Is Leach the best design available?

Good evening everyone

The topic is provocative on purpose, and though I am a newbie here, I want to see if any hard answers are available.

When I knew nothing about amplifiers and was scared to even consider building one, I just passed Leach Amp as "not powerful enough" or something, thinking that after the quoted THD was low enough and S/N ratio didn't look like a problem, all that was needed was power. Staring at Amperes felt profound at that time, compared to just staring at Watts.

I lived some years quite happily with 300W BK modules. After one got minor problems (it had always a little defect that affected the bias current), I decided it's time to grab the bull by the horns and find a design to build. I coped with a small IC kit the long months it took to build the Leachs, and while the project neared completion, I found to my amazement I was into something far more special than I had thought possible.

I like the sound of my finished Leach Amps. I less like the sound of my Seas Emblas (with custom crossover design by Gradient), but before I can afford new speakers, I can just as well enjoy them.

But there's always some insatiable greed inside men (at least I have it) that wants more. I thought for a long time that my hunger for power would only be satisfied if I built QSC USA1310 at home, which luckily never happened. Anyway, I have browsed the 'net and magazines to find if there was some kit that I ought have built more than the Leach. One single design, from which could be said that it is clearly better than Leach. Or if there are modifications the Professor ought to have included if he was really after the best and not just some design.

What I really love in Leach Amp, not a feature of the amplifier itself, is that nobody is marketing it. There are no advertisements, no long wide flattering sales speeches on the Profesor's site which aim at making you want it.

I have no "golden ears" so I just might be imagining that one amplifier is better than another, though sometimes while listening, something just comes to my mind and tells me that "there is _something_ in this sound." I don't know if I could spot it in double-blind tests or compare it to anything.

It's a fact that many good designs use different solutions than the ones used in the Leach Amp. And I am not that green I'd go on asking, which one is better, the one Leach uses or the one used in the other, and why does he, and on the other hand why doesn't he... or "why aren't all the world's amplifiers the same and the best?"

But after all that "incredible imaging, detailed highs, clear mids, deep and powerful bass, unsurpassed presence, stunning audiophile performance, extreme fidelity, shocking resolution, balanced mix of accuracy with sweetness and amazing spatial cues" has settled, there only remains the question, what could be done better?

I am not into making the Leach Amp cost many times more, by using premium, matched and hand-picked parts everywhere. I could build a separate regulated supply for the LTP and voltage gain stage, but would anyone recommend this?

Does Pass A40 (for which I have all parts, even a 1000mm piece of Fischer SK110 - but no chassis - waiting) have something better to offer than Leach? Not anything in the line of sugar, spice or feeling, but something measurable, and if not measurable, at least something that one can imagine or speculate a cause for. A cause that makes sense. There is one for every effect that makes sense.

The same I want to ask about AKSA and every single famous, extraordinary and special kit amplifier or design you can come up with. Are there anything I could pick and mix from it into Leach Amp? Or if that gives better results, do the same the other way around.

I hope you understood what I am after. I am a perfectionist. That's not an easy trait to live with. Sometimes it's no-good and confusing, and it seldom means I'm doing my best at any given time. It's mostly just a nagging fear that I missed an easy and obtainable way to do it better. Maybe that's even a new definition for perfectionism. If I wanted better Leach Amps, I could have matched the voltage gain stage and all current gain stages in addition to the LTP, which the Professor says needs matching anyway. But would I have heard the difference? Would anyone?

Is Leach Amp's internal bandwidth good enough (or GBP hign enough), or could it be better? Does it have high-order distortion low enough? Would I lose anything if I swapped my Leach Amps for AKSA or A40? Or the Elektor IGBT design or modified Crescendo...

If they are all equally (quantitively) good, is there anything you would call "bad" in Leach Amp, or is it equally "good" too? Are they all so good that it comes down to matter of tastes? I find it much easier to believe in supernatural than to believe there exists such thing as a flawless (enough) amplifier. At least after browsing the Aspen site, unless AKSA happens to be it.

More later, maybe

-Kimmo Sundqvist
Perfection, slippery and hard to hold

Kimmo Sundqvist,

There is always likely to be something new and potentially better to build. Thus the chase for perfection is never ending. It is only stopped occasionally by lack of money or regaining ones senses from time to time.

I believe I can say that I can speak from experiance on this issue. I have chassed perfection for years but she is a fast runner and hard to catch.

John Fassotte
Alaskan Audio

Is the Leach design the best available?

Certainly not. And I've been using a Leach design for around
twenty five years.

But it's a practical design that's been thoroughly documented
and debugged. I'm rather pleased with the sound quality of
Leach's current ver. 4.5

I don't know what is "best", just that there are lot of designs
out there with similar topology; amplifier design hasn't changed
much since good complementary power transistors became
available twenty years ago.

I do know that I liked his choice of a nearly perfectly symmetrical
design from input to output, and the generous amount of power supply decoupling for each stage; in that sense I think it is a better design than the ASKA.

It's possible that a pure Class A design might come closest to
"best", but whether it's practical is another question entirely.
I chose to stick with a Class B amplifier design to keep heat
dissipation issues to a sane level.

One major change in the design implementation might be
the choice of output transistors. I'm trying to obtain a small
batch of MJ21193/4 devices which have better current linearity
than the specified MJ15003/4, and are available in TO-3 metal
cases. Does anyone know of a source for these parts?

Beyond that, I'm thinking about how I could design a new amplifier chassis that actually >looks< good, and uses the
2SA1302/2SC3281 or better devices which have even better
current linearity. These are not available in TO-3 cases, so I
can't use them in my existing chassis/heatsink combination.

Another worthwhile area is building a low noise power supply
with good regulation, but that's really just a choice of parts
used rather than basic design. I don't think a regulated power
supply is worth the trouble; the Leach design already has
good power supply rejection.

What about a complementary feedback output stage? That's
not a trivial design exercise, but it might allow some changes
elsewhere in the basic design like making it possible to change
the amount of global negative feedback necessary to achieve
lowest distortion.

As for other designs and tweaks, I can only point to Doug Self
and Randy Sloan and their books on amplifier design for hints
as to what might be possible. But we're already into a range of
low distortion that my geriatric collection of Heathkit test
equipment can't possibly measure, and I can't afford an Audio
Precision test set.

Unless I can come up with a "cheap" Audio Precision equivalent,
I'd be shooting completely in the dark with new designs. And
that's daunting.

Has anyone listened to the LC Audio zero-feedback amplifier?
Damon asks: "One major change in the design implementation might be
the choice of output transistors. I'm trying to obtain a small
batch of MJ21193/4 devices which have better current linearity
than the specified MJ15003/4, and are available in TO-3 metal
cases. Does anyone know of a source for these parts? "

I got my last batch from Newark Electronics. I beleive they are also available from Allied Electronics. I don't remember the price.

"Beyond that, I'm thinking about how I could design a new amplifier chassis that actually >looks< good, and uses the
2SA1302/2SC3281 or better devices which have even better
current linearity. These are not available in TO-3 cases, so I
can't use them in my existing chassis/heatsink combination. "

Motorola still (I think) still makes a sub for the 2SA1302/2SC3281 - the MJ1302/MJ3281. Same die with the nice extended beta at high current as well as high Ft.

"What about a complementary feedback output stage? That's
not a trivial design exercise, but it might allow some changes
elsewhere in the basic design like making it possible to change
the amount of global negative feedback necessary to achieve
lowest distortion. "

I like the idea that the CFP output stage seems to have better linearity. But, I am more comfortable with a "triple" device output stage instead of a "double". I want the load isolated from the voltage gain stage as much as possible. From what I've read and experimented with (the two books you mentioned are excellent, by the way!), triples in CFP configuration may pose stability issues. Also, for class b or class a/b, I'm still wrapping my brain around having both the drivers and the output devices in class b, versus for the emitter follower compound stage having the drivers in full class a all the time and only the output devices change state. Of course, since I really want to build a class a amp next, this is kind of a moot point :)

I wrestle with the same problem as far as distortion measurements. I usually look at distortion waveforms on my scope. I think we need a group discussion about how to work around this issue for DIYers. There was a REALLY nice project in Audio magazine (RIP) some years ago for a very serious analyzer that could be reworked with current tech (using OPA604 op amps and current buffers instead of TL071, for example) and probably have much more performance than we would need. I've also seen some writings in past issues of The Audio Amateur (now Audio Electronics) by Erno Borbely ( I think) for a distortion analyzer. There was a project in Radio Electronics some years back also. I actually made a PCB and partly built this one. In my opinion, it worked very well - it gave consistent measurements with a Audio Precision I had access to at the time. However, it was VERY tough to control the electronical noise in my DIY environment though - halogen lights, misc dimmers, hair dryers, etc. (it's tough being a married DIYer!) don't make for a clean test environment. Finally, reading through the service manual for some of the HP distortion analyzers (I think the guy that did the Audio magazine project referred to doding this also) suggest a few ideas as well. Alas, I can't seem to find my copies of this anymore :-( I can't recommend enough times that service manuals from GOOD test equipment makers like HP and Tek are EXCELLENT sources of material for study!

I'd be happy with a distortion test box that could spot check with high resolution at about 4 frequencies: 50Hz, 1KHz, 20KHz, 75KHz.

Who's game?

Nelson, what do you do when you have something to test (say an idea at home late at night) and the Audio Precision is nowhere around? Wait until later? :)

Maybe we could even ask an analog Guru like Jim Williams at Linear Technology to design a simple, high peformance THD analyzer circuit (I single him out because I remember and oscillator circuit he designed that was claimed to have THD specs in the single digit parts per million. This is incredible!). They might already have such a design sitting around somewhere; I haven't checked their web site and app notes lately. Hmmmmm ...

Give the distortion analyzer project its own thread so people will be able to find it more easily.
I'm interested.
I'd like to reiterate--for those who haven't understood what I've said on these matters--that I'm not against low distortion, per se, just the use of massive quantities of negative feedback to achieve the distortion figures. Just don't pursue low distortion as an end in itself, as you'll usually find that once you reach a certain point sound quality suffers. But up to that point it can be a useful tool.

Improving the end stage

Now that I have used the last of my MJ15003/4, it is good time to think about obtaining some MJ21193/4. Thanks to Damon for the advice.

Browsing ONsemi site I came across a "sister pair" or something, MJ21195/6. Going through the datasheets I began to wonder if using MJ21194 with MJ21195 would be better than just pairing 21193/4 or 21195/6.

It seems that if GBP is more important, using MJ21193/4 wins. They have from half to one MHz more GBP than 21195/6, depending on collector current.

If gain linearity is more important (although this applies only to very high currents), then combining 21194 with 21195 looks like a better choice. 21195 has current gain of 30 at 10A, 21193 has 20 at 10A.

But looking again, 21195 and 21193 are identical up to 5A... so I guess the extra GBP counts more anyway. What comes to quoted THD, 21193/4 and 21195/6 are practically identical.

-Kimmo S.
Grey: what a short posting from you! :) :) :)

It's a difficult and risky business to design a reasonably low distortion amplifier without objective analysis to ensure a design change is really an improvement. And a simple THD analyzer is almost as insufficient because it doesn't directly reveal the spectrum of harmonics and intermodulation products. Though it's what I have to use; at least I can 'see' inside the circuitry however dimly.

And it's a myth that "massive" amounts of feedback are necessary
to produce those low figures in existing designs. Designers are
using less now because other improvements in topology and
implementation make less feedback possible. That's a very
important aspect of Self's approach to amplifier design; clean up
the individual stages, apply feedback as necessary. The hard
part's getting useful amounts of feedback at high frequencies.

Having benchmarked a design objectively and verified its performance and stability, THEN you listen to the dang thing and
decide whether you like the sound. I think it's interesting that Marshall Leach states he prefers the sound of his amplifier without active constant current and mirror sources in the diff amps.

It's possible to design for a specific "sound" as the principle design goal, but I mistrust that approach. I think that's a large part of what the EKSA amplifier is about; it's a relatively simple design and perhaps that's a reason for its sound, but it still uses global negative feedback. Has anyone reviewed the LC Audio amplifier?

All my projects have used bipolar transistors; I haven't touched
MOSFETs at all. IGBTs apparently don't have much application here; I think they have inherantly high crossover distortion because of their turnon characteristics. So for Class B audio
amplifiers, which are the more practical approach for my needs,
I want to focus on finding the best bipolar output transistors and
use them to best effect.

As for the MJ21193/4 and 5/6; I'm still trying to get some through
Active Electronics, but I'll try Newark next. It doesn't appear that
ON Semiconductor followed through with the MJ1302/3281; they
only offer it in a plastic case MJL series, which I cannot conveniently use.

I'm assuming the parts should perform better at high frequency where less feedback is available, and that 1302/3281 and
similar Toshiba devices are probably better still; the latter have
a quoted Ft of 30Mhz.

I'd really like to use the latter in a new physical layout to nearly
eliminate point to point wiring, but it'll have to wait until I've got
money to spare for the new heatsinks and can build a new
chassis around them. That could be a long while off. Sigh.

I did some tweaks to the ver. 4.5 boards: used Black Gate caps
in the diff amp power supplies and a 220 uF N series Black Gate
for the feedback capacitor; I couldn't afford any more than that.
I used polystyrene when I could find them, and silver mica for
all of the picofarad caps. As per Self and Sloan, I used a 1uF
film cap across R36 to improve turnoff of the output transistors.

Everything else was just matching for tolerance and gain, though
I did try strapping together the diff amp transistors with some
heat sink compound to hopefully improve thermal tracking and
thus DC offset.

The power supply is massively dual mono with a total of 240,000
uF, all rails bypassed with about 20 uF of film caps. I've been
experimenting with different types and ways of reducing rectifier
switching noise; that's an on-going project. Transformers are
sealed E-core types, about 300 VA each.

The current version of Leach's design seems more airy and open;
there's better imaging and a larger sound field than before. I
thought there was improvement in detail and background noise
and hum have just about disappeared. I'm still not satisfied with
my system overall, but it has improved.
It doesn't come easily, but I am capable of brevity...
Re: Leach not liking current mirrors, etc. I try to avoid them unless they're necessary. My hypothesis is that it's related to the inherent capacitance and such in real-world current sources; non-linearities, in other words. A resistor is hard to beat when it comes to linearity, but isn't always practical, unfortunately.

Damon Hill,

Thanks for your excellent post. It was truly worth reading.

On the subject of current sources and current mirrors in differential pairs. From my point of view a well designed current source is much preferable to a simple resistor unless the resistor can be made extremely large in value. In most cases this is impractical. Thus a simple resistor will likely not provide optimum performance from a differential pair. Current sources can also provide temperature compensation for the stage.

Aside from current sources I also prefer current mirrors over simple resistors as loads for differential stages because by using them provides a lot of flexibility. One benefit is that maximum gain is obtainable from such a stage when current mirrors are used. However In many cases it is not desireable to have full obtainable gain from a differential stage. Thus we have potentially good and bad benefits at the same time. Excessive gain may have to be reduced back down with excessive amounts of feedback later. This of course leads to many other problems.

One of the other bad things about a differential pair operating at full gain due to the use of a current mirror is that it is very sensitive to loading changes caused by the following stage. It is important that the differential pair to see a constant impedance load at all frequencies that the amplifier has gain at. If not loaded properly there will likely be very large gain changes versus frequency in the differential stage.

The nice thing however about using current mirrors is that the stage gain can easily be adjusted. If less than full gain is desired from a mirrored differential pair it is just a simple matter or padding down the current mirror to some degree. It is thus possible to control the gain of the differential stage easily and set it to a desired amount and still operate the devices at their best currents and from a lower voltage source than a simple resistor load would require for the same amount of gain.

An added benefit of using padding around the current mirror is that it also helps in providing a more constant load versus frequency to the differential pair.

There is a happy medium between gain and feedback. Both of these parameters need to be controlled carefully. As Grey said feedback is a usefull tool. It should not be used to reduce an excessive amount of gain to a reasonable level. A good balance between gain and feedback is one of the prime keys to good sound reproduction. Don't overlook that differential pair for sources of problems. It can cause more hidden problems than we may think.

John Fassotte
Alaskan Audio