Maker Pro
Maker Pro

Maintaining a Vbe Multiplier's bias value

J

Jon Kirwan

Jan 1, 1970
0
Sorry, I didn't read the entire message...

However, if you want a stiff multiplier, use a TLV431 instead of a BJT.
Somewhat more expensive, but it'll be VERY stiff.

I'm still in "discrete" mode. For example, I am _less_
interested in opamp topologies and design techniques than I
am in _how_ to design opamps. There is nothing like knowing
the details about how they are designed inside to understand
the gotchas that aren't readily accessible to someone using
them.

A comparison here might be like "using a handgun" vs
"understanding how handguns are designed and built." A
gunsmith requires a very detailed knowledge and while this
level of detailed knowledge may not make them a better
shooter, that knowledge still informs them about the handgun
in ways that most shooters have little idea about. And I
think it prepares them for certain unusual circumstances a
little better.

I'm at the gunsmith level, right now. I am NOT wanting to go
shooting, just yet.
However, you don't really want to hold that value constant. You want the
voltage to compensate for the temperature of the output transistors.
Yes.

You might be able to use a diode to track the temperature change, and then use
that in the feedback loop to compensate the TLV431.

No ICs. I might like to thoroughly _understand_ the internal
design of the TLV431, first. Then I'm willing to use it.
A honking big capacitor, one that has very low impedance at your frequencies
of interest, is probably the best idea I've seen on the thread.

Well, I'm interested in focusing on the crafted design of Vbe
multipliers, right now. I can _always_ slap a cap on
whatever that winds up being, later on. So set that aside.

What also bugs me is how that darned thing is going to
interact with the larger system, eventually. I don't like
ignorantly littering a schematic with poles and zeros and
phase delays where right now I have very little idea right
what then happens when I close the outer NFB loop. I'm still
"in the trenches" and trying to understand each piece in
detail and think at that level. The capacitor is at the next
level above and is outside my "view."

Besides, it doesn't do much for LF. The Z is too high and in
parallel, ignorable.
On a related note, there was an article in a recent EDN about a self biasing
preamp which was kinda cool. Instead of trying to track the difference using
diodes or a multiplier, it used a couple of transistors and an opamp to set
the correct values at the bases of the pass transistors. It was so novel (at
least to me) that I typed it into LTSpice.
<snip>

Okay. I'm going to save it, too. I'm not ready to
assimilate it, of course. But I definitely want it around
when I _am_ ready for it.

Thanks,
Jon
 
J

Jon Kirwan

Jan 1, 1970
0
The link. Do the math, it's a hoax, good only at one current and
temperature pair... besides being Beta sensitive.

...Jim Thompson

Ah! Thanks! I can use lessons like this, too! If I can see
what you see there, then that means something. Very good way
to teach. Will keep your points in mind as I read it.

Jon
 
P

Phil Allison

Jan 1, 1970
0
"Tim Williams"
The general idea is to put the Vbe transistor on the same heatsink as the
outputs, if not glued to a transistor directly.

Unfortunately, for widely mismatched current densities, this doesn't work.

** Huh ??

More gobbledegook presented as fact.

http://webpages.charter.net/dawill/Images/Ampere.gif

In this boringly typical circuit, the 2N3904 Vbe mult. doesn't have enough
tempco to compensate the far beefier (= lower current density??) output
darlingtons.

** Proof by assertion and an isolated example known only to the poster.

Gotta love that on usenet.



..... Phil
 
T

Tim Williams

Jan 1, 1970
0
Bob Monsen said:
On a related note, there was an article in a recent EDN about a self
biasing preamp which was kinda cool. Instead of trying to track the
difference using diodes or a multiplier, it used a couple of transistors
and an opamp to set the correct values at the bases of the pass
transistors. It was so novel (at least to me) that I typed it into
LTSpice. Here it is:

Could you take a screenshot of the schematic?

Tim
 
J

Jon Kirwan

Jan 1, 1970
0
<snip>
Your circuit is an example of collector feedback. Collector feedback does
not work well with large signal swings and it lowers the input impedance. A
lower input impedance means that the drivers will also need a lower output
impedance. The bottom line is that you will not see this bias method used in
a power output stages. Another option is to use emitter feedback but for the
emitter resistors to be effective, they will drop alot of signal output and
waste power, which explains why those resistors are usually very low values,
They have very little effect unless the emitter current is large. At the DC
bias current level, they don't do a thing.
Local feedback (emitter feedback or collector feedback) both create more
problems then they solve for stabilizing the power output stage bias point.
To keep the output stage bias point stabilized, the use of overall feedback
is the standard practise. A simple typical amplifier might have a
differential input stage, followerd by a voltage amplifier, followed by the
power output stage. The output from the power stage is fed back to the
differential input stage. High open loop gain with large feedback is the key
to better stabilization of the operating points. Fix it with feedback is a
term to remember.

This point is made, again and again, so it must be true! And
if it weren't true, why else would opamps have been such a
successful building block?? So I completely buy the idea.

I am still trying to study each part, though. At some point,
I will raise my head a bit above that level and take a larger
look. But I'm not yet prepared for it, as the pieces
themselves are still too fuzzily understood. I want to
quantify those, in detail, before expanding my view. In
doing so, I hope to have a somewhat better understanding of
opamps, themselves, too. Not just from a large scale view,
but also in understanding problems within them and how to
choose among various approaches when struggling with a
specific application in mind.

I take your point. But it doesn't change, at all, my
interest in seeing what can be reasonably done with at teh
Vbe multiplier level to accomodate variations in current.
That remains interesting in and of its own right.
The typical bias chain using diodes can be made with resistors as well, but
diodes have the advantage of dropping the bias voltage while having a lower
impedance to the signal.

That makes sense.
Sometimes you will see those bypassed with a large
cap if the impedance causes to much signal loss.

I had thought of that, as well. Though, of course, I hadn't
put values to it.
Diodes can also offer temperature compensation.

Their Eg and N would seem to suggest some difficulties with
curve matching, but I agree broadly.
In any case, an output stage will have way more
current flowing in that bias chain than is actually needed as base bias
current.

This seems to argue with something I read John L. saying, but
I didn't accept it (or reject it), yet. I will need to get
there in due course. But I'm still taking pieces one at a
time.
The voltage drops developed in the bias chain will not be greatly
affected by changes in the base emitter junction because the base bias
current is small compared to the current in the bias chain.

I read this sentence a few times to try and make sure I
followed it well. If I do, and I may not, I think I
addressed this when I tried to calculate the R_ac figure.

The numbers I come up with for a 5mA "bias chain" current
with 1mA in the base bias current and 4mA in the collector,
come out as around 15 Ohms, or so. If I'm right about that,
it seems almost certain that there is _some_ response to even
modest variations in current through it. A 500uA change
yields a 7.5mV change. When I LTspice it, I get a simulation
that matches what I calculate, too.
And remember,
that " Fix it with feedback " applies here too.

Yes, the mantra is slowly deepening within me.
So variations in the power
supply have a very reduced effect on the bias point. The feedback signal is
a voltage, and enough feedback will compensate to keep the output voltage
offset at zero. It will not compensate for for excessve collector currents
or power dissaption if the offset voltage remains low.

Good point for me to remember!! Thanks.
That is why temperature compensation is used too.

I begin to see, better. Thanks, again.
In the early years of transistors, it was common to see transistor stages
using many of the techniques used with vacuum tubes. Dc coupled amplifiers
were rare, because any bias shift was amplified in further stages. Feedback
was applied locally, and overall feedback had no effect on the DC operating
points.

I seem to recall that vacuum tube amplifiers even let the
consumer modify the global NFB. But, as you say, since it
wasn't so critical to the design that was probably why it was
allowed in the first place. With BJTs, it seems now to me
that global NFB is _so_ important that such things cannot be
left as "tweeks" by some consumer playing with a knob!
The trend now is to stabilize everything with feedback. It works,
and it works well. Unless you are a purist and have some religious reason to
avoid this technique, there is no sense in reinventing the wheel.

It's not a religious reason, unless _learning_ is a religion,
I suppose. I don't mind being told that "one day when you
are ready, you will use global NFB to take care of this." I
can gather and accept it, of course. But I also cannot
believe an amplifier can be designed with bags of random
bolts tossed together and "fixed with global NFB" in the end.
There are parts in there and they need to perform some
intended function to some reasonable approximation. And I am
still working on understanding each piece as well as some
thoughts about various approaches at that level to improve
the ideas.

For example, it's important to understand not just vaguely,
but quantitatively on various scores, how a diff-amp behaves
and why I may want to have a current mirror on the tails. I
don't want to just hear "put a current mirror there" and
learn nothing then about why. Later on, when I'm looking
globally at an amplifier, I can look backwards and say, "Hmm.
That Wilson mirror is great, but I really don't need it. The
bog standard 2-BJT mirror is fine enough." But I want to say
that from _understanding_ the details, not from others merely
assuring me about it.

See the difference?

Meanwhile, I'm still interested in seeing if my quantitative
analysis was correct (or wrong) and if there are some other
topologies for it, other than the two I mentioned, that may
be interesting to look at.

Thanks,
Jon
 
J

Jon Kirwan

Jan 1, 1970
0
Could you take a screenshot of the schematic?

I'll include an ASCII version here:
: R2
: +V = 12V ,------/\/\---------------------,
: | 1k |
: | +V |
: | | |
: | \ |
: | / R1 |
: | \ 1k5 |
: | / |
: | | +V |
: C2 | ,----+ | |
: || 10uF R4 | | | 2N3904| |
: ,------||------/\/\---------+ | | | |
: I| || 100 | | Q4 e>| |/c Q3 |
: N| | | |-------| |
: | | | c/| |>e |
: | | C1 --- | | |
: --- | 10uF--- |2N3906 | |
: - V2 | | | | |
: --- SINE(0 .2 1k) | | | | | C3
: - | C5 | | | | || 470uF
: | | || 10p| | +----+-||----,O
: | +V +---||----+ | | || |U
: | | | || | | | |T
: gnd | | | | | \
: | | | |2N3904 | / R5
: \ ,-------, | | | | \ 8
: / R3 | | | +V | Q1 c\| |<e Q5 /
: \ 1k | +V | | | 2N| |-------| 2N3906 |
: / | | | | |\| | e<| |\c |
: | | |\| | '-|-\ | | | |
: | '--|-\ | | >----+----' | gnd
: | | >-+-----|+/ |
: +-------|+/ |/| LT6234 |
: ,-----+ |/| LT6234 | gnd
: | | | gnd
: --- C4 \ gnd
: --- 1uF/ R9
: | \ 1k
: | /
: | |
: gnd gnd

(This was auto-generated from my LTspice to ASCII program.)

Jon
 
B

Ban

Jan 1, 1970
0
Jon Kirwan said:
Okay. I'll do that if folks here aren't interested at all in
talking about it.
BS, a couple of good answers have come.

Jon, you should read this book, bit torrentwise
Audio Power Amplifier Design Handbook, 4th Ed. - (Malestrom)
by Doug Self, one of the deeper going but still very practical publications,
you'll love it.
Ban
 
J

Jon Kirwan

Jan 1, 1970
0
BS, a couple of good answers have come.

Agreed. We are past that question.
Jon, you should read this book, bit torrentwise
Audio Power Amplifier Design Handbook, 4th Ed. - (Malestrom)
by Doug Self, one of the deeper going but still very practical publications,
you'll love it.
Ban

I just received a copy of the 5th edition, today. I'll
start, though the author says that it assumes a certain level
of prior training. And skimming through, I agree.

Jon
 
J

Jon Kirwan

Jan 1, 1970
0
Agreed. We are past that question.


I just received a copy of the 5th edition, today. I'll
start, though the author says that it assumes a certain level
of prior training. And skimming through, I agree.

Okay!!! It has a great section on Vbe multipliers under
Chapter 15 on Thermal Compensation!! This is helpful. And
it includes a discussion on that collector resistor there and
in Chapter 7, where a chart is presented with various values
for my R3 shown and the curves over current. Nice!! It also
appears, on first glance, to confirm my impressions!! This
is very good.

Jon
 
J

Jon Kirwan

Jan 1, 1970
0
Burr-Brown was famous for using bias compensation like that in the
front ends of some of their operational amplifiers, but I doubt its
efficacy in power output stages.

I am still struggling to understand it. There are very
obvious parts that I completely understand. For example, the
divider for a "center" voltage followed by a unity gain
buffer for drive current compliance. I would guess that the
gain is determined by the NFB resistor divided by the input
impedance, which is mostly R4 in this case... so 10. I see
an opamp looking like an integrator, but I'm frankly
unfamiliar with the 4-BJT arrangement structure and need to
think about that one.
The Burr-Brown scheme is similar to a discussion here a few (seven :)
years ago...

http://analog-innovations.com/SED/IB-Cancellation-WithTwoOpAmps.pdf

I'll download it now and look when I get a moment to engage a
little thought.

Thanks,
Jon
 
J

Jon Kirwan

Jan 1, 1970
0
<snip>
I'm frankly
unfamiliar with the 4-BJT arrangement structure and need to
think about that one.
<snip>

I do "see" the emitter followers of Q3/Q5 on the schematic,
of course. It's the R1/Q1/Q4/C1 parts that I'm assuming is
the bias compensation and is the part I don't follow. The C5
looks like a very lightly applied integrator cap, which I
take is needed to avoid oscillation. And that's about where
I'm stuck.

Jon
 
J

Jon Kirwan

Jan 1, 1970
0
Study up on writing loop and nodal equations and either solving by
simultaneous equations or matrix manipulation.

I'm familiar with Norton and Thevenin and the use of three
different perspectives, branch-current, mesh, and nodal
analyses. I very much prefer to "think" with nodal analysis
and have pretty much set aside the other two approaches, now.
I'm also familiar with matrix solutions and have developed my
own programs for solving them a little easier than my TI
calculator allows for and with better accuracy in difficult
cases. I can also do Laplace, but frankly I have NOT yet
learned the shortcuts often used. So I wind up with pages of
partial fractions in the end, converting back to time domain
with tables, and seeing how things look there.

Being capable at a detailed level does not let me "see" at
the top level, just yet. Sometimes, it takes a while. The
immediate example making this point is where I didn't
_understand_ what the collector resistor _might_ do when I
first saw it in a Vbe multiplier. And I initially tried to
analyze the circuit with it, included, and I realized then
that there was a negative feedback present at the tap-off
point. But it was only when I analyzed the simpler circuit,
without it, and found the approximate R_ac for it that I then
_saw_ that this calculated R_ac closely matched the collector
resistor values I saw in the examples. That then immediately
told me the _why_!!

It's how things sometimes work for me, I guess.

Anyway, I am very glad for the suggested examples.

Thanks,
Jon
 
J

Jon Kirwan

Jan 1, 1970
0
Did something get lost in the ASCII? Otherwise collector-to-collector
as in Q1-Q4 is a no-no... one of those devices will saturate.

...Jim Thompson

I just double-checked. It's just that way in Bob's posted
LTspice schematic. And when I simulate the thing, it
produces a 2V p-p output into R5 from a .2V p-p input. Takes
a few cycles to settle on a DC center level, though.

Jon
 
T

Tim Williams

Jan 1, 1970
0
Jon Kirwan said:
I'm also familiar with matrix solutions and have developed my
own programs for solving them a little easier than my TI
calculator allows for and with better accuracy in difficult
cases. I can also do Laplace, but frankly I have NOT yet
learned the shortcuts often used. So I wind up with pages of
partial fractions in the end, converting back to time domain
with tables, and seeing how things look there.

Find "residues" in your TI, it's exactly what you need. Also, polynomial
factorization, if you aren't using it already.

Tim
 
T

Tim Williams

Jan 1, 1970
0
Jim Thompson said:
Did something get lost in the ASCII? Otherwise collector-to-collector
as in Q1-Q4 is a no-no... one of those devices will saturate.

It's a current mirror, based on hFE instead of Vbe (yuck!). When Q1 or Q4
saturates, bias current (or op-amp current) is diverted to the output
transistors, driving the load.

Tim
 
T

Tim Williams

Jan 1, 1970
0
Jon Kirwan said:
It's not a religious reason, unless _learning_ is a religion,
I suppose. I don't mind being told that "one day when you
are ready, you will use global NFB to take care of this."

There are valid reasons for doing so. High bandwidth is one: when you need
high GBW, you don't have time to wait for the signal to propagate through
the 5 or 10 or 100 transistors you have[1]. Those kinds of systems are
usually built with a lot of low-gain, high bandwidth stages -- the feedback
is local, so the phase shift per stage is small, meanwhile the overall phase
shift (from input to output) can be arbitrarily large. The disadvantage is
there's nothing global to account for distortion or DC offset. However, you
can add low frequency servos to stabilize it, getting the best of both
worlds.

[1] Unless you happen to be Linear Technology. For instance, their LT1016
stupid-fast comparator claims an internal 60GHz GBW which is unity-gain
stable (you can use it as an op-amp). Better not have anything near that
feedback node.

Tim
 
J

Jon Kirwan

Jan 1, 1970
0
Find "residues" in your TI, it's exactly what you need.

That's something I know I have yet to study and so I have NOT
looked them up in the TI manual. I'll need to study and
understand them much better than now, before trying that.
Also, polynomial
factorization, if you aren't using it already.

I use it sometimes when I have the TI nearby. I don't carry
it, everywhere, though. And I do think about problems lots
more places than where I have a TI nearby.

Besides, it's nice to have it in my head and to keep on
working on that part with practice. I remember some of the
conversion table and can actually _do_ some derivations the
hard way when I am lacking both tables _and_ a calculator. I
do like keeping up a personal skill, so that I'm not overly
frustrated when the mood strikes and the rare tools are not
handy at the moment.

I always have a finger, my brain, and some dirt no matter
where I am, unless I wind up chained to some prison wall.
Then I'll have to work on my memory, too. ;)

With your recommendation, it's now off to find a good book on
complex analysis.

Jon
 
J

Jon Kirwan

Jan 1, 1970
0
A demonstration of why not to trust simulators.

Among many other such demonstrations -- and a good part about
why _I_ need to understand these things, for myself.
What does LTspice
show for the voltages at Q1:c and Q1:e? Q4:c and Q4:e?

...Jim Thompson

I know your own preference to include diagrams, rather than
just words. So here is a link:

Forgive the coloring. I didn't mess with it to make it
prettier.

Jon
 
J

Jon Kirwan

Jan 1, 1970
0
TLV431s are very simple. They are a bandgap that sucks current until the
'ref' input voltage matches the bandgap output. I've modeled the TLV431
using the datasheet, and it is a fun exercise.

I'll take a crack at it. AofE also talks about bandgaps and
I think I understand them, given the nice discussion there.
BTW, do you have a link to that cool LTSpice -> ASCII program? I'd forgotten
that you wrote it. I've been laboring over a hot 'andy's ascii' program for
schematics that I already have in LTSpice...

Hehe. Sure.
Well, the thing about these horrible power output stages is that they can
get dicey if they get too hot. When they heat up, the Vbe goes down for
both, which tends to pass more 'shoot through' current, which heats them
more... You get this.

Yes, I _get_ it. I earlier ignored temperature when thinking
about BJT analysis, as a rule. Too much to include all at
once, I suppose. But now I keep it in some part of my mind.
So, using three diodes may actually be better than a single transistor,
assuming that they are thermally coupled with the output devices. Then, the
diodes will have a higher TC than the devices (since there are three rather
than two). So, the output current goes down when it gets hot.

Here is a simulation that shows it. Look at the shoot-through current as the
temperature goes from 0 to 150C:
<snip>

I love the fact that you are shooting these schematics to me.
But I'm not sure what "shoot-through" current to look at.
What I _do_ see is your point about the output current going
downwards as temperature increases, supporting what you
suggest about stacking the diodes 3-deep.

And I like the general lesson in the schematic, too, allowing
you to focus on the output stage. The opamp basically
represents the input stage up through, but not including, the
final stage. I need to do that for other sections as I
proceed around understanding each detail individually and
then together as a whole. Kind of a behavioral thing for
everything but the section under the microscope. I need to
take that approach more often than I do.

Jon
 
Top