Precision current mirror

[[email protected]]

Most of the fast current-output D/A converters I can find (e.g.
AD9764) can generate only up to 20mA max.

I need to provide up to 40mA, while still maintaining speed (50nS) and
accuracy (>=13 bits).

Any suggestions how can I achieve this? Perhaps a pointer to a
schematic of a an accurate current "amplifier" or mirror?

I have thought of paralleling two D/A converters (and sharing
reference voltages), but perhaps there is a simpler way.

TIA

 

[Tim Williams]

Well, you can certainly make three transistors into a 1:2 current mirror,
and generic transistors will go that fast. Tweaking the full 13+ bits
accuracy out of it is your problem. ;-)

Tim

 

[[email protected]]

Well, you can certainly make three transistors into a 1:2 current mirror,
and generic transistors will go that fast.  Tweaking the full 13+ bits
accuracy out of it is your problem. ;-)

Tim

Tim Williams should know better than to top post.

A Howland current source could provide the accuracy, but you'd need a
very fast amplifier and resistors with very low parallel capacitance
to maintain the bandwidth.

L-trimmed surface mount precision resistors at the resistance levels
you'd need should be good enough, and there are plenty of really fast
op amps around.

Driving the desired current into the source of a fast FET and taking
the constant current from the drain is another popular approach, but
40mA implies a biggish FET, and the capacitances between drain and
source and drain and gate might be a bit too big for comfort at
50nsec. 40mA would not be a problem for most MOSFETs - even the SD214
will take 50mA of drain current, though you might need two in paralle
to keep the gate-to-source voltage down to practical levels.

 

[Tim Williams]

Tim Williams should know better than to top post.

What's that supposed to mean? It was a short (three line) reply to the post
at large, no particular quote needed.

Tim

 

[[email protected]]

What's that supposed to mean?  It was a short (three line) reply to thepost
at large, no particular quote needed.

It was a pretty silly reply which did invite further comment - it
would have been tidier if your comment had ended up below the OP's
request, rather than above it.

 

[[email protected]]

..err..all of the current going into a FET source would be seen at the
drain (assuming zero gate current which in normal operation is valid).
   Therefore, it is useless by itself to "amplify" 20mA to 40mA.

So much should be obvious to the meanest intellect.

I'd started off by recommending the Howland current source, which
allows you to scale a tolerably arbitary input to generate a current
output with close to infinite output impedance. Discussions of the
Howland circuit almost always point out that it is easier to make a
unipolar current source or sink by relying or a FET or a BJT to
isolate the driving amplifier from the load.

The OP's 13-bit precision requirement more or less cuts out BJT's -
the base currents are a little too unpredictable and temperature
dependent to make such a circuit practical - but FETs and MOSFET's are
a real possibility.

I would have thought it was obvious to the meanest intellect that that
the op amp drivng the FET/MOSFET would have to boost the OP's output
up to the 40mA desired, but you have proved me wrong.

Try again.

Try thinking about what has been posted before you start trying to
teach your grandmother to suck eggs.

 

[[email protected]]

Tim Williams should know better than to top post.

A Howland current source could provide the accuracy, but you'd need a
very fast amplifier and resistors with very low parallel capacitance
to maintain the bandwidth.

L-trimmed surface mount precision resistors at the resistance levels
you'd need should be good enough, and there are plenty of really fast
op amps around.

Driving the desired current into the source of a fast FET and taking
the constant current from the drain is another popular approach, but
40mA implies a biggish FET, and the capacitances between drain and
source and drain and gate might be a bit too big for comfort at
50nsec. 40mA would not be a problem for most MOSFETs - even the SD214
will take 50mA of drain current, though you might need two in paralle
to keep the gate-to-source voltage down to practical levels.

Bill, Thanks for the suggestion.
Entering "currents sources" in google produced a lot of faff, but
entering "howland" was rather more specific. That was useful!

However, reading about Howland sources (and if I understand it right)
it appears their precision is overly sensitive on component accuracy,
and possibly unable to provide a 13 bit resolution unless I use
incredibly accurate and matched resistors.

Are there any other current source architectures, names or keywords I
could try?

TIA

 

[[email protected]]

On Wed, 30 Jul 2008 20:04:53 -0700 (PDT), [email protected] wrote:

[snip]

Try thinking about what has been posted before you start trying to
teach your grandmother to suck eggs.

That's not what YOUR grandmother sucks ;-)

Both of them are dead, as someone with any grasp of reality would
expect, and don't suck anything any more.

And I did have two of them - odd as this may seem to somebody with
West Virginia hillbilly in his background.

 

[[email protected]]

Bill, Thanks for the suggestion.
Entering "currents sources" in google produced a lot of faff, but
entering "howland" was rather more specific. That was useful!

However, reading about Howland sources (and if I understand it right)
it appears their precision is overly sensitive on component accuracy,
and possibly unable to provide a 13 bit resolution unless I use
incredibly accurate and matched resistors.

Are there any other current source architectures, names or keywords I
could try?

You can buy 0.1% 15ppm resitors off the shelf from Farnell and other
broad-line distributors. 0.1% is 10-bits.

You can trim a Howland circuit with a strategically placed trim-pot to
do better - 13-bits is probably attainable, though soldered-in
parallel trimming resistors would be more stable.

Farnell also stocked really tight-tolerance resistor arrays with very
close tolerances on the initial matching and the temperature matching.
I've not got access to a Farnell catalogue at the moment and I can't
be bothered searching their web-site.

You should have found Bob Pease's application note

http://www.national.com/an/AN/AN-1515.pdf

http://www.national.com/an/AN/AN-20.pdf#page=1

gives two - very basic - op amp based current sources that don't
depend on well-matched resistors, both of which use FETs as output
devices. MOSFETs would also serve.

 

[[email protected]]

You can buy 0.1% 15ppm resitors off the shelf from Farnell and other
broad-line distributors. 0.1% is 10-bits.

You can trim a Howland circuit with a strategically placed trim-pot to
do better - 13-bits is probably attainable, though soldered-in
parallel trimming resistors would be more stable.

Farnell also stocked really tight-tolerance resistor arrays with very
close tolerances on the initial matching and the temperature matching.
I've not got access to a Farnell catalogue at the moment and I can't
be bothered searching their web-site.

You should have found Bob Pease's application note

http://www.national.com/an/AN/AN-1515.pdf

http://www.national.com/an/AN/AN-20.pdf#page=1

gives two - very basic - op amp based current sources that don't
depend on well-matched resistors, both of which use FETs as output
devices. MOSFETs would also serve.

Thanks Bill, Yes I am aware of the application notes, and breadboarded
many of them with little success, hence my original posting.
They are either not fast enough, or require too much fiddling and
trimming with components.

However I have found a very neat circuit (fig 12 of Burr Brown
application note AB-165) It only needs two matched resistors and a 1x
gain amplifier.

http://www.desi.iteso.mx/elec/instru/current.pdf

 

[[email protected]]

Well, you can certainly make three transistors into a 1:2 current mirror,
and generic transistors will go that fast. Tweaking the full 13+ bits
accuracy out of it is your problem. ;-)

Tim

snip

I guess Jim could answer this but, if you got 3 transistor in a
package
on a single die I assume, and made a one to two transistor in
parallel
mirror how well would they match and track?

-Lasse

 

[[email protected]]

snip

I guess Jim could answer this but, if you got 3 transistor in a
package
on a single die I assume, and made a one to two transistor in
parallel
mirror how well would they match and track?

-Lasse

The main problem with the simple 3 transistor current mirror is that
it produces a current "sink" from the original current "source"

To produce a matching current "source" from another "source" you need
a more complex arrangement.