[Adam S]

I'm looking for an "mixer" circuit to detect a sine waves DC to 10MHz,
3Vp-p max, and use IF output of 5kHz. Both the LO and the signal inputs
are generated by DDSs. If large harmonics are present on the LO then
there is good possibility they will mix with non-harmonic spurs
generated by the signal DDS. RF mixers are generally square wave LO
driven (or sine waves that fully switch long tail pair transistors) and
therefore will generate lots of harmonics. This is the reason an analog
multiplier is in my mind. LO 3rd harmonic of -40dB is low enough.

Currently manufactured multiplier ICs I have found:

AD835 (expensive, easy to use)
AD633 (cost ok, but only 1MHz)
HA2556 (price ?, difficult to buy)

I looked at a discrete design with total of 19 NPN transistors + 1
op-amp. (see schematic
http://members.optusnet.com.au/~eseychell/pictures/0001.png )

SPice simulation of perfectly matched transistors gives great
results on screen, but I'm very unsure what will happen in practice.
Voltage offsets at each input will cause carrier feedthrough, but these
can be trimmed out with yet more components. Neither power consumption
or supply voltages are a concern.

Apart from the high part count is there any major problems with going
for a discrete analog multiplier ?

 

[Joerg]

Hello Adam,

If it has to be cheap look for Gilbert cell multipliers. The old MC13xx
kind.

But don't expect a whole lot of dynamic range with these. You get what
you pay for ;-)

Regards, Joerg

 

[[email protected]]

I'm looking for an "mixer" circuit to detect a sine waves DC to 10MHz,
3Vp-p max, and use IF output of 5kHz. Both the LO and the signal inputs
are generated by DDSs. If large harmonics are present on the LO then
there is good possibility they will mix with non-harmonic spurs
generated by the signal DDS. RF mixers are generally square wave LO
driven (or sine waves that fully switch long tail pair transistors) and
therefore will generate lots of harmonics. This is the reason an analog
multiplier is in my mind. LO 3rd harmonic of -40dB is low enough.

Currently manufactured multiplier ICs I have found:

AD835 (expensive, easy to use)
AD633 (cost ok, but only 1MHz)
HA2556 (price ?, difficult to buy)

Have a look at the Burr-Brown - now TI - MPY634, which is claimed to
make 10MHz.

http://focus.ti.com/lit/ds/symlink/mpy634.pdf

And I'm surprised that you have missed the very simiar Analog Devices
AD734

http://www.analog.com/UploadedFiles/Data_Sheets/66167428AD734_c.pdf

 

[Stephan Goldstein]

I'm looking for an "mixer" circuit to detect a sine waves DC to 10MHz,
3Vp-p max, and use IF output of 5kHz. Both the LO and the signal inputs
are generated by DDSs. If large harmonics are present on the LO then
there is good possibility they will mix with non-harmonic spurs
generated by the signal DDS. RF mixers are generally square wave LO
driven (or sine waves that fully switch long tail pair transistors) and
therefore will generate lots of harmonics. This is the reason an analog
multiplier is in my mind. LO 3rd harmonic of -40dB is low enough.

Currently manufactured multiplier ICs I have found:

AD835 (expensive, easy to use)
AD633 (cost ok, but only 1MHz)
HA2556 (price ?, difficult to buy)

I looked at a discrete design with total of 19 NPN transistors + 1
op-amp. (see schematic
http://members.optusnet.com.au/~eseychell/pictures/0001.png )

SPice simulation of perfectly matched transistors gives great
results on screen, but I'm very unsure what will happen in practice.
Voltage offsets at each input will cause carrier feedthrough, but these
can be trimmed out with yet more components. Neither power consumption
or supply voltages are a concern.

Apart from the high part count is there any major problems with going
for a discrete analog multiplier ?

Bill Sloman pointed out the AD734, which I believe does 10MHz.

The biggest issue with a discrete design (aside from power and size)
is likely to be caused by device matching. These Gilbert-cell things
work miracles, but are incredibly sensitive to DC parameters. Area
mismatch of the core devices will give rise to second harmonics,
which might be a problem in your system. This is easily trimmed
out but adds still more complexity.

In a recent IC I designed I used some simple Gilbert multipliers and
found that even a 0.5% mismatch produced unacceptable results,
requiring trims. And this was in an IC, where all the transistors are
close together and nominally "identical". It can only be worse in a
discrete design where two parts from the same bag could come from
very different sections of the same wafer, or even different wafers,
won't necessarily be running at the same temperature, etc.

I think you'll have far fewer headaches if you go with the IC. Both
ADI (disclaimer - I work for them) and TI offer parts suitable for
your needs.

HTH

Steve

email address is nonfunctional, please reply to group

 

[Joerg]

Hello Bill,

Have a look at the Burr-Brown - now TI - MPY634, which is claimed to
make 10MHz.

But be prepared to shell out more than 10 bucks a pop.

Regards, Joerg

 

[Adam S]

Have a look at the Burr-Brown - now TI - MPY634, which is claimed to
make 10MHz.

http://focus.ti.com/lit/ds/symlink/mpy634.pdf

And I'm surprised that you have missed the very simiar Analog Devices
AD734

http://www.analog.com/UploadedFiles/Data_Sheets/66167428AD734_c.pdf

Thanks, but I noticed the price is getting kind of steep. The AD835 is
accurate enough for me and is 1/2 price of the AD734. The old Motorola
MC1495 would be perfect, but this device was made obsolete sometime in
the 1990s. Since this is for a low volume application, and the warnings
I get from Stephan Goldstein, I think shelling out money for an IC is a
better option. I could waste a week fiddling around with a discrete
design only to find it with high levels of untrimmable distortion. The
"signal" input to a Gilbert cell is typically at the lower differential
transistor pair (Q6/Q9 in
http://members.optusnet.com.au/~eseychell/pictures/0001.png )

Provided Re of these two transistors is low relative to the emitter
degenerative resistor , R4, then distortion will be minimal. The "LO"
input port however must go through a linear->non-linear then
non-linear->linear translation, comprised of Q2/Q3 and Q5/Q7/Q8/Q9. This
input port will naturally give higher distortion and is probably the
reason why 4 quadrant multiplier ICs specify one input with greater
linearity than the other.

Adam

 

[theJackal]

I'm looking for an "mixer" circuit to detect a sine waves DC to 10MHz,
3Vp-p max, and use IF output of 5kHz. Both the LO and the signal inputs
are generated by DDSs. If large harmonics are present on the LO then
there is good possibility they will mix with non-harmonic spurs
generated by the signal DDS. RF mixers are generally square wave LO
driven (or sine waves that fully switch long tail pair transistors) and
therefore will generate lots of harmonics. This is the reason an analog
multiplier is in my mind. LO 3rd harmonic of -40dB is low enough.

Currently manufactured multiplier ICs I have found:

AD835 (expensive, easy to use)
AD633 (cost ok, but only 1MHz)
HA2556 (price ?, difficult to buy)

I looked at a discrete design with total of 19 NPN transistors + 1
op-amp. (see schematic
http://members.optusnet.com.au/~eseychell/pictures/0001.png )

SPice simulation of perfectly matched transistors gives great
results on screen, but I'm very unsure what will happen in practice.
Voltage offsets at each input will cause carrier feedthrough, but these
can be trimmed out with yet more components. Neither power consumption
or supply voltages are a concern.

Apart from the high part count is there any major problems with going
for a discrete analog multiplier ?

With an integrated product you don't know whats in there and normally
that makes me feel uncomfortable. OK you can check the data sheets
but still I would go the harder and more challenging way. Instead
of worrying about matching 9 or 16 transistors why not adopt the
simpler way used in the older radios of single active devices.The
performance wasn't/isn't bad. Some used dual gate Mosfets as mixers
.. With a dual gate config. you get amplification of 2 inputs in a
single stage instead of building seperate single arrays for
amplification. The input from the LCO is mixed with that of your RF
input and output is an amplified mixture of the product. Besides BJT
mixers give a somewhat linear response to the input versus the better
square law characteristic of the MOS.
You could still go for a much simpler single BJT mixer but for the
reasons above, lower intermodulation distortion easier biasing, lower
harmonic content, higher p-p voltage swing etc I'd go for a
MOS/FET.

"Go easy on the whisky"

theJackal

 

[Winfield Hill]

Stephan Goldstein wrote...

Bill Sloman pointed out the AD734, which I believe does 10MHz.
... I think you'll have far fewer headaches if you go with the IC.

I agree. I also recommend the AD734, despite its $22 cost.
Adam S can fill out Analog's form and get two free samples.
http://www.analog.com/en/prod/0,2877,AD734,00.html

 

[[email protected]]

With an integrated product you don't know whats in there and normally
that makes me feel uncomfortable.

Barry Gilbert of Analog Devices has published a number of papers on
"Gilbert cell" mixer's and multipliers.

A search on "Gilbert B." + translinear on scholar.google.com picks up
some 28 references.

You should be able to get pretty fair idea of what goes into the Analog
Devices mulitpliers if you read a few of them. More than enough to make
you feel comfortable, in my expereience.

<snipped bad advice>

 

[Mark]

can you give us more info about your application and why you are
concerned about LO harmonics...

square wave LO drive usually improves the linearity of a mixer...

Mark

 

[theJackal]

Barry Gilbert of Analog Devices has published a number of papers on
"Gilbert cell" mixer's and multipliers.

A search on "Gilbert B." + translinear on scholar.google.com picks up
some 28 references.

You should be able to get pretty fair idea of what goes into the Analog
Devices mulitpliers if you read a few of them. More than enough to make
you feel comfortable, in my expereience.

Ummm....
Reasons why I prefer discrete components.
1) A designer should be familiar with discrete circuit disegn
techniques , something he/she loses when using integrated circuits.
2)You compromise circuit flexibility when you use an integrated
circuit
3)Discrete communication circuits are superior to integrated circuits
in performance.

<snipped bad advice>

Your Proof?
I disagree.
Its a well known fact that MOS/FET are superior to BJT mixers.
Try reading something on that.


"Go easy on the whisky"

theJackal

 

[[email protected]]

Ummm....
Reasons why I prefer discrete components.
1) A designer should be familiar with discrete circuit disegn
techniques , something he/she loses when using integrated circuits.
2)You compromise circuit flexibility when you use an integrated
circuit
3)Discrete communication circuits are superior to integrated circuits
in performance.

Your Proof?
I disagree.
Its a well known fact that MOS/FET are superior to BJT mixers.

For what frequencies?

Try reading something on that.

Such as?

My impression is that the virtues of dual gate MOSFET mixers have
everything to do with the good behaviour of the components at
relatively high frequencies.

The OP wants good mixing - and explicityl wants to minimise spurious
high order product spurs - at frequencies up to 10MHz, A well
compensated Gilbert cell multiplier will do much better than a MOSFET
for this sort of job. He also wants to use fairly high level signals,
which won't work well with a dual -gate MOSFET mixer.

The numbers for the MOSFET would have come from "Transistor Circuits
and Applications" (Prentice-Hall series in electronic technology) by
Laurence G Cowles ISBN: 0139300732.

The ISBN is for the second edition from 1974 - I had the first edition,
until it got soaked when a pipe froze in Cambridge U.K. some twenty
years ago.

 

[Winfield Hill]

[email protected] wrote...

My impression is that the virtues of dual gate MOSFET mixers have
everything to do with the good behaviour of the components at
relatively high frequencies.

The OP wants good mixing - and explicityl wants to minimise spurious
high order product spurs - at frequencies up to 10MHz, A well
compensated Gilbert cell multiplier will do much better than a MOSFET
for this sort of job. He also wants to use fairly high level signals,
which won't work well with a dual -gate MOSFET mixer.

The numbers for the MOSFET would have come from "Transistor Circuits
and Applications" (Prentice-Hall series in electronic technology) by
Laurence G Cowles ISBN: 0139300732.

The ISBN is for the second edition from 1974 - I had the first edition,
until it got soaked when a pipe froze in Cambridge U.K. some twenty
years ago.

Can you still open the pages? Anyway, yes, bipolar transistors in
a balanced Gilbert cell should do far better than a JFET or MOSFET
mixer any day, thanks to the BJT's higher transconductance at any
given operating current. First, the BJT's higher transconductance
acts to reduce the effects of DC imbalance, and second, the BJT's
intrinsic DC offsets are much lower than for JFETs or MOSFETs.
(A caveat must be introduced for high background-noise levels.)

We should point out an excellent high-frequency bipolar Gilbert
cell IC available from Intersil, the HFA3101. The hfa3101 uses
10GHz transistors (10-50mA) that can be used at reduced currents
like 2 to 5mA, and still yield a superior performance at 10MHz.

http://www.intersil.com/cda/deviceinfo/0,1477,HFA3101,0.html and
http://www.intersil.com/data/fn/fn3663.pdf DigiKey stocks the
HFA3101B version at $4.28 each, or $2.48 qty 100. Newark has
them too, same price singles, but charges $3.65 qty 100.

The HFA3102 is also a useful part, among others in the UHF series.
http://www.intersil.com/data/fn/fn3635.pdf

 

[[email protected]]

can you give us more info about your application and why you are
concerned about LO harmonics...

square wave LO drive usually improves the linearity of a mixer...

But since this is multiplying by a square wave, it adds in the products
corresponding to all the odd harmonics of the fundamental frequency of
the LO drive - albeit with excellent linearity.

Someone who wants to cover a large range of frequencies might find it
inconvenient to have to filter them out.

 

[Winfield Hill]

Winfield Hill wrote...

[email protected] wrote...

Can you still open the pages? Anyway, yes, bipolar transistors in
a balanced Gilbert cell should do far better than a JFET or MOSFET
mixer any day, thanks to the BJT's higher transconductance at any
given operating current. First, the BJT's higher transconductance
acts to reduce the effects of DC imbalance, and second, the BJT's
intrinsic DC offsets are much lower than for JFETs or MOSFETs.
(A caveat must be introduced for high background-noise levels.)

We should point out an excellent high-frequency bipolar Gilbert
cell IC available from Intersil, the HFA3101. The hfa3101 uses
10GHz transistors (10-50mA) that can be used at reduced currents
like 2 to 5mA, and still yield a superior performance at 10MHz.

http://www.intersil.com/cda/deviceinfo/0,1477,HFA3101,0.html and
http://www.intersil.com/data/fn/fn3663.pdf DigiKey stocks the
HFA3101B version at $4.28 each, or $2.48 qty 100. Newark has
them too, same price singles, but charges $3.65 qty 100.

The HFA3102 is also a useful part, among others in the UHF series.
http://www.intersil.com/data/fn/fn3635.pdf

Experienced analog engineers will recognize the hfa3101 and 3102
part numbers from the CA3101 and '02 parts introduced by RCA in
the early days of linear ICs. RCA introduced a line of transistor
array ICs that gave analog designers access to powerful matched-
transistor design techniques, routinely used by IC designers, but
otherwise not available to the rest of us. For example, the ca3045,
ca3046 and ca3086 were five-transistor NPN arrays, and the ca3096
was an NPN + PNP array with five "high-voltage" 35V transistors.
NSC chipped in with the LM3046, pardon the pun.

As the years went by, GE, RCA and Intersil became owned by Harris,
who eventually tired of them, and spun them off as Intersil.
http://www.intersil.com/products/deviceresults.asp?i=5500&i=5502

Sadly, most of the old high-voltage (e.g. 15V) arrays have been
discontinued, but a few, like the ca3083, ca3086 and the ca3127
(with 1GHz transistors) have remained, although distributors may
not carry them. The LM3046 is still available, but only in SMD.

In place of the lost "high-voltage" arrays we've gotten a few
newer "low-voltage" parts having the same schematic and pinout,
but made with much faster transistors. The original parts had
through-hole DIP packages, well suited for experimenters, but
the new parts are SMD only.

Here's a little table of the array's NPN transistor properties
to jog our memories, and a comparison to the newer parts.

NPN ca3046
spec ca3086 ca3083 ca3096 ca3127 hfa3127
---- ------ ------ ------ ------ -------
DIP? X X X X soic only
Vceo 15 15 35 15 8 V
IC-max 50 100 50 20 65 mA
beta 100 70 390 85 130
f_T 550 450 335 1100 8000 MHz
C_cb 0.58 - 2.25 0.2 0.6 pF (0.3 on die)
C_sub 2.8 - 3.0 1.3 -

Intersil also offers the hfa3134 and hfa3135, which are NPN
and PNP matched pairs with f_T = 8.5 and 7GHz.

 

[theJackal]

[email protected] wrote...

Can you still open the pages? Anyway, yes, bipolar transistors in
a balanced Gilbert cell should do far better than a JFET or MOSFET
mixer any day, thanks to the BJT's higher transconductance at any
given operating current. First, the BJT's higher transconductance
acts to reduce the effects of DC imbalance,

Thats totally OT. There is no DC imbalance in a dual gate MOSFET

and second, the BJT's
intrinsic DC offsets are much lower than for JFETs or MOSFETs.

That is unclear but as far as I know gm/I ratio is lower for a FET
then a BJT.


"Go easy on the whisky"

theJackal

 

[otxi.com]

Hello Adam ,
What exact part number do you need?
I will get you a price.
Thanks Mike
[email protected]

 

[theJackal]

On 19 Dec 2005 16:06:28 -0800, [email protected] wrote:

..

For what frequencies?

Thats a general well known consideration. Read S.A.Maas Microwave
Mixers.
To make a direct circuit comparison is ALWAYS dangerous because they
are too many variables to consider.

Such as?

Depends exactly on what you want. If you need l proof on why MOS/ FET
Mixers are better then their bipolar counter parts Clark Hess 1971
"Communication Circuits" is the place to go.

My impression is that the virtues of dual gate MOSFET mixers have
everything to do with the good behaviour of the components at
relatively high frequencies.

Well at 10 Mhz they should be OK.
They work well at both vhf and uhf I assume.

The OP wants good mixing - and explicityl wants to minimise spurious
high order product spurs - at frequencies up to 10MHz,

One of the 3 possible modes in which you can operate a FET as a mixer
is in resistive mode changing the values of the resistance or more
specifically the conditions of the MOSFET channel between the source
and drain from complete depletion to complete invertion or infinite
resistance to zero resistance . In these conditions the FET channel
resistance is linear and very low intermodulation distortion results.

The book on Clark Hess above gives a detailed analysis using Bessel
functions on why FETs have superior distortion performance compared to
their bipolar counterparts.

A well
compensated Gilbert cell multiplier will do much better than a MOSFET
for this sort of job. He also wants to use fairly high level signals,
which won't work well with a dual -gate MOSFET mixer.

The Gilbert cell multiplier has a differential output current given
by the products of the emitter current, tanh(V1/2Vt) and tanh(V2/2Vt).
V1 and V2 being the The only way you can get an analog
multiplication of the 2 input signals is by assuming V1 and V2 are
SMALL. That is by carrying out the Taylor expansion of the tanh
expression and approximating it as V1/2Vt. Thats why predistortion is
used in Gilbert cells.

Mos have much higher p-p input signal ... compare the mathematical
expression for the output drain current in a dual gate Mosfet .
id = ....+...+ kVg1*Vg2+.... . k is a numerical constant. The
drain current has no limitations as is the case for the Gilbert cell.

The numbers for the MOSFET would have come from "Transistor Circuits
and Applications" (Prentice-Hall series in electronic technology) by
Laurence G Cowles ISBN: 0139300732.

The ISBN is for the second edition from 1974 - I had the first edition,
until it got soaked when a pipe froze in Cambridge U.K. some twenty
years ago.

Try looking at the following .
Tsironis Meierer and R. Stahlmann Dual Gate Mesfet mixers "IEEE
Transactions on Microwave theory and Techniques" Vol MTT-32,No.3
Mar.1984"

S.Weaver "TV Design considerations Using Dual Gate MOSFETS "IEEE
Transactions on broadcast and Television receivers, Vol. BTR 19 , May
1973"

You don't seem to get the point of my original post . The Op wanted
some advice on both discrete and integrated circuit implementation of
a mixer. Well while everyone was saying Gilbert cell I wanted him to
know there were other options ... None of which were mentioned by
anybody! I admitted in my first post they weren't necessarily the
best and it would be challenging to use them... but if I was him I'd
have taken that road.


"Go easy on the whisky"

theJackal

 

[[email protected]]

On 19 Dec 2005 16:06:28 -0800, [email protected] wrote:

.

Thats a general well known consideration. Read S.A.Maas Microwave
Mixers.
To make a direct circuit comparison is ALWAYS dangerous because they
are too many variables to consider.
Depends exactly on what you want. If you need l proof on why MOS/ FET
Mixers are better then their bipolar counter parts Clark Hess 1971
"Communication Circuits" is the place to go.

And he won't be much interersted in mixer performance in situations
where all of the frequencies being mixed are lower than 10MHz. Long
wave and medium wave radio transmission is still of coomercial
significance, but nobody write books about it any more.

Well at 10 MHz they should be OK.
They work well at both vhf and uhf I assume.

They do - that is their particular virtue.

One of the 3 possible modes in which you can operate a FET as a mixer
is in resistive mode changing the values of the resistance or more
specifically the conditions of the MOSFET channel between the source
and drain from complete depletion to complete invertion or infinite
resistance to zero resistance . In these conditions the FET channel
resistance is linear and very low intermodulation distortion results.

But not a low as a good analog multiplier in the frequency range where
the analog multiplier is good.

The book on Clark Hess above gives a detailed analysis using Bessel
functions on why FETs have superior distortion performance compared to
their bipolar counterparts.

But not analog multipliers - right?

The Gilbert cell multiplier has a differential output current given
by the products of the emitter current, tanh(V1/2Vt) and tanh(V2/2Vt).
V1 and V2 being the The only way you can get an analog
multiplication of the 2 input signals is by assuming V1 and V2 are
SMALL. That is by carrying out the Taylor expansion of the tanh
expression and approximating it as V1/2Vt. Thats why predistortion is
used in Gilbert cells.

Mos have much higher p-p input signal ... compare the mathematical
expression for the output drain current in a dual gate Mosfet .
id = ....+...+ kVg1*Vg2+.... . k is a numerical constant. The
drain current has no limitations as is the case for the Gilbert cell.

And the analog multipliers like the AD734 are specified for +/-10V
signals. This is a different ball-park.

http://www.analog.com/UploadedFiles/Data_Sheets/66167428AD734_c.pdf

Try looking at the following .
Tsironis Meierer and R. Stahlmann Dual Gate Mesfet mixers "IEEE
Transactions on Microwave theory and Techniques" Vol MTT-32,No.3
Mar.1984"

S.Weaver "TV Design considerations Using Dual Gate MOSFETS "IEEE
Transactions on broadcast and Television receivers, Vol. BTR 19 , May
1973"

You don't seem to get the point of my original post . The Op wanted
some advice on both discrete and integrated circuit implementation of
a mixer. Well while everyone was saying Gilbert cell I wanted him to
know there were other options ... None of which were mentioned by
anybody! I admitted in my first post they weren't necessarily the
best and it would be challenging to use them... but if I was him I'd
have taken that road.

He wanted a low frequency mixer, and you are giving him advice derived
from microwave and broadcast journals. Win and I are coming out of the
scientific instrument literature, where the OP's low frequency
interests are much better represented.

 

[theJackal]

And he won't be much interersted in mixer performance in situations
where all of the frequencies being mixed are lower than 10MHz. Long
wave and medium wave radio transmission is still of coomercial
significance, but nobody write books about it any more.


They do - that is their particular virtue.

I know that, but they have also been used for lower frequencies in
radios ... from the 60s till today . I 've already said that before.

But not a low as a good analog multiplier in the frequency range where
the analog multiplier is good.


But not analog multipliers - right?

Yes they are superior to analog multipliers too. I wrote to you the
Mathematical reasons why thats so down there . Check up the
references.

And the analog multipliers like the AD734 are specified for +/-10V
signals. This is a different ball-park.

Well you were talking about a Gilbert cell versus a dual gate Mosfet .

http://www.analog.com/UploadedFiles/Data_Sheets/66167428AD734_c.pdf

You just have to open up a radio to see a dual gate mosfet being used
.... Not a datasheet.

He wanted a low frequency mixer, and you are giving him advice derived
from microwave and broadcast journals. Win and I are coming out of the
scientific instrument literature,

Did he say he was building a scientific instrument ? No he did not.
So maybe you are talking about a subject completely new to both of
you.
You don't seem to be aware of the fact that broadcast transmissions
cover 10 Mhz.

where the OP's low frequency
interests are much better represented.

Did you read those articles? NO I guess because they mention use of
the MOS at lower frequencies.
..


"Go easy on the whisky"

theJackal