# Yo! RF dudes!

J

#### John Larkin

Jan 1, 1970
0
Suppose I have a 1206 ceramic-core inductor on a pc board with nothing
else nearby. What would be the differential effect on inductance of
adding a ground plane on the opposite side of the board? Even more
important, how might it affect the TC?

I've got a 50 MHz oscillator, using a 150 nH inductor, some fixed
caps, a Maxim flecap (coarse tune), and a varicap (fine tune.) Last
board rev, we had it almost perfectly temperature compensated, thanks
to some N750 0603 caps that are special-ordered from Japan, almost
impossible to get in reasonable quantities and time frames. Some
copper was moved on the latest board rev, more copper close to the L,
and now the center frequency and TC are different, so I was wondering
if copper proximity could explain the difference.

Oh, does anybody know of a place to get stock surface-mount NTC caps?

We can software compensate out the new TC, so the thing works, but I'd
rather fix the inherent oscillator TC. If we apply a
linear-with-temperature compensation voltage to the varicap, we can
get a zero TC near room temp. But the varicap has its own, nasty TC
that varies with capacitance, so this compensation winds up with a
parabolic TC curve. OK, we could add a compensating polynomial in the
software, if we didn't mind spending a week or so getting that right.

RF requires a lot of patience.

John

T

#### Tim Wescott

Jan 1, 1970
0
John said:
Suppose I have a 1206 ceramic-core inductor on a pc board with nothing
else nearby. What would be the differential effect on inductance of
adding a ground plane on the opposite side of the board? Even more
important, how might it affect the TC?

I've got a 50 MHz oscillator, using a 150 nH inductor, some fixed
caps, a Maxim flecap (coarse tune), and a varicap (fine tune.) Last
board rev, we had it almost perfectly temperature compensated, thanks
to some N750 0603 caps that are special-ordered from Japan, almost
impossible to get in reasonable quantities and time frames. Some
copper was moved on the latest board rev, more copper close to the L,
and now the center frequency and TC are different, so I was wondering
if copper proximity could explain the difference.

Oh, does anybody know of a place to get stock surface-mount NTC caps?

We can software compensate out the new TC, so the thing works, but I'd
rather fix the inherent oscillator TC. If we apply a
linear-with-temperature compensation voltage to the varicap, we can
get a zero TC near room temp. But the varicap has its own, nasty TC
that varies with capacitance, so this compensation winds up with a
parabolic TC curve. OK, we could add a compensating polynomial in the
software, if we didn't mind spending a week or so getting that right.

RF requires a lot of patience.

John
The effect on inductance would depend on the construction of the coil,
but if the coil is wound from end to end a conductor underneath the coil
shouldn't affect the inductance that much.

Absent a shield on the coil, such a conductor _would_ capacitively
couple to the coil windings, which would change (probably lower) your
center frequency. FR-4 has a pretty healthy temperature coefficient, so
the added capacitance would be highly variable with temperature; this

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" gives you just what it says.
See details at http://www.wescottdesign.com/actfes/actfes.html

J

#### Joerg

Jan 1, 1970
0
John said:
Suppose I have a 1206 ceramic-core inductor on a pc board with nothing
else nearby. What would be the differential effect on inductance of
adding a ground plane on the opposite side of the board? Even more
important, how might it affect the TC?

Usually it adds stray capacitance (winding to plane), with whatever TC
the dielectric of the PCB material exhibits.

I've got a 50 MHz oscillator, using a 150 nH inductor, some fixed
caps, a Maxim flecap (coarse tune), and a varicap (fine tune.) Last
board rev, we had it almost perfectly temperature compensated, thanks
to some N750 0603 caps that are special-ordered from Japan, almost
impossible to get in reasonable quantities and time frames. Some
copper was moved on the latest board rev, more copper close to the L,
and now the center frequency and TC are different, so I was wondering
if copper proximity could explain the difference.

Oh, does anybody know of a place to get stock surface-mount NTC caps?

The market for that stuff has become so small that you'd have to rely on
expensive boutique parts. Always with the risk that the next call
results in "We no longer manufacture...". I most certainly would keep
doing that electronically (as you have done).

We can software compensate out the new TC, so the thing works, but I'd
rather fix the inherent oscillator TC. ...

Allow me the question: Why?

... If we apply a
linear-with-temperature compensation voltage to the varicap, we can
get a zero TC near room temp. But the varicap has its own, nasty TC
that varies with capacitance, so this compensation winds up with a
parabolic TC curve. OK, we could add a compensating polynomial in the
software, if we didn't mind spending a week or so getting that right.

The brute force approach would be to keep that part of the circuitry
temperature-regulated. Did that a lot with oscillators in the olden
days. I always used a temperature that is guaranteed to be higher than
ambient will ever be so no TEC was required, just a resistor as a heater.

Another thing I have done but in precision phase shifters and not
oscillators was to use either FET arrays (mostly gone the boutique route
by now...) or PIN diodes to pull in a capacitance. Mostly with the
HSMP38xx series but nowadays there are some nice PIN diodes in the
Infineon BAP series. If you can get them, that is. I found that company
to be less than efficient when buying prototype qties <1000 in the US.

RF requires a lot of patience.

Sure does, but the results can be very rewarding and the solution
stunningly sparse in parts count. Sometimes

R

#### RST Engineering $$jw$$

Jan 1, 1970
0
John Larkin said:
Suppose I have a 1206 ceramic-core inductor on a pc board with nothing
else nearby. What would be the differential effect on inductance of
adding a ground plane on the opposite side of the board? Even more
important, how might it affect the TC?

The center frequency should have gone down. The TC is now a function of the
PCB material (G10/FR4?) which is wildly variant with temperature. Get the
copper out of there if you can.

RF requires a lot of patience.

....and powdered bat wings, ground toad warts, incantations, and a virgin (if
you can find one these days) sacrifice.

Jim

J

#### Joerg

Jan 1, 1970
0
Tim said:
The effect on inductance would depend on the construction of the coil,
but if the coil is wound from end to end a conductor underneath the coil
shouldn't affect the inductance that much.

Absent a shield on the coil, such a conductor _would_ capacitively
couple to the coil windings, which would change (probably lower) your
center frequency. FR-4 has a pretty healthy temperature coefficient, so
the added capacitance would be highly variable with temperature; this

Now just imagine the poor Europeans who have those inductors floating
about in some kind of RoHS compliant goo of unknown performance.

J

#### Joerg

Jan 1, 1970
0
RST said:
The center frequency should have gone down. The TC is now a function of the
PCB material (G10/FR4?) which is wildly variant with temperature. Get the
copper out of there if you can.

Sometimes it helps stamping it out or drill a hole underneath. In the
old days when inductors had real leads we just placed them a bit higher.
Of course then you have to deal with microphonics...
...and powdered bat wings, ground toad warts, incantations, and a virgin (if
you can find one these days) sacrifice.

I use wee whiffs of incense...

J

#### John Larkin

Jan 1, 1970
0
The center frequency should have gone down. The TC is now a function of the
PCB material (G10/FR4?) which is wildly variant with temperature. Get the
copper out of there if you can.

...and powdered bat wings, ground toad warts, incantations, and a virgin (if
you can find one these days) sacrifice.

Jim

Yes, the frequency shift direction and tc are consistant with extra
board capacitance. I once measured some FR4 capacitance and got +900
ppm per degree C. The oscillator frequency tc apparently shifted from
near zero to -70 ppm when the extra ground was added. I guess I'll
measure the critical node capacitance on a bare pcb and do the math,
just to see if the numbers are close. Luckily, there's a pattern I can
cut out of the backside copper that should seriously reduce C and not
trash any vias. Good thing we only bought 50 boards!

Still, I'm curious about the effect of a ground plane on the inductor.
I can, and will, measure the effect on L, but it's more difficult to
measure the effect on the inductor's TC.

John

J

#### Joerg

Jan 1, 1970
0
John said:
Yes, the frequency shift direction and tc are consistant with extra
board capacitance. I once measured some FR4 capacitance and got +900
ppm per degree C. The oscillator frequency tc apparently shifted from
near zero to -70 ppm when the extra ground was added. I guess I'll
measure the critical node capacitance on a bare pcb and do the math,
just to see if the numbers are close. Luckily, there's a pattern I can
cut out of the backside copper that should seriously reduce C and not
trash any vias. Good thing we only bought 50 boards!

Still, I'm curious about the effect of a ground plane on the inductor.
I can, and will, measure the effect on L, but it's more difficult to
measure the effect on the inductor's TC.

It's most likely miniscule compared to the capacitance (bottom of
winding to plane). It would be easy to measure the added capacitance if
you can get something like a HP4191 input pod really close to that
point. Dremel away the suspect area, measure, place copper tape over it,
measure again.

Question: Is there some headroom to just solder silver wire to the pads
and then the inductor to those, so it comes 1/4" or so off the board?
Those wires should be beefy and stiff to avoid microphonics, else you'd
have an earthquake detector.

J

#### Jamie

Jan 1, 1970
0
John said:
Suppose I have a 1206 ceramic-core inductor on a pc board with nothing
else nearby. What would be the differential effect on inductance of
adding a ground plane on the opposite side of the board? Even more
important, how might it affect the TC?

I've got a 50 MHz oscillator, using a 150 nH inductor, some fixed
caps, a Maxim flecap (coarse tune), and a varicap (fine tune.) Last
board rev, we had it almost perfectly temperature compensated, thanks
to some N750 0603 caps that are special-ordered from Japan, almost
impossible to get in reasonable quantities and time frames. Some
copper was moved on the latest board rev, more copper close to the L,
and now the center frequency and TC are different, so I was wondering
if copper proximity could explain the difference.

Oh, does anybody know of a place to get stock surface-mount NTC caps?

We can software compensate out the new TC, so the thing works, but I'd
rather fix the inherent oscillator TC. If we apply a
linear-with-temperature compensation voltage to the varicap, we can
get a zero TC near room temp. But the varicap has its own, nasty TC
that varies with capacitance, so this compensation winds up with a
parabolic TC curve. OK, we could add a compensating polynomial in the
software, if we didn't mind spending a week or so getting that right.

RF requires a lot of patience.

John
You may have some SR problems at some spectrum..

M

#### Mike Monett

Jan 1, 1970
0
[...]
I've got a 50 MHz oscillator, using a 150 nH inductor, some fixed
caps, a Maxim flecap (coarse tune), and a varicap (fine tune.)
[...]

We can software compensate out the new TC, so the thing works, but
I'd rather fix the inherent oscillator TC. If we apply a
linear-with-temperature compensation voltage to the varicap, we
can get a zero TC near room temp. But the varicap has its own,
nasty TC that varies with capacitance, so this compensation winds
up with a parabolic TC curve. OK, we could add a compensating
polynomial in the software, if we didn't mind spending a week or
so getting that right.
RF requires a lot of patience.

If the oscillator frequency is critical enough to require
compensation, why not lock it to a nearby crystal? You've already
got the varicap, why spend a week or so fiddling with compensation?

There has to be some crystal oscillator nearby. An hour's work
should get the divider and loop filter. Build, debug document and
ship in an afternoon.

You do this often enough. Why is there a problem in this case?

Regards,

Mike Monett

J

#### john jardine

Jan 1, 1970
0
John Larkin said:
Suppose I have a 1206 ceramic-core inductor on a pc board with nothing
else nearby. What would be the differential effect on inductance of
adding a ground plane on the opposite side of the board? Even more
important, how might it affect the TC?

I've got a 50 MHz oscillator, using a 150 nH inductor, some fixed
caps, a Maxim flecap (coarse tune), and a varicap (fine tune.) Last
board rev, we had it almost perfectly temperature compensated, thanks
to some N750 0603 caps that are special-ordered from Japan, almost
impossible to get in reasonable quantities and time frames. Some
copper was moved on the latest board rev, more copper close to the L,
and now the center frequency and TC are different, so I was wondering
if copper proximity could explain the difference.

Oh, does anybody know of a place to get stock surface-mount NTC caps?

We can software compensate out the new TC, so the thing works, but I'd
rather fix the inherent oscillator TC. If we apply a
linear-with-temperature compensation voltage to the varicap, we can
get a zero TC near room temp. But the varicap has its own, nasty TC
that varies with capacitance, so this compensation winds up with a
parabolic TC curve. OK, we could add a compensating polynomial in the
software, if we didn't mind spending a week or so getting that right.

RF requires a lot of patience.

John
Not an RF person but would see the Copper under the coil as the total
problem.
The tuned circuit is now part tuned by the added yucky 1%decC TC, Q=80, FR
material distributed capacitance but in addition this Copper "screening" (by
traditional mechanisms) is also dropping the L value by say 5% and the Q
value by say 5%.

J

#### John Larkin

Jan 1, 1970
0
[...]
I've got a 50 MHz oscillator, using a 150 nH inductor, some fixed
caps, a Maxim flecap (coarse tune), and a varicap (fine tune.)
[...]

We can software compensate out the new TC, so the thing works, but
I'd rather fix the inherent oscillator TC. If we apply a
linear-with-temperature compensation voltage to the varicap, we
can get a zero TC near room temp. But the varicap has its own,
nasty TC that varies with capacitance, so this compensation winds
up with a parabolic TC curve. OK, we could add a compensating
polynomial in the software, if we didn't mind spending a week or
so getting that right.
RF requires a lot of patience.

If the oscillator frequency is critical enough to require
compensation, why not lock it to a nearby crystal? You've already
got the varicap, why spend a week or so fiddling with compensation?

There has to be some crystal oscillator nearby. An hour's work
should get the divider and loop filter. Build, debug document and
ship in an afternoon.

You do this often enough. Why is there a problem in this case?

Regards,

Mike Monett

This is a 50 MHz oscillator that is started by an external trigger.
There's a DSP servo loop, in an fpga, that does in fact lock it to a
quartz reference, but preserves the phasing relative to the trigger.
So the frequencies are locked, and the phase is also locked to
whatever the phase difference was when the LC started.

The oscillator is coarse-tuned by the flecap at powerup, and then the
varicap takes over. We need to keep the varicap pull range fairly
small to meet our noise/jitter requirements. What we don't like is
that the inherent TC of the oscillator was tuned to near zero, and a
minor layout change messed it up enough to lose lock at temperature
extremes. Again, I have a trim dac and a temperature sensor on board,
and the firmware is tweaking the baseline varicap voltage to mostly
null out the temperature drift, but we don't understand what's going
on.

I measured the hot node capacitance on a bare board and it's only 1.2
pF, which accounts for only 10 ppm of the 70 ppm/degC we're seeing, so
it's still strange.

But who can "Build, debug document and ship in an afternoon."?

John

J

#### Joerg

Jan 1, 1970
0
John said:
[...]
I've got a 50 MHz oscillator, using a 150 nH inductor, some fixed
caps, a Maxim flecap (coarse tune), and a varicap (fine tune.)
[...]

We can software compensate out the new TC, so the thing works, but
I'd rather fix the inherent oscillator TC. If we apply a
linear-with-temperature compensation voltage to the varicap, we
can get a zero TC near room temp. But the varicap has its own,
nasty TC that varies with capacitance, so this compensation winds
up with a parabolic TC curve. OK, we could add a compensating
polynomial in the software, if we didn't mind spending a week or
so getting that right.
RF requires a lot of patience.

If the oscillator frequency is critical enough to require
compensation, why not lock it to a nearby crystal? You've already
got the varicap, why spend a week or so fiddling with compensation?

There has to be some crystal oscillator nearby. An hour's work
should get the divider and loop filter. Build, debug document and
ship in an afternoon.

You do this often enough. Why is there a problem in this case?

Regards,

Mike Monett

This is a 50 MHz oscillator that is started by an external trigger.
There's a DSP servo loop, in an fpga, that does in fact lock it to a
quartz reference, but preserves the phasing relative to the trigger.
So the frequencies are locked, and the phase is also locked to
whatever the phase difference was when the LC started.

The oscillator is coarse-tuned by the flecap at powerup, and then the
varicap takes over. We need to keep the varicap pull range fairly
small to meet our noise/jitter requirements. ...

How come? Is noise getting into the varicap drive?

... What we don't like is
that the inherent TC of the oscillator was tuned to near zero, and a
minor layout change messed it up enough to lose lock at temperature
extremes. Again, I have a trim dac and a temperature sensor on board,
and the firmware is tweaking the baseline varicap voltage to mostly
null out the temperature drift, but we don't understand what's going
on.

I measured the hot node capacitance on a bare board and it's only 1.2
pF, which accounts for only 10 ppm of the 70 ppm/degC we're seeing, so
it's still strange.

Who knows what else is going on. Eddy currents in the added plane etc.
Can't you just raise the inductor by 1/4" and see if the problem will go
away, or at least push it outside the critical range?

But who can "Build, debug document and ship in an afternoon."?

Some software companies seem to do that 8-D

It's like with bananas and tomatoes, the ripening process happens at the
customer.

M

Jan 1, 1970
0
J

#### John Larkin

Jan 1, 1970
0
John said:
[...]

I've got a 50 MHz oscillator, using a 150 nH inductor, some fixed
caps, a Maxim flecap (coarse tune), and a varicap (fine tune.)

[...]

We can software compensate out the new TC, so the thing works, but
I'd rather fix the inherent oscillator TC. If we apply a
linear-with-temperature compensation voltage to the varicap, we
can get a zero TC near room temp. But the varicap has its own,
nasty TC that varies with capacitance, so this compensation winds
up with a parabolic TC curve. OK, we could add a compensating
polynomial in the software, if we didn't mind spending a week or
so getting that right.

RF requires a lot of patience.

John

If the oscillator frequency is critical enough to require
compensation, why not lock it to a nearby crystal? You've already
got the varicap, why spend a week or so fiddling with compensation?

There has to be some crystal oscillator nearby. An hour's work
should get the divider and loop filter. Build, debug document and
ship in an afternoon.

You do this often enough. Why is there a problem in this case?

Regards,

Mike Monett

This is a 50 MHz oscillator that is started by an external trigger.
There's a DSP servo loop, in an fpga, that does in fact lock it to a
quartz reference, but preserves the phasing relative to the trigger.
So the frequencies are locked, and the phase is also locked to
whatever the phase difference was when the LC started.

The oscillator is coarse-tuned by the flecap at powerup, and then the
varicap takes over. We need to keep the varicap pull range fairly
small to meet our noise/jitter requirements. ...

How come? Is noise getting into the varicap drive?

Noise *always* gets into varicap drive. And varicaps have rotten TCs,
and the TCs change with voltage, which is a fresh bag'o worms. Which
is why my linear oscillator TC, compensated by a linear temperature
correction voltage, results in a parabolic freq:temp curve!

Who knows what else is going on. Eddy currents in the added plane etc.
Can't you just raise the inductor by 1/4" and see if the problem will go
away, or at least push it outside the critical range?

I might try hacking the ground plane away from the back side. Anything
I try takes hours in the temp chamber to evaluate.

John

R

#### RST Engineering $$jw$$

Jan 1, 1970
0
As I heard the story as a youngster (mommy having worked for the brewer in
question), a famous St. Louis brewer who was running out of tank brewing
space poured malt, hops, water, and yeast into a railway tank car and sent
it to Milwaukee where it was turned around and sent back to St. Louis as
fermented beer.

Jim

J

#### Jim Thompson

Jan 1, 1970
0
And some wines. Aged in transit.

Regards,

Mike Monett

What? Thunderbird ?

...Jim Thompson

M

#### Mike Monett

Jan 1, 1970
0
[...]
The oscillator is coarse-tuned by the flecap at powerup, and then
the varicap takes over. We need to keep the varicap pull range
fairly small to meet our noise/jitter requirements. What we don't
like is that the inherent TC of the oscillator was tuned to near
zero, and a minor layout change messed it up enough to lose lock
at temperature extremes. Again, I have a trim dac and a
temperature sensor on board, and the firmware is tweaking the
baseline varicap voltage to mostly null out the temperature drift,
but we don't understand what's going on.

Sounds like the poor varicap is pushed to the limit. How about a
slightly larger one if it would still allow you to meet the phase
noise. Or maybe several in parallel. Called Massively-Parallel.
Rohde shows examples in figures 5 and 8 of

http://www.synergymicrowave.net/articles/PDF/Voltage-Controlled-
Oscillators.pdf (132KB)

(sorry for the wrap)

Also, if I'm not mistaken, the flecap has pretty low Q. This might
also be worth looking at to try to reduce the effect. Other circuit
on the tank.

I found using a simple constant current source in place of the
emitter resistor in a Colpits really improves the tank Q under load.

It also helps align the emitter current pulse with the peak of the
sine wave from the tank, which improves the phase noise. Hajimiri
and Lee discuss this in

http://www.chic.caltech.edu/Publications/phase_tutor.pdf

Also, Rohde patented a simple bias stabilization technique shown in
figure 15 of

http://www.synergymicrowave.net/articles/PDF/Oscillators.pdf (915KB)

I believe Matjaz Vitmar, S53MV, also described it long before Rohde
applied for the patent.

I've tried it, and it works very well. Other feedback arrangements
to stabilize the bias also help improve the phase noise.

So there's lots to look at that might help.

[...]
But who can "Build, debug document and ship in an afternoon."?

Depends on how much you can cut and paste from previous projects.
Betcha Joerg does it all the time

Regards,

Mike Monett

J

#### Joerg

Jan 1, 1970
0
John said:
John Larkin wrote:

[...]

I've got a 50 MHz oscillator, using a 150 nH inductor, some fixed
caps, a Maxim flecap (coarse tune), and a varicap (fine tune.)

[...]

We can software compensate out the new TC, so the thing works, but
I'd rather fix the inherent oscillator TC. If we apply a
linear-with-temperature compensation voltage to the varicap, we
can get a zero TC near room temp. But the varicap has its own,
nasty TC that varies with capacitance, so this compensation winds
up with a parabolic TC curve. OK, we could add a compensating
polynomial in the software, if we didn't mind spending a week or
so getting that right.

RF requires a lot of patience.

John

If the oscillator frequency is critical enough to require
compensation, why not lock it to a nearby crystal? You've already
got the varicap, why spend a week or so fiddling with compensation?

There has to be some crystal oscillator nearby. An hour's work
should get the divider and loop filter. Build, debug document and
ship in an afternoon.

You do this often enough. Why is there a problem in this case?

Regards,

Mike Monett

This is a 50 MHz oscillator that is started by an external trigger.
There's a DSP servo loop, in an fpga, that does in fact lock it to a
quartz reference, but preserves the phasing relative to the trigger.
So the frequencies are locked, and the phase is also locked to
whatever the phase difference was when the LC started.

The oscillator is coarse-tuned by the flecap at powerup, and then the
varicap takes over. We need to keep the varicap pull range fairly
small to meet our noise/jitter requirements. ...

How come? Is noise getting into the varicap drive?

Noise *always* gets into varicap drive. And varicaps have rotten TCs,
and the TCs change with voltage, which is a fresh bag'o worms. Which
is why my linear oscillator TC, compensated by a linear temperature
correction voltage, results in a parabolic freq:temp curve!

You might have to shield that portion with the varicaps and the loop for
them. Just had to do the same thing. A laser diode driver where where a
fraction of a microamp of noise would throw a whole system off the rocker.
I might try hacking the ground plane away from the back side. Anything
I try takes hours in the temp chamber to evaluate.

If it's lots of boards that might be more work than raising them.