Linearising FETs: How to choose a FET?

[Joel Kolstad]

I've been looking at the "linearization trick" for getting a FET to behave as
an approximately linear resistor as described in AoE, 2nd ed., pg. 139-140. I
can reproduce the graphs on pg. 141, although I've come to realize that one of
the reasons they look so good is because they only took Vds out to half a volt
(JFET) and 200mV (MOSFET) -- if you run the circuits through SPICE and let Vds
run higher, things aren't nearly as pretty.

In any case, what parameters should I be looking for on a data sheet to
suggest whether a given JFET or MOSFET is a good candidate for linearization?
I had SPICE run through a handful of different randomly chosen FETs, and some
are markedly better than others (the VN13 that AoE uses is among the better
MOSFETs... and while I don't have a VCR2N model, I suspect it would be among
the better JFETs). For that matter, other than preference and gate current,
in general are JFETs going to be better than MOSFETs for this application?
I've been getting that impression...

What I'd like to do is to build some electronically controllable RF
attenuators good to the HF range (30MHz); non-linearities in the resistance
will translate into distortion (the creation of spurs). I was also hoping to
be able to pass some 10dBm, which is 1V peak into 50 ohms, so I was thinking
I'd be biasing the FETs at around 1.5V or so. Should this be pretty doable?

Thanks in advance for advice.

I'll post a representative plot of a linearized FET over on ABSE...

---Joel Kolstad

 

[Tony Williams]

I've been looking at the "linearization trick" for getting a FET
to behave as an approximately linear resistor as described in
AoE, 2nd ed., pg. 139-140. I can reproduce the graphs on pg.
141, although I've come to realize that one of the reasons they
look so good is because they only took Vds out to half a volt
(JFET) and 200mV (MOSFET) -- if you run the circuits through
SPICE and let Vds run higher, things aren't nearly as pretty.

In any case, what parameters should I be looking for on a data
sheet to suggest whether a given JFET or MOSFET is a good
candidate for linearization?

Look for JFETs with the highest Vgs(off).

Siliconix AN73-1 is *the* app note to read first.
It includes an abstract on how to estimate and
reduce thd (from 12% to 0.1% in the given example).

 

[Joel Kolstad]

Siliconix AN73-1 is *the* app note to read first.

Will do. From what I've downloaded, it looks like AN105 is an updated
version (and is a lot easier to find as a complete PDF).

---Joel

 

[Ban]

I've been looking at the "linearization trick" for getting a FET to
behave as an approximately linear resistor as described in AoE, 2nd
ed., pg. 139-140. I can reproduce the graphs on pg. 141, although
I've come to realize that one of the reasons they look so good is
because they only took Vds out to half a volt (JFET) and 200mV
(MOSFET) -- if you run the circuits through SPICE and let Vds run
higher, things aren't nearly as pretty.
In any case, what parameters should I be looking for on a data sheet
to suggest whether a given JFET or MOSFET is a good candidate for
linearization? I had SPICE run through a handful of different
randomly chosen FETs, and some are markedly better than others (the
VN13 that AoE uses is among the better MOSFETs... and while I don't
have a VCR2N model, I suspect it would be among the better JFETs). For
that matter, other than preference and gate current, in general
are JFETs going to be better than MOSFETs for this application? I've
been getting that impression...
What I'd like to do is to build some electronically controllable RF
attenuators good to the HF range (30MHz); non-linearities in the
resistance will translate into distortion (the creation of spurs). I
was also hoping to be able to pass some 10dBm, which is 1V peak into
50 ohms, so I was thinking I'd be biasing the FETs at around 1.5V or
so. Should this be pretty doable?
Thanks in advance for advice.

I'll post a representative plot of a linearized FET over on ABSE...

---Joel Kolstad

I think most MOSFET spice models are not meant to be used in the
subthreshold region. Generally for audio frequencies JFETs seem to be
preferred, with a high Vgs cutoff voltage. They are much lower in noise.
I have here a couple of UHF-attenuators with GaAs-Fets by Ma/com. They work
from DC up to the GHz range.
Download the databook_final.pdf by Interfet Corporation.

 

[Joel Kolstad]

Will do. From what I've downloaded, it looks like AN105 is an updated
version (and is a lot easier to find as a complete PDF).

I took a look... it's the same technique that Win describes in AoE.
Interestingly, AN73-1 (the older paper) has considerably more information
than AN105, although AN105 is handy in that they give a table for modern
replacements of the VCR2N/4N7N.

I really ought to get some sleep now...

---Joel

 

[Tony Williams]

I took a look... it's the same technique that Win describes in
AoE. Interestingly, AN73-1 (the older paper) has considerably
more information than AN105, although AN105 is handy in that
they give a table for modern replacements of the VCR2N/4N7N.

The Siliconix original papers, by authors such as J.S Sherwin
or W.Gosling (Prof William Gosling afair), are always better
than later re-drafts. They give more complete information and
tend to be more scrupulous about providing source references.

 

[Joerg]

Hello Joel,

What I'd like to do is to build some electronically controllable RF
attenuators good to the HF range (30MHz); non-linearities in the resistance
will translate into distortion (the creation of spurs). I was also hoping to
be able to pass some 10dBm, which is 1V peak into 50 ohms, so I was thinking
I'd be biasing the FETs at around 1.5V or so. Should this be pretty doable?

Check out the BF998 dual gate. I did some experiments with it for a
publication where I need a variable resistance. Still have to finish
that publication but then I got a call from a client.

 

[Joerg]

Hello Ban,

I think most MOSFET spice models are not meant to be used in the
subthreshold region. Generally for audio frequencies JFETs seem to be
preferred, with a high Vgs cutoff voltage. They are much lower in noise.
I have here a couple of UHF-attenuators with GaAs-Fets by Ma/com. They work
from DC up to the GHz range.

You can obtain some good models on sites like the Yahoo LTSpice group,
in the files section. And from companies like Philips.

For the GHz range I prefer PIN diodes but they won't do much good under
a few MHz or with large amplitudes.

 

[Jim Thompson]

I don't have a copy of AoE in front of me so I don't know what this
linearising trick is, but I have looked at using FETs in the triode region
(i.e. VGD > VTH) as attenuators in a PI configuration. I think the trick
is to find a small enough FET so that you can get the desired resistance
with a sufficiently large VGS. That way, the conductivity of the FET which
is roughly proportional to (VGS-VTH) will not vary by such a large
percentage during the RF cycle. Another thing which can help linearity is
to put a large resistor in series with the gate, so that the gate voltage
tracks the average of the drain and source voltage as far as RF is
concerned.

HP used to make III-V chips with FET pi-attenuators on them, I have a data
book about them. They have a nice op-amp feedback circuit which operates
on a dummy version of the RF attenuator and which adjusts the gate voltages
so that the attenuation and characteristic impedance are right.

By the way, if you could tolerate using switched stepped attenuators instead
of continuously variable attenuation, then it is much easier to achieve low
distortion, In that case you could switch in and out the attenuator stages
using FST3125 or similar, though the on-resistance will limit the minimum
attenuation that you can get to.

Chris

I vaguely remember some method of adding part of the signal into the
gate path that considerably improved linearity.

...Jim Thompson

 

[Chris Jones]

I've been looking at the "linearization trick" for getting a FET to behave
as
an approximately linear resistor as described in AoE, 2nd ed., pg.
139-140. I can reproduce the graphs on pg. 141, although I've come to
realize that one of the reasons they look so good is because they only
took Vds out to half a volt (JFET) and 200mV (MOSFET) -- if you run the
circuits through SPICE and let Vds run higher, things aren't nearly as
pretty.

In any case, what parameters should I be looking for on a data sheet to
suggest whether a given JFET or MOSFET is a good candidate for
linearization? I had SPICE run through a handful of different randomly
chosen FETs, and some are markedly better than others (the VN13 that AoE
uses is among the better MOSFETs... and while I don't have a VCR2N model,
I suspect it would be among
the better JFETs). For that matter, other than preference and gate
current, in general are JFETs going to be better than MOSFETs for this
application? I've been getting that impression...

What I'd like to do is to build some electronically controllable RF
attenuators good to the HF range (30MHz); non-linearities in the
resistance
will translate into distortion (the creation of spurs). I was also hoping
to be able to pass some 10dBm, which is 1V peak into 50 ohms, so I was
thinking
I'd be biasing the FETs at around 1.5V or so. Should this be pretty
doable?

Thanks in advance for advice.

I'll post a representative plot of a linearized FET over on ABSE...

---Joel Kolstad

I don't have a copy of AoE in front of me so I don't know what this
linearising trick is, but I have looked at using FETs in the triode region
(i.e. VGD > VTH) as attenuators in a PI configuration. I think the trick
is to find a small enough FET so that you can get the desired resistance
with a sufficiently large VGS. That way, the conductivity of the FET which
is roughly proportional to (VGS-VTH) will not vary by such a large
percentage during the RF cycle. Another thing which can help linearity is
to put a large resistor in series with the gate, so that the gate voltage
tracks the average of the drain and source voltage as far as RF is
concerned.

HP used to make III-V chips with FET pi-attenuators on them, I have a data
book about them. They have a nice op-amp feedback circuit which operates
on a dummy version of the RF attenuator and which adjusts the gate voltages
so that the attenuation and characteristic impedance are right.

By the way, if you could tolerate using switched stepped attenuators instead
of continuously variable attenuation, then it is much easier to achieve low
distortion, In that case you could switch in and out the attenuator stages
using FST3125 or similar, though the on-resistance will limit the minimum
attenuation that you can get to.

Chris

 

[Joerg]

Hello Jim,

I vaguely remember some method of adding part of the signal into the
gate path that considerably improved linearity.

Half the signal is fed back into the gate.

 

[Terry Given]

I vaguely remember some method of adding part of the signal into the
gate path that considerably improved linearity.

...Jim Thompson

Jim Williams discusses this trick in one of his books, where he develops
a modern version of the lamp-stabilised wein bridge oscillator

Cheers
Terry

 

[Joel Kolstad]

Hi Chris,

Thank you for your comments; they've been helpful.

I don't have a copy of AoE in front of me so I don't know what this
linearising trick is, but I have looked at using FETs in the triode region
(i.e. VGD > VTH) as attenuators in a PI configuration.

I started looking at MMIC attenuator data sheets, and it seems like it's a
common technique. IP3 usually isn't very impressive, however -- the method is
a lot more applicable to receivers than transmitters.

Another thing which can help linearity is
to put a large resistor in series with the gate, so that the gate voltage
tracks the average of the drain and source voltage as far as RF is
concerned.

Ah, so that's why the MMIC data sheets show gate resistors...

HP used to make III-V chips with FET pi-attenuators on them, I have a data
book about them.

Hittite has something like that:
http://www.hittite.com/product_info/product_specs/attenuators/hmc346ms8g.pdf
.... nice little device (the real information on it is in an accompanying app
note).

By the way, if you could tolerate using switched stepped attenuators instead
of continuously variable attenuation, then it is much easier to achieve low
distortion, In that case you could switch in and out the attenuator stages
using FST3125 or similar, though the on-resistance will limit the minimum
attenuation that you can get to.

I'll keep that in mind... I probably need about 10-12 bits of switching, but
at HF I imagine I can still get away with the PCB area needed to do that.

---Joel

 

[Joerg]

Hello Joel,

I can't see all posts in this thread because my ISP's server dropped
some. If it's a controlled RF attenuator you are looikng for and cost
isn't the first priority look at Mini Circuits. They make stuff like that.

http://www.mini-circuits.com/appnote/an70004.pdf