802.15.4 RF issues

[linnix]

We are using the Microchip MRF24JA40 802.15.4 RF transceiver,
might look into Atmel's AT86RF231 and Freescale's MC13202 as well. We
have one working board talking to a Microchip pre-made module, but two
non-working boards. The only difference is the crystal. The working
one is 5 mm SMD and the non-working one is 11 mm 2 pins. Could it be
so sensitive to crystal selection? Do we need to pre-screen them for
productions?

The Freescale chip allows crystal calibration, but won't solve drift
problem. My non-working board detects signals occasionally, but not
good enough to pull in any packet. It seems to be drifting in and out
of the channel.

Also, the 1/4 length for 2.5GHz is 3cm. Most WiFI antenna seems to be
bigger. Are they 3/4 length? Would 1-1/4 be better? We are just
using a wire coiled on top of the PCB.

 

[[email protected]]

We are using the Microchip MRF24JA40 802.15.4 RF transceiver,
might look into Atmel's AT86RF231 and Freescale's MC13202 as well.  We
have one working board talking to a Microchip pre-made module, but two
non-working boards.  The only difference is the crystal.  The working
one is 5 mm SMD and the non-working one is 11 mm 2 pins.  Could it be
so sensitive to crystal selection?  Do we need to pre-screen them for
productions?

a different crystal will probably need different capacitors to be
spot
on frequency. Try measuring the frequency, 802.15.4 requires
something
like +/-40ppm

The Freescale chip allows crystal calibration, but won't solve drift
problem.  My non-working board detects signals occasionally, but not
good enough to pull in any packet.  It seems to be drifting in and out
of the channel.

I believe if you get the caps right wor the crystal you should able
to
stay with in +/40ppm over temperature and aging

Also, the 1/4 length for 2.5GHz is 3cm.  Most WiFI antenna seems to be
bigger.  Are they 3/4 length?  Would 1-1/4 be better?  We are just
using a wire coiled on top of the PCB.

all kinds of different antennas as long as it is matched I'd assume
what
ever their put on give an acceptable performance

but if devices are close I don't think it matters much

I worked on developing the freescale 802.15.4 stuff but that was many
years ago

-Lasse

 

[linnix]

a different crystal will probably need different capacitors to be
spot
on frequency. Try measuring the frequency, 802.15.4 requires
something
like +/-40ppm

The non-working one says 20ppm, not sure if it's stability or
tolerence. However, the cheap stuff (10 for $1.95 including shipping
from Hong Kong) might be fake. The working one (80 cents + $3
shipping from digikey) is actually 30ppm. Will try to order more test
crystals later.

I believe if you get the caps right wor the crystal you should able
to
stay with in +/40ppm over temperature and aging




all kinds of different antennas as long as it is matched I'd assume
what
ever their put on give an acceptable performance

but if devices are close I don't think it matters much

I worked on developing the freescale 802.15.4 stuff but that was many
years ago

-Lasse

I have a 9cm wire talking to the Microchip's PCB E antenna. My signal
(on the devices) is much weaker than the Microchip module
(coordinator), but seems to be OK.

 

[[email protected]]

The non-working one says 20ppm, not sure if it's stability or
tolerence.  However, the cheap stuff (10 for $1.95 including shipping
from Hong Kong) might be fake.  The working one (80 cents + $3
shipping from digikey) is actually 30ppm.  Will try to order more test
crystals later.

my point is that for each different type of xtal you need to measure
the resulting frequency and adjust the caps, it can be much more than
40ppm
off

I have a 9cm wire talking to the Microchip's PCB E antenna.  My signal
(on the devices) is much weaker than the Microchip module
(coordinator), but seems to be OK.

-Lasse

 

[linnix]

Pay attention to parasitic capacitances on the PCB, too.  If you've got
ground plane underneath your crystal lands, then you've got capacitors,
and those capacitors' temperature behavior is no better than the PCB
material's.

Yes, will have better layout for next version.

See if your manufacturer of choice has any app notes on layout for good
oscillator stability.

They suggested 4 layers with 4 ground plane (Digital, Analog, Crystal
and RF). But that would be too expensive. I am cutting corners with
2 layers. Perhaps i am just lucky with the one working board.

 

[linnix]

my point is that for each different type of xtal you need to measure
the resulting frequency and adjust the caps, it can be much more than
40ppm
off

Until i get the 2.5GHz frequency counter, i just have to keep trying
different crystals and caps. I wonder if it's because of the ceramic
SMD vs. metal case HC-59 crystal, in terms of para. caps.

 

[[email protected]]

Until i get the 2.5GHz frequency counter, i just have to keep trying
different crystals and caps.  I wonder if it's because of the ceramic
SMD vs. metal case HC-59 crystal, in terms of para. caps.

you can just measure the xtal frequency

-Lasse

 

[linnix]

you can just measure the xtal frequency

-Lasse

Would it be possible to temperature compensate the crystal with the
MC13202? That might force us to switch part.

 

[[email protected]]

Would it be possible to temperature compensate the crystal with the
MC13202?  That might force us to switch part.

http://cache.freescale.com/files/rf_if/doc/app_note/AN3251.pdf

seems to indicated that is wouldn't be necessary but with the
switchable load caps and temperature measurement I guess you could

-Lasse

 

[Joerg]

Yes, will have better layout for next version.


They suggested 4 layers with 4 ground plane (Digital, Analog, Crystal
and RF). ...

Four (!) different ground planes? That is usually a recipe for disaster.

... But that would be too expensive. I am cutting corners with
2 layers. Perhaps i am just lucky with the one working board.

As Lasse wrote, check for correct burden caps. They may very well have
to be different between crystal versions. If the burden caps are wrong
oscillation start may be recalcitrant and your frequency can we off.
Easy way to check if you don't have a good frequency counter: Program a
timer to spit out 1MHz. Then borrow a shortwave receiver, pull in WWV or
WWVB or 5, 10 or 15MHz, listen to the beat. If the beat is in the kHz
range you won't be a happy camper with channel lock up at 2.45GHz.

 

[Martin Riddle]

The non-working one says 20ppm, not sure if it's stability or
tolerence. However, the cheap stuff (10 for $1.95 including shipping
from Hong Kong) might be fake. The working one (80 cents + $3
shipping from digikey) is actually 30ppm. Will try to order more test
crystals later.


I have a 9cm wire talking to the Microchip's PCB E antenna. My signal
(on the devices) is much weaker than the Microchip module
(coordinator), but seems to be OK.

Make sure the board is cleaned properly. The No clean flux can cause
problems for crystals, especially under the crystal.

Cheers

 

[Oliver Betz]

you can just measure the xtal frequency

where "just measure" means using a high impedance (low capacitance)
probe or an output not connected to the crystal.

Oliver

 

[Oliver Betz]

:

[...]

Make sure the board is cleaned properly. The No clean flux can cause
problems for crystals, especially under the crystal.

could you explain this further, what kind of influence are you talking
about?

Oliver

 

[Oliver Betz]

Jon Elson wrote:

[...]

This usually doesn't show up right away. It develops over weeks or months.
Some fluxes, like "no-clean" can be pretty high impedance, especially right
after soldering or reflow. Then, they absorb water, and the conductivity
goes up. Many logic inverter-type oscillators need some amount of
conductance between the two crystal terminals to put the inverter input
near the transition. But, too much conductance and the oscillator
may fail to start, or may hop in frequency. In some cases I've seen
conductance between adjacent chip pins in the hundreds of Ohms, which
would certainly kill the Q of an oscillator. Cleaning the boards carefully

hundreds of ohms between nets will kill most circuits, won't they?

If they don't kill them immediately, they will do slowly by
electrochemical migration.

with a toothbrush and solvent will fix the problem. It can sometimes

Cleaning boards is a demanding task, especially if they are soldered
with "no clean" flux. No clean often means "no chance to clean".

Oliver (still curious what Martin meant)

 

[Oliver Betz]

Jamie wrote:

[...]

Think about that, the flux gets heated in the process and many fluxes
are made with organics that can break down to carbon. Also, years ago, I
had some flux that I stuck the probes of a cap meter into and I was very
surprised at the dialectic value it had..

I think the air gap between the case and the board will contribute
much more uncertainty.

Oliver

 

[[email protected]]

:

[...]

Make sure the board is cleaned properly. The No clean flux can cause
problems for crystals, especially under the crystal.

could you explain this further, what kind of influence are you talking
about?

Oliver

I have not been followed this discussion with great attention, but
there are several issues related to PCB material humidity and
temperature dependencies in oscillators.

When building free running HF oscillators (e.g. VFOs) never use two
(or multi) layer constructions near the LC resonant circuit. The stray
capacitance between the resonant LC components and the PCB ground
plane affects the frequency. Unfortunately, this stray capacitance
varies with the air humidity and temperature, which affects the
dielectric constant of the PCB material and hence affect the
frequency.

For simple oscillators in the UHF/microwave range, a free running
oscillator made of 1/4 wavelength PCB traces are extremely sensitive
to these issues.

Anyway, one should remember that a frequency drift of 1 Hz at 25 MHz
is 100 Hz at 2.45 GHz. I was once debugging a GHz signal source based
on some VHF overtone crystals and wondered, why the frequency was
shifting every few seconds. I finally discovered that I was breathing
on the crystal, that cased the frequency drift .

When using HF fundamental or VHF overtone crystals as a reference to
frequency multipliers or PLLs, you really need to pay attention to the
PCB material and layout around the crystal oscillator.

 

[Oliver Betz]

Hello Jon,

not announcing a Followup-To: is bad.

So again for c.a.e:

hundreds of ohms between nets will kill most circuits, anyhow.

If they don't kill them immediately, they will do slowly by
electrochemical migration.

Cleaning boards is not easy. No clean often means "no chance to
clean".

Oliver

 

[linnix]

Regarding 2.5GHz test. If i can test the 16M/20M clock, why brother with 2..5G?

Not Zigbee. Just native 802.15.4. Zigbee requires at least 20K code, which would be difficult to fit in a 16K µC.

Somewhere in a poorly written 10,000 page document with no index?

Most 802.15.4 chip set docs are available without NDA. I read the spec forMRF24J40, AT86RF231 and MC13202 several times. Unfortunately, only the MC13202 provides a Clock Out interface based on the crystal.

 

[linnix]

Are you sure you can test at 16/24/32Mhz without affecting the

frequency? Touch either lead of the crystal and you're going to move

the frequnecy.

There is a separate Clock Out (16M, 8M, 4M or 23K) pin, in addition to the two crystal pins. I can just feed this as the µC clock input, and calibrate it again an external low-freq source. I guess Motorola Corp (sorry, Freescale) has more experiences in making RF chips.