ben-wpan around the world

Werner Almesberger werner at openmoko.org
Sat Mar 5 21:28:47 EST 2011


The project is moving closer to production.

I made five of the latest atben and five of the latest atusb boards,
plus three atusb-pgm. Then I set up my RF lab again and recorded the
spectrum around the constant wave test signal. Finally, I started
sending out samples.

I didn't take a picture of all the boards, but here are the last two
I have left:

http://downloads.qi-hardware.com/people/werner/wpan/tmp/2boards-20110305.jpg

The others look quite the same, with a slightly more brilliant tin
surface on the atusb boards. (See below.) These are atben-20110219
and atusb-20110214 boards.


You can behold the measurement results in all their beauty here:

http://downloads.qi-hardware.com/people/werner/wpan/20110303/

Click on the thumbnails to enlarge.

atusb-sil is an older atusb board (atusb-20101229-3), still of the
SiLabs-based design. Likewise, atben-20110115-1 and atben-20110123-1
are older designs I included for reference.

atusb-20110214-* and atben-20110219-* are the latest batch. The
2 x 16 very similar-looking graphs on the right show the spectrum
emitted by each board when sending a constant wave at +3 dBm (2 mW)
at +/- 0.5 MHz of the center frequency of each IEEE 802.15.4
channel. The y axis is normalized such that the peak is at 0 dB.

Other information included in the graphs is the center frequency,
the difference of the peak from the center, and the absolute
magnitude of the peak.

The frequency difference can also be seen as a slight shift of the
peak from the center. Frequency differences are caused by variations
in the crystals in the atben/atusb boards and also by any error in
my USRP2's oscillator. IEEE 802.15.4-2003/2006 requires an accuracy
of +/- 40 ppm. The USRP2 has about 20 ppm [1] (+/-, I supposed),
which leaves a range of +/- 20 ppm for the measured tolerance. The
crystal I used has a nominal tolerance of +/- 15 ppm.

[1] http://gnuradio.org/redmine/wiki/1/USRP2GenFAQ#What-is-the-USRP2-reference-clock-stability

As one can see from the data, all clocks are well within
specification.

The magnitude of the peak is shown as a number and also as a
downward-growing green or red bar a bit left of the center. More
is better here.

Returning to the table, the leftmost column shows a summary of the
spectra across all transmit frequencies and below a graph with the
peak power for each frequency.

In the summary, the green line shows the maximum power, the blue
line the average power, and the red line the minimum power. The
averaging algorithm I used isn't too great, so its role is more a
decorative one. The peaks show us if there are any gremlins.

Note that the center frequency of the USRP2 was set to +0.5 MHz
off the transmit frequency. This causes artefacts at +0.5 MHz and
+1.0 MHz from the peak. The other peaks should be for real.

Some of the spectra look a bit like porcupines, with spikes of up
to -30 dB in 0.25 MHz intervals. They may be indicative of the
build quality of the respective board, and may also indicate poor
antenna matching.

I didn't test yet whether they cause a real degradation of
performance. It may be possible to reduce them by adding a small
capacitor. The design already provides a place for this option.
IEEE 802.15.4 limits emissions outside +/- 3.5 MHz off-center to
-20 dB from the peak, so there's no problem with meeting the
standard.

Last but not least, the power-over-transmit-frequency graph shows
how the system performs at different frequencies within the
2.4-2.5 MHz band. My test equipment is not calibrated, so some
distortions will also come from there. The ideal result would be
a flat line, as high a possible (relative to boards tested with
the same setup - absolute values are not very meaningful.)

The atbens look pretty good. The atusbs are also reasonably flat
but seem to favour higher frequencies. This may also be an
artefact of the test setup. (In the case of atusb, also the PC it
is plugged into becomes part of the RF system. I used an OQO 01+,
which is basically a small metal brick.)

All things considered, performance is quite smooth over the whole
range and the individual devices from the latest run don't vary
too much among each other.


Next, there are the mechanical characteristcs. My production has
somewhat variable tolerances, and they show in board dimensions:

Board                   Length          Width           Comment         Dest.
----------------        -------------   -------------   --------------  -------
nominal                 37.6            16.3
atusb-20110214-3        37.0    -1.6%   15.8    -3.1%   bad tin         richard
atusb-20110214-4        37.1    -1.3%   16.1    -1.2%   bad tin         werner
atusb-20110214-5        37.5    -0.3%   16.2    -0.6%                   roh
atusb-20110214-6        37.0    -1.6%   15.9    -2.5%                   adam
atusb-20110214-7        37.5    -0.3%   16.2    -0.6%   DNR C11/C13     adam

Board                   Length outside  Length inside   Width outside   Dest.
----------------        --------------  -------------   -------------   -----
nominal                 35.5            26.3            9.9
atben-20110219-1        35.5      -     26.3      -     9.9       -     adam
atben-20110219-2        35.5      -     26.3      -     9.9       -     roh
atben-20110219-3        35.4    -0.3%   26.3      -     9.9       -     richard
atben-20110219-4        35.5      -     26.2    -0.4%   9.9       -     adam
atben-20110219-5        35.4    -0.3%   26.3      -     9.9       -     werner

All dimensions are in mm. The atbens are all quite good while the
atusbs still suffer some small variations. I tried a different PCB
cleaning process for atusb boards 3 and 4, which turned out to
chemically alter the tin, making it quite difficult to solder.

Also, atusb board 7 suffered copper delamination at C11/13 (RF
power bypassing) during some rework, so these capacitors should
not be reworked further.

In general, I had a hard time making this last batch of atusb
boards. The large ground areas, the number of vias, and also some
issues with layer registration (well, that's just my clumsiness)
all contribute to making the boards difficult to solder. Luckily,
most of these problems disappear with industrially made PCBs. The
atben were fun to make.


As indicated above, I've already assigned boards for all the
recipients of samples. The boards for Adam and Joachim are already
on their way, the ones for Richard will follow next week.

- Werner




More information about the discussion mailing list


interactive