The Larkin front end

[martin griffith]

Hi John
Just a few questions from the ABSE post

what are the typical values of L4 and L5, I guess 100uH?
do you ever use a common mode i/p filter?


martin

 

[John Larkin]

Hi John
Just a few questions from the ABSE post

what are the typical values of L4 and L5, I guess 100uH?
do you ever use a common mode i/p filter?

They are surface-mount ferrite beads, I think rated 300 ohms (at 100
MHz, which how they usually rate beads.) Beads are nice because the Q
is so low, they pretty much never resonate.

I haven't ever use a common-mode filter for stuff like this. The
intent was to just squash any RF, common mode or differential.
Nowadays, cell phones and things are everywhere.

One of my guys has an older, low-band cell phone, and whenever he
walks into my office my PC speakers make horrible raspberry noises.

John

 

[[email protected]]

They are surface-mount ferrite beads, I think rated 300 ohms (at 100
MHz, which how they usually rate beads.) Beads are nice because the Q
is so low, they pretty much never resonate.

If only that were true. I had a ferrite bead in one of my designs that
had an inductance of 1.6uH, and resonated nicely with a 100nF capacitor
at 400kHz, with a Q of about 5.

I ended up having to add a wound 100uH SMD inductor in series, with 22R
of damping resistance. Dead embarassing. The quick and dirty fix was
4u7F tantalum capacitor in parallel with the 100nF ceramic - the ESR of
that tantalum did all the damping required, but the exta capacitance
chewed up current whenever we intermittently activated the low-power
system involved.

I haven't ever use a common-mode filter for stuff like this. The
intent was to just squash any RF, common mode or differential.
Nowadays, cell phones and things are everywhere.

One of my guys has an older, low-band cell phone, and whenever he
walks into my office my PC speakers make horrible raspberry noises.

In my day it was the electron beam microfabricator that wrote the wrong
stuff on a $2,000 mask if the security guy went down the corridor with
his two-way radio. Fixable, eventually.

 

[martin griffith]

On 7 Nov 2006 01:53:53 -0800, in sci.electronics.design
[email protected] wrote:

Hi Bill

I ended up having to add a wound 100uH SMD inductor in series, with 22R
of damping resistance

do you mean the 22R was in parallel to the inductor?


martin

 

[[email protected]]

On 7 Nov 2006 01:53:53 -0800, in sci.electronics.design
[email protected] wrote:

Hi Bill

do you mean the 22R was in parallel to the inductor?

I didn't specify, but it was in series with both the inductors (who
were also in series).

The whole thing was intended to keep the switching spikes from a
comparator out of the the op amp that was providing one of the inputs
to the comparator - always worth doing, and particularly so in this
case.

 

[Joerg]

Hello John,

They are surface-mount ferrite beads, I think rated 300 ohms (at 100
MHz, which how they usually rate beads.) Beads are nice because the Q
is so low, they pretty much never resonate.

Might not be a problem with most SMT versions but it is with thru-hole:
They can exhibit have mechanical resonance. Or in more popular speak a
"rattle".

 

[Rich Grise]

Hello John,


Might not be a problem with most SMT versions but it is with thru-hole:
They can exhibit have mechanical resonance. Or in more popular speak a
"rattle".

Hey, there's something that'd never occurred to me - if the core is loose,
it will actually become like a little motor, right?

Thanks!
Rich

 

[John Larkin]

If only that were true. I had a ferrite bead in one of my designs that
had an inductance of 1.6uH, and resonated nicely with a 100nF capacitor
at 400kHz, with a Q of about 5.

I ended up having to add a wound 100uH SMD inductor in series, with 22R
of damping resistance. Dead embarassing. The quick and dirty fix was
4u7F tantalum capacitor in parallel with the 100nF ceramic - the ESR of
that tantalum did all the damping required, but the exta capacitance
chewed up current whenever we intermittently activated the low-power
system involved.


In my day it was the electron beam microfabricator that wrote the wrong
stuff on a $2,000 mask if the security guy went down the corridor with
his two-way radio. Fixable, eventually.

Some years ago, the head of R&D for a big NMR company called me and
asked if we wanted to make sample temperature controllers for them.
They'd been using units from Oxford Instruments for years, and were
having problems, one of which was astounding EMI sensitivity (the
other problem being that they had apparently lost the Z80 source code,
and couldn't fix bugs!)

I got one of the Oxfords, and found that I could shut it down from
across the room with one of those old GR unit oscillators driving a
banana lead antenna. There were multiple, very narrow bands in the
150-400 MHz range where it was extraordinarily sensitive. Turns out
the sensitivity was hugely enhanced by internal wiring and PCB
resonances. This is the range where beads work well.

We built a new controller with much better layout and bypassing, and
it was about 20:1 less sensitive. When we ran the internal
thermocouple leads (t/c connector to pcb) through a double-hole
ferrite bead, we got another 10:1, presumable because it blocked entry
and killed the Q of that path. We've shipped close to 3000 units by
now.

John

 

[Joel Kolstad]

We built a new controller with much better layout and bypassing, and
it was about 20:1 less sensitive. When we ran the internal
thermocouple leads (t/c connector to pcb) through a double-hole
ferrite bead, we got another 10:1, presumable because it blocked entry
and killed the Q of that path.

I wouldn't be surprised if, even with all this, your build cost is still less
than The Big Guy's.

We've shipped close to 3000 units by
now.

Congratulations!

 

[[email protected]]

Some years ago, the head of R&D for a big NMR company called me and
asked if we wanted to make sample temperature controllers for them.
They'd been using units from Oxford Instruments for years, and were
having problems, one of which was astounding EMI sensitivity (the
other problem being that they had apparently lost the Z80 source code,
and couldn't fix bugs!)

I got one of the Oxfords, and found that I could shut it down from
across the room with one of those old GR unit oscillators driving a
banana lead antenna. There were multiple, very narrow bands in the
150-400 MHz range where it was extraordinarily sensitive. Turns out
the sensitivity was hugely enhanced by internal wiring and PCB
resonances. This is the range where beads work well.

We built a new controller with much better layout and bypassing, and
it was about 20:1 less sensitive. When we ran the internal
thermocouple leads (t/c connector to pcb) through a double-hole
ferrite bead, we got another 10:1, presumable because it blocked entry
and killed the Q of that path.

Actually, that is a minimal balun, and gives you good differential
rejection at frequencies where it has a significant impedance.

If you want more turns, you can use a small ferrite torroid - if you
confine the wires to a single layer and space them at about their own
diameter, you are supposed get the maximum availalbe resonant
frequency. I've not had occasion to try it yet, but the theory seemed
to make sense.

We've shipped close to 3000 units by now.

Nice work.

 

[mark]

Some years ago, the head of R&D for a big NMR company called me and
asked if we wanted to make sample temperature controllers for them.
They'd been using units from Oxford Instruments for years, and were
having problems, one of which was astounding EMI sensitivity (the
other problem being that they had apparently lost the Z80 source code,
and couldn't fix bugs!)

I got one of the Oxfords, and found that I could shut it down from
across the room with one of those old GR unit oscillators driving a
banana lead antenna. There were multiple, very narrow bands in the
150-400 MHz range where it was extraordinarily sensitive. Turns out
the sensitivity was hugely enhanced by internal wiring and PCB
resonances. This is the range where beads work well.

now.

John

John
Any good articles re EMI suppresion you can point us to? Which authors do
you feel do a good job explaining the problem and solutions?
(I'll have to remember ferrite beads next time I come across a narrow band
sensitivity. Thanks for the tip)
M Walter

 

[Eeyore]

Any good articles re EMI suppresion you can point us to? Which authors do
you feel do a good job explaining the problem and solutions?
(I'll have to remember ferrite beads next time I come across a narrow band
sensitivity. Thanks for the tip)

Ferrite beads are rubbish for EMI on differential inputs since their values are
never matched and the resulting imbalance can cause more trouble than previously
existed.

I know because I've run real tests.

Graham

 

[John Larkin]

Ferrite beads are rubbish for EMI on differential inputs since their values are
never matched and the resulting imbalance can cause more trouble than previously
existed.

I know because I've run real tests.

Does that mean I have to give all the money back?

John

 

[[email protected]]

Ferrite beads are rubbish for EMI on differential inputs since their values are
never matched and the resulting imbalance can cause more trouble than previously
existed.

I know because I've run real tests.

That why you can buy two-hole beads, not to mention six hole beads,
rings and tubes - I doubt if the ferrite is perfectly uniform within
these parts, but the matching is a lot closer than it would be from
part to part.

 

[John Larkin]

That why you can buy two-hole beads, not to mention six hole beads,
rings and tubes - I doubt if the ferrite is perfectly uniform within
these parts, but the matching is a lot closer than it would be from
part to part.

I don't understand this "matching" thing. If you lowpass filter both
sides effectively, you're way better off, even if the lowpass
filtering isn't exactly the same on both sides. The transistors in the
front-end of an opamp are essentially square-law RF rectifiers. If you
have 100 mv of RF, they will rectify it big-time. If you have 1 mv,
they barely will at all. 100^2 = 10000, and 10000:1 is usually
considered to be a pretty big improvement.

The really insidious RF is in the 100+ MHz range, where amp internal
feedback and bias mechanisms can't track, and things go nonlinear...
every junction drives stray capacitance to form a peak detector.
Ferrite beads attenuate the RF and kill resonant Q's.

Wound inductors can be useful, too, but you've got to make sure they
don't resonate anywhere dangerous, make sure their parasitic
capacitance doesn't let hf stuff blow through, and make sure they're
not hum detectors.

John

 

[Eeyore]

I don't understand this "matching" thing.

You're not as smart as I once thought.

If you lowpass filter both
sides effectively, you're way better off, even if the lowpass
filtering isn't exactly the same on both sides. The transistors in the
front-end of an opamp are essentially square-law RF rectifiers. If you
have 100 mv of RF, they will rectify it big-time. If you have 1 mv,
they barely will at all. 100^2 = 10000, and 10000:1 is usually
considered to be a pretty big improvement.

Consider the inherent rejection of common mode signals in a differential front end. It
doesn't work so well if the signals are different sizes ( rather obviously ).

I'll use 1% caps here too.

Graham

 

[SioL]

Consider the inherent rejection of common mode signals in a differential front end. It
doesn't work so well if the signals are different sizes ( rather obviously ).

I'll use 1% caps here too.
Graham

I think I understand John.
If the first stage is a balanced opamp input, its a biq question how "balanced" its function
is as a rf detector, the interference signals are likely way out of the operational range
of that opamp. Thermocouple sounds like a very slow device. Pretty much killing
anything even remotely RF should work pretty well.

SioL

 

[Eeyore]

I think I understand John.
If the first stage is a balanced opamp input, its a biq question how "balanced" its function
is as a rf detector, the interference signals are likely way out of the operational range
of that opamp. Thermocouple sounds like a very slow device. Pretty much killing
anything even remotely RF should work pretty well.

A ferrite bead won't do that.

Graham

 

[John Larkin]

You're not as smart as I once thought.



Consider the inherent rejection of common mode signals in a differential front end. It
doesn't work so well if the signals are different sizes ( rather obviously ).

I'll use 1% caps here too.

Graham

If you're referring to capacitive imbalances turning common-mode stuff
into differential, that's a different issue. In the case of load
cells, with ballpark 200 ohm drives and 0.1 uF caps, and a deep 50/60
Hz notch in the delta-sigma adc response, it's not an issue at all; do
the math. The ferrites are dead shorts at these frequencies. With a 25
mV full-scale load-cell signal, I can weigh things to 5-digit noise
and linearity with *unshielded* leads.

Line rejection is a trivial issue here, something obvious that simple
math can handle. RF rectification is far more insidious, less
predictable, and more dangerous (a little AC noise is nothing compared
to hard railing the ADC) and what you see on the schematic is just
part of the story.

Another thing I don't understand is why you insist on calling me
stupid when you disagree with me. I've never called you stupid.

John

 

[John Larkin]

A ferrite bead won't do that.

Graham

The only complication being that it does.

John