Maker Pro
Maker Pro

Audio amplifier beginnings, part II

J

Joel Kolstad

Jan 1, 1970
0
So here's the current schematic: http://oregonstate.edu/~kolstadj/amp3.gif .
I replaced the 'resistive' current source with a 'real' one and added a
potentiometer for tweaking the bias. (Somehow I prefer two diodes and a pot
to a Vbe mutiplier.)

It's certainly getting a lot better, I think. Remaining questions:

-- The voltage gain seems a little difficult to predict correctly. The
current source should have an output impedance of gm*Ro*R2, and with
gm*Ro=Va/Vt, and the data sheet saying a 2N3906 has an Early voltage (Va) of
40V, I get 40/26m*56=86k. This would predict, then, a gain of ~86k/100 (R3
being the 100 ohm resistor), or 860. SPICE simulations show a gain of ~325.
Not that I'm too wirred -- any gain over about 10 is fine with me.

-- Biasing Q5 correctly is annoying, involving tweaking the exact valve of
R7. Even with a potentiometer, I doubt this is a reliable way to build the
actual circuit? (I am planning to physically build this thing shortly.) I
could set up a current mirror off of Q6 to bias Q5, but I'm thinking that
now's the time to replace the input with a differential pair so that the
feedback will automatically bias Q5 appropriately, and it's only a small
number of additional transistors as well.

-- I still don't see where The Art of Electronics gets their '20%' number at
the bottom of page 93 from. :)

Thanks for all the help so far!

---Joel Kolstad
 
W

Winfield Hill

Jan 1, 1970
0
Joel Kolstad wrote...

Better, but still not great. You want R3 to be half the value
of R2, think about it. This puts a lot of pressure on the bias
scheme for Q5. Add a diode-connected transistor in series with
R7 and change R7's value appropriately. For that matter diode-
connected bd139 and bd140 can replace D2 and D4.

OK, now look at your no feedback circuit. What determines its
gain? What keeps it biased at 0V out for 0V in? So you've got
a lot of fixen' up to do, and we haven't even started talking
about its high distortion levels...

Thanks,
- Win
 
W

Walter Harley

Jan 1, 1970
0
Joel Kolstad said:
..

Joel, you're doing a lot of reinventing of the wheel from first
principles... the topology and design issues of a class-AB moderate-power
amplifier is probably one of the best-explored areas in all of analog
electronics. By all means do the experimentation, but be aware that the
answers you're looking for (and the wrong paths you're going down) are
well-documented in many books, including the one that I referenced earlier.
You would be really well-served by reading one. At the very least, grab a
schematic for one of the many high-fidelity project audio amps on the web,
study it, and try to understand what it has that yours doesn't and why they
bothered.

One reason to use a Vbe multiplier rather than the diodes + current source +
pot solution that you have is that it's easier to thermally couple it to the
output transistors. With poor thermal coupling, it is very hard to control
your bias properly.

You mention wanting to take feedback through a differential input pair, as a
way of controlling the bias. I think you'll find it doesn't achieve that -
you can control DC offset that way, but not quiescent current. The feedback
will, however, give you a substantial improvement in linearity and
bandwidth. (Of course, you'll need to add frequency compensation to avoid
oscillation.)

By the way, you do realize that with the input stage shown, you've got a
high-pass rolloff at about 65Hz? That's another thing that a differential
input stage can help fix for you.
 
J

Joel Kolstad

Jan 1, 1970
0
Hi Walter,

Walter Harley said:
Joel, you're doing a lot of reinventing of the wheel from first
principles... the topology and design issues of a class-AB moderate-power
amplifier is probably one of the best-explored areas in all of analog
electronics. By all means do the experimentation, but be aware that the
answers you're looking for (and the wrong paths you're going down) are
well-documented in many books, including the one that I referenced
earlier. You would be really well-served by reading one.

I haven't read any of the books solely on audio amplifier design, but I have
gone through about three different circuits books (TAoE, Sedra & Smith, and
Analog Electronics by T. E. Price). I'm finding there's a big difference
between just reading a book and analyzing circuits presented vs. designing
your own -- after all, someone had to figure out just what they all the
exact component values needed to be, even if they were using a tried and
true topology. I'd also comment that -- other than TAoE -- the circuits
books emphasize items such as efficiency and gain calculations rather than
what's necessarily a practical implementation (see
http://oregonstate.edu/~kolstadj/amp2.gif , given in Price's book... I
suppose it's anyone's guess whether or not it's meant to be a workable
circuit, but it has the same problem with distortion that my first circuit
did).
At the very
least, grab a schematic for one of the many high-fidelity project audio
amps on the web, study it, and try to understand what it has that yours
doesn't and why they bothered.

I have been to a large handful, and it is clear that the likes of Elliott
Sound Products (Rod Elliott) and The Amplifier Institute (Douglas Self) are
run by folks who have a very, very good understanding of how amplifiers
ought to be designed and could whip up something better than many commercial
offerings on their lunch break on the back of a McDonald's napkin. On the
other hand, there are also some web sites that are just 'circuits'
collections, and it's a lot less clear whether or not the circuits in
question are well-designed.

My basic approach here has been to start with the topologies that are well
known, and I've been asking questions when it's unclear to me exactly why
something is the way that it is. I think you'll see that it's very unusual
for me to just post a question without also taking a stab at a plausible
answer.
One reason to use a Vbe multiplier rather than the diodes + current
source + pot solution that you have is that it's easier to thermally
couple it to the output transistors. With poor thermal coupling, it is
very hard to control your bias properly.

I was (quite consciously) not planning to thermally couple the bias
circuitry to the output transistors. TAoE has a discussion of this on page
93, including a numeric example, and I find the deviation in bias current
over temperature (with output transmitter emitter degeneration) acceptable.
This is also discussed on Rod Eliott's page where he comments that his
quiescent current was surprisingly high -- something he hadn't noticed, but
that other people had commented on. Although he did produce an improved
version of his circuit, he also mentions that there were many of his
original design produced and the bias current instability just isn't that
big of a deal.

BTW, Rod Eliott's original circuit (http://sound.westhost.com/project03.htm)
uses resistors (rather than a current source) for biasing, making his
amplifier rather non-linear; he then uses global feedback to correct for
this. But it's interesting to me that here, again, we have a guy who
designed a commerical product decades back doing it the 'sub-optimal' way,
and it all still works. I think it's just as interesting to find what sort
of circuit 'hacks' you can get away with -- being fully cognizant of what
you're doing -- as it is to know the 'correct' way of designing a circuit.
You mention wanting to take feedback through a differential input pair,
as a way of controlling the bias. I think you'll find it doesn't achieve
that - you can control DC offset that way, but not quiescent current.

I think you might have misunderstood which bias I'm talking about -- I meant
the bias of Q5, presently set up by R6 and R7, not the bias of the power
transistors. Q6 provides a pretty stiff current source of ~10mA, and --
ignoring thermal variations -- D2, D4, and VR1 (a potentiometer) allow the
quiescent current to be adjusted nicely.
The feedback will, however, give you a substantial improvement in
linearity and bandwidth. (Of course, you'll need to add frequency
compensation to avoid oscillation.)

Yes, I hadn't gotten to frequency compensation yet.
By the way, you do realize that with the input stage shown, you've got a
high-pass rolloff at about 65Hz?

Really? SPICE says the thing works from effectively DC to ~35kHz before
falling off. Capacitor C1 is _one Farad_ :) -- just for the sake of
simulation.

Thanks again for the input!

---Joel
 
W

Walter Harley

Jan 1, 1970
0
Joel Kolstad said:
I haven't read any of the books solely on audio amplifier design, but I have
gone through about three different circuits books (TAoE, Sedra & Smith, and
Analog Electronics by T. E. Price). I'm finding there's a big difference
between just reading a book and analyzing circuits presented vs. designing
your own

Yes. I don't mean to accuse you of intellectual laziness; it's quite clear
from your postings that you're thinking hard about what you're doing, and
doing the relevant theoretical reading and work. I'm just trying to pave
your way by pointing out that there are books that address the practical
implementation issues and gotchas that seem to be presenting you with
stumbling blocks (as they have everyone else who has gone down this road,
which is why the books exist).

I have been to a large handful, and it is clear that the likes of Elliott
Sound Products (Rod Elliott) and The Amplifier Institute (Douglas Self) are
run by folks who have a very, very good understanding of how amplifiers
ought to be designed and could whip up something better than many commercial
offerings on their lunch break on the back of a McDonald's napkin. On the
other hand, there are also some web sites that are just 'circuits'
collections, and it's a lot less clear whether or not the circuits in
question are well-designed.

Quite so. The basic requirement for someone to post a circuit is the
ability to run an ftp client, not the ability to design good circuits.
There are many proudly-presented circuits out there that would never have
worked had they ever been tried. Elliott and Self have indeed put lots of
time into exploring the issues and designing good circuits.

[...] the bias current instability just isn't that
big of a deal.

Depends what kind of distortion levels you're going for and how much spare
heat dissipation capacity you've got, I guess.

BTW, Rod Eliott's original circuit (http://sound.westhost.com/project03.htm)
uses resistors (rather than a current source) for biasing, making his
amplifier rather non-linear; he then uses global feedback to correct for
this.

If you introduce 10% distortion, and you want your amp to have 0.01%
distortion, then you need an open-loop gain of at least 1000x more than the
closed-loop gain at all frequencies of interest, yes? There are some
advantages to having a relatively linear open-loop path.

The current source also helps isolate you from distorted waveforms on the
power rail, I think. You really don't want the signal on the power rail
coupling into your output, because it's a mix of 120Hz hum and
half-wave-rectified output, very ugly.

The later designs of a good designer do indeed generally improve on the
earlier designs. That's what gives me hope :)

But it's interesting to me that here, again, we have a guy who
designed a commerical product decades back doing it the 'sub-optimal' way,
and it all still works. I think it's just as interesting to find what sort
of circuit 'hacks' you can get away with -- being fully cognizant of what
you're doing -- as it is to know the 'correct' way of designing a circuit.

Well, we responders are somewhat hampered by not knowing what your goals in
this project are. Recapitulation of the history of audio amp design?
Improving your soldering skills, or your circuit analysis skills? Building
a good-sounding amp? All legitimate goals, but different. If your goal is
a nice hike, then my offering you a lift is not useful.

This is why I keep pushing Self's book, in any event. He doesn't just say
"here's how to do it"; for each stage, he considers a variety of different
topologies and options that have been historically used, analyzes and
explores each one, and then draws a conclusion as to which one is best. So
you learn a lot about what you can get away with, but also a lot about what
the best current solution is.

I think you might have misunderstood which bias I'm talking about

Yes. My apologies.

Really? SPICE says the thing works from effectively DC to ~35kHz before
falling off. Capacitor C1 is _one Farad_ :) -- just for the sake of
simulation.

Ah, so. Missed that; assumed you were using realistic values throughout,
and thought you meant 1uF. Again, my apologies.

-walter
 
J

Joel Kolstad

Jan 1, 1970
0
Hello,

Thanks for the response; I'll try to make this brief!
[...] the bias current instability just isn't that
big of a deal.

Depends what kind of distortion levels you're going for and how much spare
heat dissipation capacity you've got, I guess.

I've got 'big hokin' heat sinks' that came surplus and didn't come with
specs, but they sure look like they're 50 watters to me. :) (Which is
about what I want going into the load -- it's interesting how class AB
amplifiers have the 'hump' in the power transistor dissipation curve that --
if only you can get over it -- you might approach better than 50%
efficiency...)

After looking at Rod's circuit again, I need to apologize for this claim,
because it's incorrect: He uses a boot-strapped collector load resistor
(described in TAoE on page 97), so his 'current source' is a lot better than
my original one and that in Price's book (which, as mentioned before, was
not necessarily intended as a complete circuit).
If you introduce 10% distortion, and you want your amp to have 0.01%
distortion, then you need an open-loop gain of at least 1000x more than
the closed-loop gain at all frequencies of interest, yes? There are some
advantages to having a relatively linear open-loop path.

Somewhere I was reading that the goal is to start with as linear of an
amplifier as you can before applying feedback, although I can imagine that
for the sake of complexity or cost, sometimes this criteria is partially
ignored.

If I end up with 1% distortion with 50W here, I'll be happy!
This is why I keep pushing Self's book, in any event. He doesn't just say
"here's how to do it"; for each stage, he considers a variety of different
topologies and options that have been historically used, analyzes and
explores each one, and then draws a conclusion as to which one is best.
So you learn a lot about what you can get away with, but also a lot about
what the best current solution is.

It definitely sounds worth reading; I'll pick it up and take a look next
time I'm at Powell's. Although -- as I'm sure is true of most people --
often it helps to at least partially attempt a problem and get a little
frustrated with it prior to looking at detailed explanations of better
methods, problems with yours, etc. Hence why people go back and re-read a
lot of their books from time to time -- lots of new nuanes can be gleaned
with each iteration!

---Joel
 
W

Walter Harley

Jan 1, 1970
0
Joel Kolstad said:
If I end up with 1% distortion with 50W here, I'll be happy!

Set your sights higher than that! A solid-state class AB amp with 1%
distortion will sound noticeably blurry; it won't be an amp you actually
want to use. (Much of the distortion will be crossover distortion, which is
particularly ugly sounding and disproportionately affects moderate listening
levels.)

A reasonable target that will give you an enjoyable device is 0.1% at 1kHz,
driving 10W. (10W is pretty loud.) You should be able to hit that without
too much difficulty, but it will force you to pay attention to things like
proper wiring layout.
 
Top