trimpot guts

[John Larkin]

This is a Bourns 3314G. They work very well, but the hairpins on the
ends add inductance, roughly 7-10 nH, and the substrate is fairly
capacitive, so used as a pot there's a lot of leak-through when the
output is supposed to be zero. With a 100 ohm pot, a 30 ps input step
makes a 70 ps output blip of about 60% amplitude. That corresponds to
a serious amount of transmission in vaguely the 5 GHz sort of area. So
it might be usable to 1 GHz maybe.

ftp://66.117.156.8/3314G_guts.JPG

I'm looking for a better pot, or we might try using a phemt as a shunt
element and control gain with dc gate voltage, like a classic jfet
attenuator. I need gain trimming in a dc-1 GHz+ signal chain, and pin
diodes and most attenuator ic's don't work down to dc.

John

 

[Jeff Liebermann]

This is a Bourns 3314G. They work very well, but the hairpins on the
ends add inductance, roughly 7-10 nH, and the substrate is fairly
capacitive, so used as a pot there's a lot of leak-through when the
output is supposed to be zero. With a 100 ohm pot, a 30 ps input step
makes a 70 ps output blip of about 60% amplitude. That corresponds to
a serious amount of transmission in vaguely the 5 GHz sort of area. So
it might be usable to 1 GHz maybe.

ftp://66.117.156.8/3314G_guts.JPG

I'm looking for a better pot, or we might try using a phemt as a shunt
element and control gain with dc gate voltage, like a classic jfet
attenuator. I need gain trimming in a dc-1 GHz+ signal chain, and pin
diodes and most attenuator ic's don't work down to dc.

John

Does it need to be a rotary, user adjustable, potentiometer, or can
you use a "slide" type of pot for factory adjustments? See:
<http://www.google.com/patents?id=djUlAAAAEBAJ&dq=5339065>
Allen-Bradley (and probably others) make these.

 

[Joerg]

This is a Bourns 3314G. They work very well, but the hairpins on the
ends add inductance, roughly 7-10 nH, and the substrate is fairly
capacitive, so used as a pot there's a lot of leak-through when the
output is supposed to be zero. With a 100 ohm pot, a 30 ps input step
makes a 70 ps output blip of about 60% amplitude. That corresponds to
a serious amount of transmission in vaguely the 5 GHz sort of area. So
it might be usable to 1 GHz maybe.

ftp://66.117.156.8/3314G_guts.JPG

Why pots? I never use pots, creates service headaches and adds cost in
production. What's that shiny surface you placed them on?

I'm looking for a better pot, or we might try using a phemt as a shunt
element and control gain with dc gate voltage, like a classic jfet
attenuator. I need gain trimming in a dc-1 GHz+ signal chain, and pin
diodes and most attenuator ic's don't work down to dc.

Another option can be dual-gates such as the BF998. My favorite used to
be dies such as the SD5400 with several on there because you can
DC-servo them. But then they became expensive and hard to find boutique
parts :-(

Why don't they make something like the CD4007 in high-speed logic?
Depending on your circuit you might be able to use an unbuffered
high-speed inverter/buffer.

 

[[email protected]]

Why don't they make something like the CD4007 in high-speed logic?
Depending on your circuit you might be able to use an unbuffered
high-speed inverter/buffer.

PIN diodes?

 

[Joerg]

PIN diodes?

Nope, John needs to regulate all the way down to DC. Of course, one
could use PIN diodes plus a CD4007 but then it would really become
unorthodox.

 

[[email protected]]

Nope, John needs to regulate all the way down to DC. Of course, one

Oh yeah, "step". My bad.

 

[**THE-RFI-EMI-GUY**]

This is a Bourns 3314G. They work very well, but the hairpins on the
ends add inductance, roughly 7-10 nH, and the substrate is fairly
capacitive, so used as a pot there's a lot of leak-through when the
output is supposed to be zero. With a 100 ohm pot, a 30 ps input step
makes a 70 ps output blip of about 60% amplitude. That corresponds to
a serious amount of transmission in vaguely the 5 GHz sort of area. So
it might be usable to 1 GHz maybe.

ftp://66.117.156.8/3314G_guts.JPG

I'm looking for a better pot, or we might try using a phemt as a shunt
element and control gain with dc gate voltage, like a classic jfet
attenuator. I need gain trimming in a dc-1 GHz+ signal chain, and pin
diodes and most attenuator ic's don't work down to dc.

John

Is there any way to treat the internal structure like a transmission
line? Put a ground plane behind the pot and use stripline to match the
unit to the rest of the circuit?

--
Joe Leikhim K4SAT
"The RFI-EMI-GUY"©

"Treason doth never prosper: what's the reason?
For if it prosper, none dare call it treason."

"Follow The Money" ;-P

 

[Joerg]

Is there any way to treat the internal structure like a transmission
line? Put a ground plane behind the pot and use stripline to match the
unit to the rest of the circuit?

And then a guy at Bourns improves design reliability and ...

 

[Ben Jackson]

I'm looking for a better pot, or we might try using a phemt as a shunt
element and control gain with dc gate voltage, like a classic jfet
attenuator. I need gain trimming in a dc-1 GHz+ signal chain

Are you trying to make 10 units or 10,000 units? If you can afford the
extra time, you could use two parallel SMT resistors, one installed and
one not, and use 1% resistors (glued to toothpicks for handles) to trim
it. When you find the one you need, solder it in. With the right test
rig you could measure the un-trimmed amplitude and compute the value to
insert.

 

[Howard Swain]

Hi John,

This is a Bourns 3314G. They work very well, but the hairpins on the
ends add inductance, roughly 7-10 nH, and the substrate is fairly
capacitive, so used as a pot there's a lot of leak-through when the
output is supposed to be zero. With a 100 ohm pot, a 30 ps input step
makes a 70 ps output blip of about 60% amplitude. That corresponds to
a serious amount of transmission in vaguely the 5 GHz sort of area. So
it might be usable to 1 GHz maybe.

ftp://66.117.156.8/3314G_guts.JPG

I'm looking for a better pot, or we might try using a phemt as a shunt
element and control gain with dc gate voltage, like a classic jfet
attenuator. I need gain trimming in a dc-1 GHz+ signal chain, and pin
diodes and most attenuator ic's don't work down to dc.

John

How about Bourns 3329H?
Note that the leads come straight down out of the pot and are only
0.1" apart.

How many dB range do you need?
Are you doing L pad, T pad, pi pad, or ??
Or changing gain-determining resistors in an amp?

 

[John Larkin]

Hi John,



How about Bourns 3329H?
Note that the leads come straight down out of the pot and are only
0.1" apart.

I actually have some of them, and will try them. I'd prefer surface
mount, to avoid via capacitance, but I guess we could surface-mount
them, too, if we bent the leads.

How many dB range do you need?
Are you doing L pad, T pad, pi pad, or ??
Or changing gain-determining resistors in an amp?

It's just


opamp output----------+
|
|
p
o<-------------- next opamp + input
t
|
|
gnd


where I actually need a "gain" between the opamps of between 0.25 and
maybe 0.8, so I can adjust for photodiode sensitivity. System
bandwidth is roughly 1.2 GHz.

John

 

[Joerg]

I actually have some of them, and will try them. I'd prefer surface
mount, to avoid via capacitance, but I guess we could surface-mount
them, too, if we bent the leads.


It's just


opamp output----------+
|
|
p
o<-------------- next opamp + input
t
|
|
gnd


where I actually need a "gain" between the opamps of between 0.25 and
maybe 0.8, so I can adjust for photodiode sensitivity. System
bandwidth is roughly 1.2 GHz.

1.2GHz with a potmeter?

<goose bumps appearing...>

How about these?
http://www.analog.com/UploadedFiles/Data_Sheets/ADL5391.pdf

Ok, a bit noisy and sets you back five bucks in qties. I you go with the
pot I'd buy tons of them so you don't get a black eye when Bourns
"optimizes" the potmeter design and all your calcs go out the window.

 

[John Larkin]

1.2GHz with a potmeter?

Sure, why not? The 3314G looks like it may work OK at that frequency,
although it would be cool to find something faster. Hey, I could add a
lowpass filter in front of my scope and see how that looks. I have an
870 MHz MiniCircuits gaussian filter around here somewhere. This pot
definitely bleeds through around 3 GHz or so.

<goose bumps appearing...>

How about these?
http://www.analog.com/UploadedFiles/Data_Sheets/ADL5391.pdf

Ok, a bit noisy and sets you back five bucks in qties.

The DC situation looks dicey, too. All you RF guys ignore anything
below a gazillion Hertz.

1 GHz design is easy; DC-to-1GHz is harder.

John

 

[Joerg]

Sure, why not? The 3314G looks like it may work OK at that frequency,
although it would be cool to find something faster. Hey, I could add a
lowpass filter in front of my scope and see how that looks. I have an
870 MHz MiniCircuits gaussian filter around here somewhere. This pot
definitely bleeds through around 3 GHz or so.

Or use a TV amplifier. Most of those roll off rather quickly around a GHz.

The DC situation looks dicey, too. All you RF guys ignore anything
below a gazillion Hertz.

1 GHz design is easy; DC-to-1GHz is harder.

True. That's why I sometimes split DC from the RF path early on. You
mentioned that it is to accommodate photodiode tolerances. Since those
rarely exceed +/-15% couldn't you just use a resistive divider where the
resistor to ground is a series combo of a real resistor and the DS path
of a zippy RF FET? The BF998 comes to mind.

 

[John Larkin]

Or use a TV amplifier. Most of those roll off rather quickly around a GHz.



True. That's why I sometimes split DC from the RF path early on. You
mentioned that it is to accommodate photodiode tolerances. Since those
rarely exceed +/-15% couldn't you just use a resistive divider where the
resistor to ground is a series combo of a real resistor and the DS path
of a zippy RF FET? The BF998 comes to mind.

Yesterday we were playing with a circuit (on a whiteboard) that would
use two gaasfets and an opamp. We would dump some dc current into the
drain of one fet and the opamp would servo its gate voltage to get the
saturated drain voltage to some target. Apply that same gate voltage
to the other fet, and it becomes a stabilized Rds-on to ground, part
of an attenuator. Since the fets on one reel tend to match very well,
that should work.

The little NEC fet we're using elsewhere (NE3509) make a nice variable
resistor. It behaves just about like a super-hot jfet. Rds-on is about
6 ohms at zero gate voltage, 10 at -0.1, 30 at -0.2. Drain capacitance
is almost constant at about 0.35 pF, about the same as a surface-mount
resistor. Helluva part for 85 cents.

But that's a lot of junk to replace one trimpot.

Hey, just for fun while I was testing trimpots, I soldered two
hardlines to a piece of copperclad, with a small gap between their
center conductors, which I bridged with a 47 ohm, 0805 resistor. Then
I did a TDR/TDT test on the mess. The thru signal had a 30 ps risetime
(which is about what the scope is good for) and looked almost perfect,
with a tiny bit of ringing, a few per cent maybe.

John

 

[Joerg]

Yesterday we were playing with a circuit (on a whiteboard) that would
use two gaasfets and an opamp. We would dump some dc current into the
drain of one fet and the opamp would servo its gate voltage to get the
saturated drain voltage to some target. Apply that same gate voltage
to the other fet, and it becomes a stabilized Rds-on to ground, part
of an attenuator. Since the fets on one reel tend to match very well,
that should work.

That's how I usually do that. But I won't rely on reel tracking. In
cases where it's not all that hot you might want to consider these,
under 10c:

http://www.rohm.com/products/databook/tr/pdf/um6k1n.pdf

Of course the 9pf start to hurt at a few hundred MHz but sometimes when
you only need a dB or two you can set it in series with a resistor. Man,
I wish the SD5400 hadn't become boutique-ware. Stuff was so easy when
you could buy them for around two bucks.

The little NEC fet we're using elsewhere (NE3509) make a nice variable
resistor. It behaves just about like a super-hot jfet. Rds-on is about
6 ohms at zero gate voltage, 10 at -0.1, 30 at -0.2. Drain capacitance
is almost constant at about 0.35 pF, about the same as a surface-mount
resistor. Helluva part for 85 cents.

But that's a lot of junk to replace one trimpot.

However, you have to factor in the field returns where something changed
after the unit got banged or the fighter pilot turned freighter pilot
nailed it to the runway again.

Hey, just for fun while I was testing trimpots, I soldered two
hardlines to a piece of copperclad, with a small gap between their
center conductors, which I bridged with a 47 ohm, 0805 resistor. Then
I did a TDR/TDT test on the mess. The thru signal had a 30 ps risetime
(which is about what the scope is good for) and looked almost perfect,
with a tiny bit of ringing, a few per cent maybe.

Sound like the electronics version of shooting a can down the road ;-)

 

[[email protected]]

This is a Bourns 3314G. They work very well, but the hairpins on the
ends add inductance, roughly 7-10 nH, and the substrate is fairly
capacitive, so used as a pot there's a lot of leak-through when the
output is supposed to be zero. With a 100 ohm pot, a 30 ps input step
makes a 70 ps output blip of about 60% amplitude. That corresponds to
a serious amount of transmission in vaguely the 5 GHz sort of area. So
it might be usable to 1 GHz maybe.

ftp://66.117.156.8/3314G_guts.JPG

I'm looking for a better pot, or we might try using a phemt as a shunt
element and control gain with dc gate voltage, like a classic jfet
attenuator. I need gain trimming in a dc-1 GHz+ signal chain, and pin
diodes and most attenuator ic's don't work down to dc.

John

Apologies for using google groups...

Have you looked at this attenuator?
http://www.peregrine-semi.com/content/products/wireless/wireless_pe4309.html

In an ancient (paper) HP data book I found a DC-coupled (but specced
to 26GHz) analogue variable PI-attenuator made from GaAs MESFETs, the
part number was HMMC-1001. There are some interesting ideas in the
datasheet. There is a replica attenuator on the same die and they use
op-amps to servo the gate voltages so that the replica attenuator has
the right attenuation and input impedance at DC, and therefore
(assuming reasonable device matching across the die), so does the RF
attenuator. I don't think Avago will admit that these devices ever
existed so that part is mostly just interesting for the datasheet.

Chris

 

[John Larkin]

Apologies for using google groups...

Have you looked at this attenuator?
http://www.peregrine-semi.com/content/products/wireless/wireless_pe4309.html

Again, these RF guys like to take liberties with the word "DC". The
test circuit has 100 pF caps at the chip input and output, yet the
attenuation accuracy spec is "DC <= 1GHZ", and the graphs all cruise
smoothly down to 0 Hz. There's probably a big internal DC bias on the
switches, but they don't say what it is.

I used some Hittite terminating RF switches where the datasheet
features said, in huge type, "DC-3 GHZ OPERATION". The internal
off-channel terminations quit working below 100 MHz, at which the off
channel signals blow straight through. They blamed me for not reading
the return loss curves carefully enough to deceipher their definition
of DC.

In an ancient (paper) HP data book I found a DC-coupled (but specced
to 26GHz) analogue variable PI-attenuator made from GaAs MESFETs, the
part number was HMMC-1001. There are some interesting ideas in the
datasheet. There is a replica attenuator on the same die and they use
op-amps to servo the gate voltages so that the replica attenuator has
the right attenuation and input impedance at DC, and therefore
(assuming reasonable device matching across the die), so does the RF
attenuator. I don't think Avago will admit that these devices ever
existed so that part is mostly just interesting for the datasheet.

We're considering doing this with discrete phemts, possibly servoing
the gate voltage against a dummy part. The NEC fet I'm using has
absurd (0.35 pF) drain capacitance, so should be good to many GHz, but
its Gm is so high, pinchoff around -0.4 volts, that Rds-on is only
linear to +-0.1 volts or so, so it's only good for very low-level
signals, again stuff that would keep the RF boys happy. If I can find
a not-going-obsolete phemt or mesfet with low drain capacitance,
reasonable Rds-on, and pinchoff greater than a volt or so, we might
reconsider doing it active.

The alternate is to just use two resistors as a simple voltage divider
and hope that the photodiode gain is close and, if it's not, let the
test people start soldering. I don't like to do that.

Or maybe I could build a step attenuator, with resistors and using the
fets as switches. In the fiberoptics business, nothing is ever more
accurate than +-10% or so, if that, so we wouldn't need a lot of steps
to get the amp gain close enough.

John

 

[Joerg]

Again, these RF guys like to take liberties with the word "DC". The
test circuit has 100 pF caps at the chip input and output, yet the
attenuation accuracy spec is "DC <= 1GHZ", and the graphs all cruise
smoothly down to 0 Hz. There's probably a big internal DC bias on the
switches, but they don't say what it is.

I used some Hittite terminating RF switches where the datasheet
features said, in huge type, "DC-3 GHZ OPERATION". The internal
off-channel terminations quit working below 100 MHz, at which the off
channel signals blow straight through. They blamed me for not reading
the return loss curves carefully enough to deceipher their definition
of DC.



We're considering doing this with discrete phemts, possibly servoing
the gate voltage against a dummy part. The NEC fet I'm using has
absurd (0.35 pF) drain capacitance, so should be good to many GHz, but
its Gm is so high, pinchoff around -0.4 volts, that Rds-on is only
linear to +-0.1 volts or so, so it's only good for very low-level
signals, again stuff that would keep the RF boys happy. If I can find
a not-going-obsolete phemt or mesfet with low drain capacitance,
reasonable Rds-on, and pinchoff greater than a volt or so, we might
reconsider doing it active.

Then you'd rely on close enough in-reel tolerances. That's dicey because
they will reel up stuff from the next wafer in the middle of it every so
many reels.

No chance to servo or clamp onto sync signals, carriers or something
else in the signal itself? With my last FO project I lucked out and
could do just that. Or course with a LED and a message to the host if
the sync could not be detected for x milliseconds because the fiber got
ripped out or something.

Regards, Joerg

http://www.analogconsultants.com/

 

[Fred Bartoli]

John Larkin a écrit :

Again, these RF guys like to take liberties with the word "DC". The
test circuit has 100 pF caps at the chip input and output, yet the
attenuation accuracy spec is "DC <= 1GHZ", and the graphs all cruise
smoothly down to 0 Hz. There's probably a big internal DC bias on the
switches, but they don't say what it is.

I used some Hittite terminating RF switches where the datasheet
features said, in huge type, "DC-3 GHZ OPERATION". The internal
off-channel terminations quit working below 100 MHz, at which the off
channel signals blow straight through. They blamed me for not reading
the return loss curves carefully enough to deceipher their definition
of DC.



We're considering doing this with discrete phemts, possibly servoing
the gate voltage against a dummy part. The NEC fet I'm using has
absurd (0.35 pF) drain capacitance, so should be good to many GHz, but
its Gm is so high, pinchoff around -0.4 volts, that Rds-on is only
linear to +-0.1 volts or so, so it's only good for very low-level
signals, again stuff that would keep the RF boys happy. If I can find
a not-going-obsolete phemt or mesfet with low drain capacitance,
reasonable Rds-on, and pinchoff greater than a volt or so, we might
reconsider doing it active.

How about using the R-R linearizing trick? (one R between DG and one
between GS)
Of course you'll need low values, like 100R or lower, but given the low
targeted RDSon it'll surely won't matter.