Digging around trying to find some already drawn schematics for the tube audio thread, I ran into some words I'd written about the old Dan Meyer Tiger .01, with my plans for improvement, which I did wind up doing. I have a pcb layout for the new one, but am looking for the schematic file among my older computers (I did do this decades ago...). In my opinion, this is as good as you can do in an audio amp, or as good as has ever made any difference, in both "golden ear" and plain old measuring tests -- I've always been a bit agnostic on that flame war. We used ears and gear both, always. One informs the other -- the gear tells you want to listen for, and the ear tells you what to measure, I just don't see the conflict except as a way to sell magazines. To do the latter golden ear tests, we did something a little novel, which was to hang two very high quality mics (actually we tried several different types) outdoors, and listened to nature and an acoustic band inside through them. We could not find (at the time) any recording media that didn't swamp the errors in even the original Tiger design, much less the new one. In tests, we actually couldn't hear anything wrong at all that wasn't attributable to the microphones themselves.
Here are the words I wrote lo, these many years back just before I embarked on a year or so of tweaking and testing (which was after a long time, years, thinking about it). The basic Tiger design was wondrous enough, and intimidating to attempt to "improve" so it took awhile to actually find ways to make it better. I feel like the end result was close to the pinnicle of my analog design career, which improved all the numbers about 5x or more, and took the thing to stable down to DC and up to 4 mhz at full power.
I note VMike said he had oscillation problems with his Tigers. We did too. Turns out that the ground isolation washer for the input jacks is crucial, lead dress has to be as they described in the kit manual, AND -- you either need long, separate cables to the preamp, or put a couple of large ferrites on them (each one separately). You can't hard ground *anything* together in the original design, it seems.
///////////////////////////////////////////////////////
An Analysis of The Tiger 207 Circut
This document will attempt to analyze the Tiger circuit developed by
Dan Meyer in a tutorial fashion. Much baloney is spouted by "audiophiles"
about amplifier design, and this tutorial is designed to cut out the
unscientific, vague stuff and give a common ground for constructive
discussion. We will explore the circuit for this excellent amplifier
from input to output and bring up points as they occur. A good text
to read before proceeding is the Op Amp Tutorial in National Semi's
Linear Applications Book, which gives the math and theory behind some of
the statements made here.
Most amplifiers need some kind of "conditioning" circuit at the input.
The Tiger is no exception, and has one worthy of note. The input circuit
provides an impedance that goes down with increasing frequency, and never
becomes a high - Q reactance to either end. This helps insure stability in
the driving source, and helps eliminate extraneous high frequencies.
Many amp designers forego such an input lowpass filter so they can show a
pretty square-wave plot. However, most of us don't listen to square waves.
The distortion of most designs increases with frequency, as the loop gain
goes down. In the days when everyone's best source material was vinyl,
a shocking amount of above-20kc energy was often supplied to the power amp,
particularly when mistracking occurred. Why care? Distortion of even a
very good amp may reach 10% or more at 30 kc. While you can't hear that,
any distortion implies inter-modulation, and you CAN hear the resulting
low-frequency products. I believe that this effect is largely responsible
for audible differences in amps that test about the same. It is no doubt
a factor in the effect we all know about where an amp will sound better
with a particular preamp, and a preamp that sounds rotten with one amp
sounds great with another. Note that all the points marked "in gnd" on
the schematic are tied closely on the pc board, and tied to the input jack.
This insures that the negative feedback is isolated from high-current
ground-loop paths and is actually comparing the output with the input. A wire
connects the input jack ground back to the center tap of the power supply,
which is the "star ground" point. More on this later.
Proceeding to the active part of the input circuit, we see a rather
novel configuration. Rather than the usual single differential pair we
have two. There are several benefits. First, the input base currents
will tend to cancel, making it easier to provide good DC performance at
a reasonably high input impedance. Second, this configuration allows for
push-pull drive to be generated, rather than the usual single-ended drive.
This will tend to cancel out all of the even-order harmonic distortion
products right from the start. I'm a strong believer in eliminating
errors at their source, rather than using negative feedback as a cure-all.
" Get it as good as possible first, THEN use NFB to improve it!"
Why are there emitter resistors in the diff pairs? These provide
transconductance reduction in the input stage. A look at the National
Semi article mentioned above produces the surprising conclusion that a
lower input transconductance will allow a higher ultimate slew rate in
a closed loop amplifier. This is because a higher transconductance requires
a lower pole frequency to roll off the gain before phase errors accumulate
and make the loop unstable. The idea is to make the maximum amount of
current available to slew the pole capacitor without producing too much
gain. The LM318 uses this technique and was the fastest slewing op amp
of it's time. The downside of this approach is that resistor mismatch
will cause a dc offset. In audiophile land, we can afford precision
resistors, however. The main reason you want a high slew rate in an
amplifier is to avoid ever slewing. Slewing implies that the input stage
is in saturation (clipping) which will cause distortion. I do not believe
that you need a ridiculously high slew rate for good performance, however.
It's much better to eliminate spurious high frequency inputs, rather
than have response into the RF region. Who needs the possibility of tweeter-
frying high frequency oscillation, which is easier to produce from stray
capacitance at high bandwidths?!
The negative feedback circut is novel in another way. Note that
the large electrolytic cap (ugh) is NOT in the high current path. This
will tend to eliminate any distortion produced by the cap's "ESR". On that
note, why do audiopholes call that part of the impedance which isn't pure
capacitive reactance "ESR"? If resistance was all it was, no one would
worry about it much. Non-linear dielectric absorption in mylar and ceramics,
"diode effects" in electrolytics, etc are the real culprits in "capacitor
sound" and if there is no or negligable signal voltage drop across the cap,
you can successfully use even electrolytics in a good design. The NFB
uses a 47k resistor for the DC portion to match the design input impedance
in an attempt to produce a low DC offset. In the AC portion, we see the usual
voltage divider with a capacitor across it to roll off un-needed high frequency
gain. This amp is actually stable a a closed loop gain of one, which is not
necessary, but is nice, as high power signals have a way of coupling to
input signals, causing oscillation. The less gain, the less chance of this
occurring.
Note the 2.2k resistors in series with the current sources for the
diff pairs. This helps eliminate the "tail pole" caused by capacitance
in the current source transistors. This is also why the bases of these
transistors do not have a series resistor. Adding series resistance at this
point produces a miller integrator with the base collector capacitance, and
one wonders why you see a series resistor so often in lesser designs.
The voltage gain stage after the diff pairs are nothing special, except
that there are two in a push-pull configuration. When the front end is
well balanced these provide a nice constant current across the bias network.
The emitter resistors are larger than usual for this stage, to make input
stage bias current less critical, and to allow the use of small-signal
transistors for higher quality. We might also note that the less variable
the power dissipation in this and the input stages, the less the VBE of
the transistors will shift as the waveform goes up and down. You might not
think of this as an important effect, but it is the reason early opamps
latched up a lot, and can produce effects up to a few hundred hertz. Ever
noticed the "bathtub" shaped distortion curves in many amps? This is the
reason that many amplifiers have a distortion rise at low frequencies.
Now that some fairly sexy transistors are available in TO-220 packages,
we might improve our amplifiers by using them in the "low-level" stages.
This would also make it easier to insure that the transistors in a
differential stage are at the same temperature as each other!
The bias network is a fairly conventional (now!) VBE multiplier.
Dan Meyer intended this to be a kit, and maybe that's why there is an extra
diode on the pc board in parallel with the heatsink temp-sensor diode.
Wires break, diodes that are thermally cycled sometimes open up, and
kit builders sometimes put the parts in wrong. A failure in the bias
net will instantly cook almost any amp, no matter what the protection
circut, so he used an extra-reliable one.
The output stage is one of the places where this design really shines.
Firstly, it is configured to have gain. This immediately makes the job
of the pre-driver stages much easier, as they don't have to swing to the
power supply rails. This reduces the heating effects mentioned earlier as
well. The collectors-together configuration acknowledges the fact that
transistors are most linear in current mode. An emitter follower has that
logarithmic base-emitter voltage to current ratio IN SERIES with your
signal. Talk about a dumb idea! The gain of this stage is set by the
emitter resistor networks in the first driver transistors. Their ground
is referenced to the output connector for the most accurate "measurement"
of what the speaker load sees, and also to keep high currents out of the
input circut. The speaker connector has a seperate wire back to the supply
to implement this. The output stage has a gain of about 3, set by these
resistors. Note the 220 ohm resistors in series with the collectors of the
drivers. This protects the base-emitter junction of the next transistor in
line from excessive current. A darlington stage follows the first driver,
so it can use relatively high impedances, and a smaller, faster transistor.
The darlington's first stage contributes it's output current directly
to the output so it gets there without the delay of a slow output stage.
The 47 ohm series resistors protect the transistor if the prot circuitry acts.
This darlington allows the use of a 100 ohm base turnoff resistor for
the output device to help it turn off fast. Remember that this amp was
designed a long time ago and the transistors of today were not available
then. The MJ802 and MJ4502 used in this design had a then-amazing 2 mhz
Ft, which is still pretty good, but you could still burn this amp out
(just about the only way you could) with a high-frequency square wave.
This is because transistors have a storage time and don't turn off as fast
as you can turn them on. This could result in both output transistors
being on at the same time, a disaster. Looking at the place where the
output transistor collectors come together, we find an unusual pair of
resistors. These serve to isolate the stage from the output somewhat, and
provide some feedback for the biasing system. The .1uf caps across them
prevent a high frequency, high Q circuit from developing with the inductance
of the wirewound resistor. The isolation helps make the amp stable with
a short circuit as some signal can still get back to the first driver.
The principle of making each stage as linear as possible and using a small
amount of NFB in each stage before closing the loop is one I endorse
heavily.
The protection circuit is a conventional volt-amp sensor which shorts
out the drive when limits are exceeded. It's about the simplest thing that
works ok, but I hate them all. If they work, they inevitably don't allow
the full safe area of the output devices to be utilized, and of course,
they produce distortion when activated. My philosophy is to use such
large output devices that the line fuse will blow-- the power supply can't
melt them! With the better current-bandwidth devices now available, this
is possible without having to wimp out the power supply.
The output network has the usual resistor-loaded inductor and RC network
to keep the output stage from seeing extremes of impedance caused by your
( unknown ) load. It works ok, and that's all you need. Since I'm careful,
I sometimes omit this from my systems.
THOUGHTS ON POTENTIAL IMPROVEMENTS
Why try to improve on something that nearly everyone agrees is the
best anyway. ( Hear a tiger before you write that nasty letter! )
We audiophiles have always liked to push the limits, and though amps are
already the best thing in any chain, better is better. The Tiger had
somewhat limited power, and tended to overheat when playing loud into
4 ohms. I have a friend who likes to make 1 ohm ribbon speakers, and
that would never do as a load for a Tiger. Some of us like inefficient
speakers (not me!), and the 60 watts or so from a Tiger just isn't enough.
The old ampzilla (which was worse at low levels) was thought superior
at high levels, but I and several others are convinced that that was an
artifact of the higher power available. ( The AIR distorts at high levels
anyway, see Beraneks ACOUSTICS for Info). I just like to see the heads
turn when I put on a recording of a quarter being dropped, about the only
REAL source material available in these decadent studio-driven days!
Incidently, check out some of the movies available in VHS Hi Fi, which isn't
all that hifi, but at least many of the sounds are real, because the guys
who make them don't know a "studioed" way to make a car crash sound or what-
ever. They just crash the cars and record it. Novel Idea, Huh?
Anyway, I digress. To improve the Tiger circuit I suggest the following:
1. Use better, TO-220 transistors in the front end, and burn more current
to get a faster potential slew rate. These transistors should be bolted
together on a slab of aluminum to ensure temp tracking and a very slow
thermal time constant.
2. Match all the resistors! Dan got better than .01% distortion with 10%
matching......
3. See above for transistors. These should be matched for different
parameters depending on where in the circuit they are. VBE and current
gain ( at the operating current ) are the important param's for the
diff pairs. Hfe should match npn-npn and pnp-pnp, as well as
npn-pnp for the front end to really cancel base current at the input.
The pre-drivers and drivers should match if possible too. These
transistors should also be as fast as possible (really throughout).
Dan used transistors with about a 60 mhz ft, and much better are now
available.
4. For more power, parallel more outputs. The Crown DC-300 does this
successfully with separate emitter resistors for the output transistors,
and it should work here too. Then maybe you could get rid of the protection
circuitry.
5. After the above, you could DC couple the thing throughout, and probably
make the main pole frequency higher for even higher slew rate. I've
thought a lot about making a DIGITAL offset adjuster that would sense
output offset during quiet periods and send a correction signal to a D/A
converter to adjust offset. This would give good DC performance even
in the presence of pre-amp offset and eliminate the nasty caps! I feel
that we can do a good enough job without this, however.
6. (And most controversial!) Change the output devices to FETs. Yes,
it's true that there are no complementary FETs, which is also basically
true of transistors. However, you could match transconductance in a
number of ways, unequal "emitter" resistors, for instance, and FETS
are FAR more bulletproof than transistors. FETs don't have a second
breakdown effect, period. They are also MUCH faster. I can turn a
40 amp fet on and off with < 30 ns risetimes, which is only possible with
delicate RF transistors that cost a LOT more. Remember, it is the
falling loop gain (necessitated by slow outputs) that causes distortion
to rise at high frequencies in most all amps. Here we have a chance
to make everything fast, which means we don't have to roll off the
gain as soon to prevent phase errors from making our amp into an
oscillator. Fets also parallel much better than transistors from a
practical point of view. Incidentally, has anyone seen a .2 ohm pot
anywhere? I thought not, but you could make a tapped resistor out of
pc board track for the purpose of getting rid of even order harmonic
distortion in output stages! Anyway, you would have to make a few
other changes to use FETS in the outputs, such as in the bias
network. You might use led(s) in place of the diode, for instance,
to add some constant (bandgap) voltage to the diode's. You don't
want the bias to go down with temp on FETs the way you need to
do with transistors. Note that the bias feedback inherent in the
Tiger circuit makes this easier to get working. While it's true that
FETs aren't as linear as transistors, especially at the rails, most
people like the "tube" sound anyway. I think the "tube" sound
is partly due to tubes "graceful" logarhithmic clipping, which is
what FETs will give. Also, most tube amps seem to have a better
phase margin than most transistor amps, which means less ringing,
cleaner high end rolloff, etc. And of course tubes are FAST.
Don't try making a 30 mhz transmitter with the output device from
your Transistor amp, but you can with just about any tube! Or any
FET, if you can handle driving it's input capacitance.
I've made some changes in the basic schematic to reflect what you
might do to make a FET version. This must be considered incomplete, but
note the following points:
1. I no longer close the loop to gain of one at HF. ( internal loops too)
2. The bias net has one or more leds in place of diodes. ( I'll test this)
3. No more protection circut.
4. Output darlington now is emitter follower, as FETs won't swing the
rails otherwise. This transistor will need a better heatsink.
5. Some resistor values are different, and more will change during
the optimization process, to reflect the much lower transconductance
of the fets (1.6 mho vs approx 10 mho for the transistor version)
//////////////////////////////////////////////
Anyone have a copy of the original schematic? I probably do, somewhere, but it might take a month of looking (full time!) to find it at this point. I don't even have a clue which of 5 buildings it might be in, and I've never been super organized...
I didn't wind up using big die front end transistors. One reason Dan had that output stage with gain is that high quality low level transistors of the time, and still when I was doing this, tended not to be able to handle the higher voltages needed in a design where the drivers had to swing the rails. I replaced the silly series R's and zener across the front end with real voltage regulators, as the resistors had tended to burn out and create spectacular failures, for one thing, and I was never comfortable with those input rails kind of floating.
One of the big tricks in my improvement was matching of the outputs. At the time (and maybe still) there were no really matched N and P fets, with the N ones always having higher transconductance so....cheap trick...I used a small source resistor in the N side only to reduce the distortion without feedback. I also munged the bias circuit, (and never truly got it right) for the different characteristics of the fets. The result was something that tended to try and heat the outputs up to a fixed temperature, so the bias current was fairly high when at low
levels. When you were cranking the thing for long periods, the heatsinks didn't warm up much more than that, and the bias current went down a little, but still stayed good. I felt at the time that rather than fool with that forever, just go with it -- when one of these was cranked enough for that to matter, you were into air-distortion range (and possible ear damage) and you sure couldn't hear the effect of temporarily reduced bias -- it never went close to crossover distortion anyway. I'll add more here as I dig around and find it -- would be a good thing to capture what's in the older computers in my network (easy now with big new disks in the newer ones) so I can reclaim the space they take up anyway.
That's going to take some backing and forthing because of course the old disks are IDE and not the faster versions of that. Hopefully I have at least one running well enough to
read them and get on the network...
At one point we were running a software development shop here and the "Good for that time" computers tended to accumulate, I must have 20 or so -- now not so amazing, one of them is even a P-II. A couple are worth keeping, as we got the special P-III tualitin chips Intel custom made for the LA School system. They were the forerunner of the chips later used in laptops, and only draw 8-9 watts at 900 mhz or so. They had much faster level 2 cache (full speed) than the normal pentiums of the day, to boot, and it really made a difference.
For a person on solar power, that's pretty sweet if you can stand the fact that you have to use older software that doesn't have today's bloat and the assumption that there are cycles to burn.
But that's another rant -- the stuff we wrote wasn't bloated one bit and wasted no cycles, and so got incorporated into a lot of products out there (still!) because it gave better performance to the products that used it than their competitors could manage. Can you imagine writing in assembly language today? We did when it mattered, and it usually did for signal processing kinds of things, or getting a pentium to be able to eat continuous data flows from even an 8 mhz uP blowing out 115k baud...most PC's would not handle that at the time unless you kept bursts to below the uart fifo size with long times between bursts!