distortion circuit using
|> less feedback (with MOS) than you could with a BJT design.
This is not true. MOS devices are lower in gm, and higher in open-loop
distortion.
|> MOSFETS are also capabable of higher frequency operation...
Again, this has been disproved many times. As an example, however, note
that in the RAM business, the fastest stuff is ECL, not CMOS. It's that
old low-gm problem again. If you saturate the BJT, however, it will behave
much slower, and perhaps this is what John Lipp means. In that he is correct.
|> (the main reason for MOS) has to do with easy of design.
I feel that there is some advantage in MOS due to high current gain, but
probably the more significant advantage for audio is that MOS circuits clip more
gently. They don't clip as harshly, and recover from clipping much faster.
|> In addition, power sharing for MOSFETS works better than
|> BJT's and thus designing an output stage with several devices in parrallel is
|> trivial (it only requires one resistor in addition to the power transistor).
This is almost true. High power stages are touchy with any device. There
is lots of room for funny effects with BJTs and MOS in parallel.
|> Another ... MOSFET advantage is the lack of thermal
|> run-away and no region of secondary breakdown (the two are related).
Quite true. Good MOSFET amplifiers are very robust.
|> ... advantage ... for a BJT are ... assured high-frequency
|> stability (MOSFETS can oscillate if the circuit layout/design is poor).
A darlington connected BJT will tend to oscillate if loaded with a capacitor.
The lower gm of MOS actually makes MOS a more stable output stage.
|> BJT op-amp circuits have lower offset voltage as a rule, but I don't think this
|> applies to discrete designs. Their could be a noise or other issues that I am
|> unaware of.
Lower offset can be an advantage in linearity. If you have nonlinearity
in the output stage, this will be attenuated by the feedback from the gain
of the preceeding stage. However, nonlinearity in the input stage due
to mismatch in the input pair will introduce slight distortion that no
feedback can reduce. It is summed directly with the input.
BJTs are lower noise than MOS devices, but this rarely matters for amplifiers.
It is very valuable in a phono preamp, however.
--
Bob Neidorff; Unitrode I. C. Corp. | Internet: neid...@uicc.com
7 Continental Blvd. | Voice : (US) 603-424-2410
Merrimack, NH 03054-0399 USA | FAX : (US) 603-424-3460
Bob Myers
> that in the RAM business, the fastest stuff is ECL, not CMOS. It's that
> old low-gm problem again. If you saturate the BJT, however, it will behave
> much slower, and perhaps this is what John Lipp means. In that he is correct.
Bad example, Bob; the main reason that ECL is the speed champ is that it IS
a non-saturating logic. Your other points are correct, but let's not spread
false impressions here.
Bob Myers KC0EW Hewlett-Packard Co. |Opinions expressed here are not
Systems Technology Div. |those of my employer or any other
my...@fc.hp.com Fort Collins, Colorado |sentient life-form on this planet.
Chris O'Neill
MAT...@LURE.LATROBE.EDU.AU (BYRNES,Graham) writes:
>In <1993Feb25.1...@mtu.edu> li...@mtu.edu writes:
>> MOSFETS are also capabable of higher frequency operation, but we are talking
about
>> the mega Hz range of operation; this is well out of the 20-20k band that
audio
>> is interested in.
>
>It *is* important: it allows the dominant pole of the amp to be placed higher,
>allowing more feedback at the high end of the audio band.
>>
>> I would theorize that the main motivation for using MOSFETS in amplifier
power
>> stages has nothing to do with this though.
>I disagree :-)
>> It has to do with easy of design. A
>> BJT that is designed for high-current has a very low current gain ("beta"),
often
>> around 10. Thus, if you make an amplifier that can deliver 50 amps, the
stage
>> that drives the output devices must be able to supply 5 amps! Thus you have
to
>> have a modest power amplifier to drive the output stage of your real
amplifier!!!
>So you need a compound output stage which is slower still, see comment above.
I would like to clarify what this situation is about and whether having an
extra transistor before the output necessarily produces a "slower" arrangement
than a MOSFET.
The idea of "slower" doesn't look very useful to me in circuit design. I
prefer to find out what the poles and zeros are and from these work out the
circuit behaviour including what happens under feedback.
With a BJT audio amp the output transistor is usually set up in common
collector configuration so the current gain of this will have a pole at ft/beta
and a zero at ft. ft might be 1 Mhz and beta 20, hence a pole at 50 khz and
zero at 1 Mhz.
Now suppose you have a pre-output transistor driving the output in a Darlington
pair arrangement and that this transistor has an ft of 50 Mhz and a beta of 50.
In a Darlington pair, the poles in the overall current gain are at ft/beta for
each transistor and the zeros are at ft for each transistor.
So for the above transistors, you have poles at 50khz and 1 Mhz and zeros at 1
Mhz and 50 Mhz in the overall current gain. The 1 Mhz pole and zero cancel
each other so the overall current gain has a net pole at 50 khz and zero at 50
Mhz.
Now with a MOSFET, since the input is nearly pure capacitance driven by a high
output resitance driver circuit, there is a pole at a fairly low frequency (5
khz if the driver circuit output resistance is 10 k ohms and MOSFETs total
equivalent input capacitance is 3 nF). The capacitance that is mainly
responsible for shorting the input current to a MOSFET in common drain
configuration (assuming that is what is being used) is the gate-drain
capacitance. In this case the overall current gain just has the pole at 5 khz,
and no zeros. If the transconductance of the MOSFET is 1/(0.5 ohms), the
current gain from MOSFET input to MOSFET output drops to 1 at 100 Mhz.
The overall comparison is that the MOSFET gives more gain in the audio
frequencies than the transistors, but that the transistors in Darlington pair
have just one significant pole just like the MOSFET. The cancelling pole and
zero in the Darlington pair example here might not exactly cancel in real
examples, but as long as the zero is below the pole or not too far above it,
the pole won't cause any stability problem.
So I don't think it's necessarily true that MOSFETs reduce stability problems
just because they have far more current gain per device than transistors.
However, allowing a simpler circuit might reduce problems with stray
capacitance and lead inductance (might :-).
>> MOSFETS only require enough current to charge and discharge their input
capacit-`
>> ances. The current level to do this is on the order of 10 mA to 100 mA,
dependent
>> on device geometry.
One thing that makes me wonder about MOSFET circuits is that you've got a lot
of gate-drain capacitance to charge (in CD configuration) and this puts a limit
on the output slew rate. If the circuit is not designed carefully, the driver
transistor can be driven into transient overload (and hence generate TID). Of
course, the circuit design can avoid this, but I think a lot of circuits don't.
Chris O'Neill
BYRNES,Graham
[comments by me & others cut]
> extra transistor before the output necessarily produces a "slower" arrangement
> than a MOSFET.
>
> The idea of "slower" doesn't look very useful to me in circuit design. I
> prefer to find out what the poles and zeros are and from these work out the
> circuit behaviour including what happens under feedback.
>
> With a BJT audio amp the output transistor is usually set up in common
> collector configuration so the current gain of this will have a pole at ft/beta
> and a zero at ft. ft might be 1 Mhz and beta 20, hence a pole at 50 khz and
> zero at 1 Mhz.
>
> Now suppose you have a pre-output transistor driving the output in a Darlington
> pair arrangement and that this transistor has an ft of 50 Mhz and a beta of 50.
> In a Darlington pair, the poles in the overall current gain are at ft/beta for
> each transistor and the zeros are at ft for each transistor.
>
> So for the above transistors, you have poles at 50khz and 1 Mhz and zeros at 1
> Mhz and 50 Mhz in the overall current gain. The 1 Mhz pole and zero cancel
> each other so the overall current gain has a net pole at 50 khz and zero at 50
> Mhz.
>
I volunteer my (mis-)understanding so that perhaps we can work out who!
When you say pole, I presume you mean a term 1/(s-u) in the Laplace transform
of the transfer function, which is indeed a pole in the sense of complex
analysis at s=u. However since s=iw, w being frequency, the pole occurs
at a purely imaginary freq, ie there is no real freq where the response
approaches infinity. The manifestation of an imaginary pole is a break-point
in the frequency response, ie something like P=1/(w^2+u^2). Which is certainly
what happens in a simple model of the power gain of a transistor at Ft/beta.
Ok, so if we are talking about the same thing, the type of zero that could
cancel this out would be a Laplace space expression (s-u), with corresponding
power transfer function (w^2+u^2). That is, point where the power transfer
starts to increase (or level out in combination with a pole at lower freq).
Now how can such a thing occur at Ft ? Are you saying that as you push an
emitter follower beyond its Ft, its gain never drops below unity?
If this is the case I'm surprised, but then I'm not an EE...
In fact my impression was that in a real transistor the gain was often < 1
before Ft, which is specced as the extrapolation of the gain curve
measured at a (much) lower frequency.
If this is not the case, and the level out point of the output coinciding
with the breakpoint of the driver can produce a smooth 1st order
response, why is the Ft specified for Darlingtons always so low? Your argument
puts the Ft of the pair at the Ft of the driver!
> Now with a MOSFET, since the input is nearly pure capacitance driven by a high
> output resitance driver circuit, there is a pole at a fairly low frequency (5
> khz if the driver circuit output resistance is 10 k ohms and MOSFETs total
> equivalent input capacitance is 3 nF).
yes, but who on earth would drive a power mosfet from a 10k impedance? A more
realistic source Z would be 200, and less is very plausible.
The capacitance that is mainly
> responsible for shorting the input current to a MOSFET in common drain
> configuration (assuming that is what is being used) is the gate-drain
> capacitance. In this case the overall current gain just has the pole at 5 khz,
> and no zeros. If the transconductance of the MOSFET is 1/(0.5 ohms), the
> current gain from MOSFET input to MOSFET output drops to 1 at 100 Mhz.
The capacitance I usually see specced is the gate-source cap. (unless I'm
confusing drains and sources again :-)). Is 3nF realistic for G-D?
>
> The overall comparison is that the MOSFET gives more gain in the audio
> frequencies than the transistors, but that the transistors in Darlington pair
> have just one significant pole just like the MOSFET. The cancelling pole and
> zero in the Darlington pair example here might not exactly cancel in real
> examples, but as long as the zero is below the pole or not too far above it,
> the pole won't cause any stability problem.
>
> So I don't think it's necessarily true that MOSFETs reduce stability problems
> just because they have far more current gain per device than transistors.
> However, allowing a simpler circuit might reduce problems with stray
> capacitance and lead inductance (might :-).
>
> >> MOSFETS only require enough current to charge and discharge their input
> capacit-`
> >> ances. The current level to do this is on the order of 10 mA to 100 mA,
> dependent
> >> on device geometry.
>
> One thing that makes me wonder about MOSFET circuits is that you've got a lot
> of gate-drain capacitance to charge (in CD configuration) and this puts a limit
> on the output slew rate. If the circuit is not designed carefully, the driver
> transistor can be driven into transient overload (and hence generate TID). Of
> course, the circuit design can avoid this, but I think a lot of circuits don't.
>
> Chris O'Neill
It's pretty common to have the max slew rate specced these days. Even
Electronics Australia (local hobby magazine, for those outside oz) managed
to put together something capable of 50V/uS without too much drama. Of
course if most of the capacitance is G-S, the slew rate can be a lot lower
into a capacitive load.
Regards,
Graham B
Chris O'Neill
Yes, sorry I was being a bit jargony. When I said a pole at 1 Mhz in this
context, I meant a pole on the negative real axis that produces a break point
at 1 Mhz in the frequency response.
>The manifestation of an imaginary pole is a break-point
>in the frequency response, ie something like P=1/(w^2+u^2). Which is certainly
>what happens in a simple model of the power gain of a transistor at Ft/beta.
>
>Ok, so if we are talking about the same thing, the type of zero that could
>cancel this out would be a Laplace space expression (s-u), with corresponding
>power transfer function (w^2+u^2). That is, point where the power transfer
>starts to increase (or level out in combination with a pole at lower freq).
>
>Now how can such a thing occur at Ft ? Are you saying that as you push an
>emitter follower beyond its Ft, its gain never drops below unity?
>If this is the case I'm surprised, but then I'm not an EE...
Yes, this is what I'm saying. You might get an idea from the fact that the
current gain of an emitter follower is 1+hfe, so if hfe goes to zero, you still
have a current gain of 1.
In physical terms, above ft, the base-emitter diffusion capacitance is shorting
out the signal current so the base and emitter are effectvely shorted together.
The only other connection to the transistor, the collector, is isolated from
the base and emitter by relatively small reverse-biased junction capacitances,
so overall the transistor just looks like a shorted base and emitter with a
relatively small capacitance to the collector (which is AC ground). This
shorted base and emitter give a current gain of one.
>In fact my impression was that in a real transistor the gain was often < 1
>before Ft, which is specced as the extrapolation of the gain curve
>measured at a (much) lower frequency.
>If this is not the case, and the level out point of the output coinciding
>with the breakpoint of the driver can produce a smooth 1st order
>response, why is the Ft specified for Darlingtons always so low?
Well, I'm not familiar with the ft of many Darlingtons, the few I know have fts
that aren't real low or real high either, but perhaps in Darlingtons they don't
worry too much about the performance of the individual transistors so they are
starting out with transistors neither of which have really high fts.
>Your argument
>puts the Ft of the pair at the Ft of the driver!
That's right.
>> Now with a MOSFET, since the input is nearly pure capacitance driven by a
high
>> output resitance driver circuit, there is a pole at a fairly low frequency
(5
>> khz if the driver circuit output resistance is 10 k ohms and MOSFETs total
>> equivalent input capacitance is 3 nF).
>yes, but who on earth would drive a power mosfet from a 10k impedance? A more
>realistic source Z would be 200, and less is very plausible.
In the amp circuits I'm thinking of, the driver output is common emitter class
A push-pull with 10 ma quiescient current. What output resistance would you
expect this to have?
But it doesn't matter if this has a high resitance anyway because it just
lowers the frequency of the dominant pole and this doesn't cause any problem.
>>The capacitance that is mainly
>> responsible for shorting the input current to a MOSFET in common drain
>> configuration (assuming that is what is being used) is the gate-drain
>> capacitance. In this case the overall current gain just has the pole at 5
khz,
>> and no zeros. If the transconductance of the MOSFET is 1/(0.5 ohms), the
>> current gain from MOSFET input to MOSFET output drops to 1 at 100 Mhz.
>The capacitance I usually see specced is the gate-source cap. (unless I'm
>confusing drains and sources again :-)). Is 3nF realistic for G-D?
You've made me check the various MOSFET capacitances of the devices I had in
mind because in the above I was assuming that gate-drain capacitance was nearly
as much as gate-source capacitance. This might have been generally true of
MOSFETs 15 years ago but since then pwer MOSFETs have been produced that have
very low gate-drain capacitances compared with gate-source capacitances (e.g.
2SJ48 Cgd = 40 pF at Vgd = 5V and decreases rapidly with increasing Vgd, Cgs =
500 pF). These values mean that gate-source capacitance affects response much
more than gate-drain capacitance, even though in source follower configuration,
the effective input capacitance equals Cgd + (1-K)*Cgs where K is the voltage
gain which is close to one. (The voltage gain for a resitive load R equals
R/(R+1/Gm) where Gm is the transconductance of the MOSFET.)
Hence instead of what I said above, I should have said an effective input
capcitance to the MOSFETs of 3nF * 1/9 = 330 pF. (4 ohm load with 2 1S
transconductance MOSFETs in parallel.)
>It's pretty common to have the max slew rate specced these days. Even
>Electronics Australia (local hobby magazine, for those outside oz) managed
>to put together something capable of 50V/uS without too much drama. Of
>course if most of the capacitance is G-S, the slew rate can be a lot lower
>into a capacitive load.
Forget what I said in the old message above about the slew rate limitation. I
didn't know that the gate-drain capacitance in the devices used was so low.
Regards
Chris O'Neill
Thomas W. Matthews
would like to relate one supposed advntage of MOSFETS that
I haven't seen mentioned here. In at least some designs,
the MOSFET output drivers do not need overcurrent
protection. It has been argued that the extra
current sense resistor and switch transistor that
comprise the overcurrent protection circuitry degrade
performance.
The MOSFETS apparently do need protection from overheating,
hence there are temperature sensors on the heat sinks of my
Hafler DH-220.
Has anyone looked into this? And what about the topic
of overcurrent protection in general?
Tom Matthews
Kirk Lindstrom
>emitter follower beyond its Ft, its gain never drops below unity?
>If this is the case I'm surprised, but then I'm not an EE...
>
>before Ft, which is specced as the extrapolation of the gain curve
>measured at a (much) lower frequency.
>
----------
Actually, transistors don't roll-off as nicely as the model suggests.
In most cases, they have another parameter called Fmax which can be
10 to 40% higher than Ft. In short, the gain can still be higher than 1.0
beyond Ft. Using the simple model for transistor gain-bandwidth works
well for transistors running with gains over 10, but it starts to "tail
away" from simple theory well before GBW=1.
Kirk out
JOHN I LIPP
> the TO-220 encapsulation without any trouble, but I guess that large temp
> variations could have an effect ( read:does have an effect ) on an accuratly
> set bias level, so for max-fi ;-) the temp should be held constant...
>
> like any silicon device, MOSFETS CAN be overheated, but they do not have
> the thermal-runaway or SOE problems of BJT'S..
>
> ------------------------------------------------------------------------------
> stu...@ccu1.aukuni.ac.nz
>
>
> >>>>More, Louder, Faster<<<<
> ------------------------------------------------------------------------------
The reason for tempurature sensor may be bias stabalization (vs. tempurature).
This is a pretty standard part of amplifier design; the sensor is simply a
transistor coupled to the heat sink so that its temperature and the associated
parameter changes track those of the output transistors and can be used to
"cancel" them.
It is also possible that it is part of a protection scheme; the amplifier gets
too hot and the device switches on to shut-down the amplifier or turns on a fan.
Kirk Lindstrom
>would like to relate one supposed advntage of MOSFETS that
>I haven't seen mentioned here. In at least some designs,
>the MOSFET output drivers do not need overcurrent
>protection. It has been argued that the extra
>current sense resistor and switch transistor that
>comprise the overcurrent protection circuitry degrade
>performance.
>
>The MOSFETS apparently do need protection from overheating,
>hence there are temperature sensors on the heat sinks of my
>Hafler DH-220.
>
>Has anyone looked into this? And what about the topic
>of overcurrent protection in general?
>
>Tom Matthews
I think you make a good point. I've NEVER designed a MOSFET amp, but
I would think that not needing overcurrent protection would be an
advantage in that you don't need to spend money on getting a high
^^^^^^^^^^^
quality current sense resistor. Probably why you see FETs in car
audio. I think first order current regulation can be done with either
-the transformer (for 120V amps) or
-the switcher (for car amps)
Problem with transistors is that as they get hot, the base emitter voltage
drops which draws more current, gets hotter, and so on until all the
current flows thru (get ready mathmaticians) a single point meaning
infinite current densities or (usually) meltdown. Designers (at least
me) put resistors in the emitters of power transistors and gang many
in parallel so they don't have thermal run-a-way. I think a FET just
defaults to the channel resistance so it would be cheaper to get the
same max current.
If done correctly, I don't see how "emitter degeneration" resistors
would make the sound worse since they are actually "linearizing"
the transistor some (thermal and voltage feedback).
Kirk out
JOHN I LIPP
>
>I think you make a good point. I've NEVER designed a MOSFET amp, but
>I would think that not needing overcurrent protection would be an
>advantage in that you don't need to spend money on getting a high
> ^^^^^^^^^^^
>quality current sense resistor. Probably why you see FETs in car
>audio. I think first order current regulation can be done with either
>-the transformer (for 120V amps) or
>-the switcher (for car amps)
ANY output stage needs current protection of SOME sort UNLESS it is designed to
run into a short. In many amplifiers this protection is simply a set of fuses
in the power supply rails (example: Adcom 535).
Why would a current sense resistor have to be high quality (other than maybe
non-inductive)? I don't believe that the tolerance on the thermal overload
level is so small that you need tight component tolerance for current sensing???
Also bear in mind that a current limiting circuit limits current by stopping a
voltage rise in the output stage. If done abruptly, this causes a high-frequency
spike to be generated by the (inductive) speaker load. MOSFETs are very sensitive
to this kind of thing; exceed the rated output voltage and it becomes a diode
(connected to a large capacitor with lots of charge... dead, very dead, MOSFET).
This is the reason for the parallel LR output isoltion network on most amplifiers
(in addition to isolation from capacitive loads). Designs with gentle current
limiting sometimes get away without them (if they can handle capacitive loads).
> [stuff deleted]
>
>If done correctly, I don't see how "emitter degeneration" resistors
>would make the sound worse since they are actually "linearizing"
>the transistor some (thermal and voltage feedback).
More important than linearization, they insulate against device parameter var-
iations and (as you indicate) provide bias feedback. They are a "practical"
solution to biasing stabilization, and with negative feedback, any output
resistance they might add is reduced considerably.
Kirk Lindstrom
>>I think you make a good point. I've NEVER designed a MOSFET amp, but
>>I would think that not needing overcurrent protection would be an
>>advantage in that you don't need to spend money on getting a high
>> ^^^^^^^^^^^
>>quality current sense resistor. Probably why you see FETs in car
>>audio. I think first order current regulation can be done with either
>>-the transformer (for 120V amps) or
>>-the switcher (for car amps)
>
>ANY output stage needs current protection of SOME sort UNLESS it is designed to
>run into a short. In many amplifiers this protection is simply a set of fuses
>in the power supply rails (example: Adcom 535).
>
>Why would a current sense resistor have to be high quality (other than maybe
>non-inductive)? I don't believe that the tolerance on the thermal overload
>level is so small that you need tight component tolerance for current sensing???
>
Well, let me think.
- Use a cheap 20% resistor and your max o/p current varies by the same
amount (+/- 20%) - tough to get high ratings on the bench with normal
production stuff.
-You mentioned the inductance problem. If you don't use wirewound the
resistor could be fairly large. Board space is expensive (VERY!)
My main thinking was that, say for a $200 30x4 WPC car amp like the rf460sd,
even adding 4 $0.10 resistors eats alot of profit margin.
The way I do it with LED and/or laser drivers on 8 GHz Ft ICs is to parallel
many big driver transistors and "feather" the emitter metal of each to add
a few tenth Ohms and not cause "hot spots". This is VERY, VERY expensive
as it uses alot of expensive silicon. (I'm also talking about a driver stage
that puts out 160mA into a short in 400 picoseconds so these extreme
techniques may not be required for audio - but the ideas often scale since
the hard part is designing for high dI/dt {and low side-effects!}).
>Also bear in mind that a current limiting circuit limits current by stopping a
>voltage rise in the output stage. If done abruptly, this causes a high-frequency
>spike to be generated by the (inductive) speaker load. MOSFETs are very sensitive
>to this kind of thing; exceed the rated output voltage and it becomes a diode
>(connected to a large capacitor with lots of charge... dead, very dead, MOSFET).
>This is the reason for the parallel LR output isoltion network on most amplifiers
>(in addition to isolation from capacitive loads). Designs with gentle current
>limiting sometimes get away without them (if they can handle capacitive loads).
How does a parallel LR o/p isolation ckt work?
Also, I've designed (again all on silicon) regulated output stages and I
usually sense the current and then lower the gain further back in the
overall gain chain to keep radiated emissions and ringing a minimum. I'd
assume audio designers would do the same. Is that what is meant by "soft
clipping" circuits?
>/***************************************************************************\
>* John Lipp * Loudspeaker design pseudo-expert and software developer *
>\***************************************************************************/
Kirk out
=> "We are what we pretend to be." - Kurt Vonnegut Jr.
+---------------------------------------------------------------------+
| Kirk Lindstrom - OCD Product R & D | Hewlett-Packard Co. M/S: 91UA |
| Engineer/Scientist, Hardware | |
|------------------------------------| Optical Communication Division |
| kirk_li...@sj.hp.com | |
| Kirk Lindstrom / HP0100/UX | 370 W. Trimble Rd. |
| ph 408 435 6404 | fax 408 435 6286 | San Jose, CA 95131-1096 |
+---------------------------------------------------------------------+
lenahan,grant f
> In article <1993Mar8.2...@ccu1.aukuni.ac.nz>, stu...@ccu1.aukuni.ac.nz (Stuart Woolford) writes:
> > >The MOSFETS apparently do need protection from overheating,
> > >hence there are temperature sensors on the heat sinks of my
> > >Hafler DH-220.
>
> This is a pretty standard part of amplifier design; the sensor is simply a
> transistor coupled to the heat sink so that its temperature and the associated
> parameter changes track those of the output transistors and can be used to
> "cancel" them.
No, the DH-200/220 sensors are for overheadting protection. They cut
the mains power.
>
> It is also possible that it is part of a protection scheme; the amplifier gets
> too hot and the device switches on to shut-down the amplifier or turns on a fan.
Basically, yes.
Within the normal range of DH-200/220 bias, and given the existing heatsinks,
the MOSFETs used have a linear TC to NTC. Ruaway does not occur.
Grant