Don's source follower

[Don Tillman]

   From: "tomg" <efm3 at mediaone.net>
   Date: Mon, 19 Jun 2000 01:11:13 -0700

   Nifty Don, Looks a little like the fet followers Buchla used in 
   his LDR BP filter. Thanks for sharing.

The followers composed of an N channel FET and a PNP transistor?  If
that's the case, no, it's very different.

   From: "Michael Lloyd" <michaellloyd at lucent.com>
   Date: Mon, 19 Jun 2000 11:40:23 -0400

   Nifty circuit! You asked if any has seen it before. I've seen it in
   my old (from about 1982) harmon kardon CD301 cassette deck. The
   audio circuits (except dolby, of course) used discrete devices
   instead of the monolithic op amps.

Heh-heh, I happen to own a 1982 vintage Harman Kardon CD401 cassette
deck.  It may or may not use the same circuit; I'll check the
schematics... Okay, the line amp uses something similiar, on top is a
NFET/PNP compound and on the bottom is a PFET/NPN compound.  That's
the same basic idea.  Is that what the CD301 does?

   I'm curious; what are the ten features. I don't think I could come
   up with ten :-)

1.

With no load the thing is pretty much distortionless, at least as far
as a basic FET model goes.  Compare that to the standard source
follower (imagine an illustration here) where the square-law FET
transfer curve shows up in the output.  A standard source follower
with a current sink on the drain terminal (imagine that illustration
here) is also distortionless with no load.  But...

2. 

Compared to a source-follower-with-a-current-source circuit, running
with the same quiescent current, this circuit will put out several
times more drive current and have a pretty low offset.

3.

When there is a load, the even order distortion products cancel.  And
since an FET is close to a perfect square law device, almost all the
distortion cancels.

4. 

If the devices aren't matched exactly, the remaining distortion is
probably the least annoying possible.  Perhaps even preferred.

5.

Clipping is graceful.

6. 

With an input resistor, current limiting into low impedance loads
appears to be graceful too.  (I haven't given it a rigorous test yet
though.) 

7.

No common-mode range issues.  No crazy behavior like when you run a
TL071 with the inputs within 1.5 volts of the supply rails.

8.  

It's easy to cascode the FET drains if you think it's a source of
nonlinearity.  I don't know if it is or not in real life situations,
but that could be a potential improvement.

9.

You can replace the two resistors with a pot and zero out any
remaining offset by hand.

10.

The usual "I hate opamps argument": 

In the classic opamp voltage follower the signal has to go through
about 6 transistors, each with an exponential characteristic, each
exponential characteristic generating distortion products of all
harmonics, and the transistor stages generally alternate polarity so
the exponential characteristics multiply, and than a lot of feedback
is applied around the whole thing.  Lots of high order intermodulation
distortion.  

Here there's one transistor stage between input and output.
Distortion is removed by the basic nature of the devices and by
parallel symmetry.

11.

The circuit requires hand matching of the FETs.  I used to consider
hand matching an engineering wart, but lately I've realized that hand
matching enables many interesting topologies that wouldn't be possible
otherwise.

12. 

The output impedance isn't real low, about 1000 ohms.  Not good for
any precision control type of applications, but for audio it's
fine.


Like I said, I'm just starting to use this in a number of designs.  I
don't have enough experience with it to declare it the best thing ever
constructed with four components, but it seems to have a lot of
potential.

I'm imagining having to build up a test device with 50 complementary
source followers in a row to check signal degradation.  But I'm not
really in the mood to do that, so I was wondering if there's any
experience with it around.

  -- Don

-- 
Don Tillman
Palo Alto, California, USA
don at till.com
http://www.till.com