dna@dsd.UUCP (06/04/84)
To: All packet stations
From: Tom Clark, W3IWI
Subject: Eye Patterns for packet radio testing
Date: 03 June 1984
I have been asked to explain a little more about eye-patterns for
testing the performance of demodulators. In my original note on the
FT726 performance I recommended the following setup:
---------------------
| Triggerable |
| Scope |
| |
Demodulator ------->| Vert In |
Analog Output | |
| |
Clipped Digital ---->| Trigger In |
Data | |
---------------------
The analog output from the demod should be taken off at the last point
in the demod that the signal is still analog -- i.e. just before the
clipper that converts it into digital format. On the TAPR TNC, this
analog output is available on the XR2211 on pin 8. Similar points exist
in commercial demods. The digital data should be the digital signal you
are actually going to feed into your UART or your HDLC chip. On the
TAPR TNC, use the pull-up resistor on pin 27 of the 1933 as a tap. (I
am not referencing the test points and/or header pins on TAPR since the
Beta and Kit versions have different designators).
The decision on whether the analog signal is a 0 or a 1 is made in
some sort of clipper thresholding device. This is in the 2211 on TAPR,
it is often done in a 741 or 339 in other modems. This clipper is
pretty dumb -- it only knows up and down. To make its life easiest, it
wants to see a clean analog signal. The eye-pattern is the easiest way
to test what type of signal it is being given. The following is a
"perfect" eye-pattern (pardon the computer graphics!), assuming you are
triggering on the + going edge :
-------------------------------------------------------
/ \ \ / \
/ \ \ / \
/ \ \ / \
/ \ \ / \ /
!---------!---------\---------!---------X---------!---------X
\ / \ / \
\ / \ /
\ / \ /
\ / \ /
-----------------------------------
The "eyes" are open, zero crossings are clean and repeatable. The first
"eye" (or loop) has only a single loop since you triggered on the +
going edge. For all succesive loops, the typical data stream as some
cases when adjacent data bits are the same and some cases when they are
different. When adjusting you modem, note that the DC level represented
by the vertical position on the scope can be used to make the zero-
crossings (denoted by X's) occur at the same voltage level as the
clipper's 0/1 threshold. This is the best way to set a modem on
frequency. This is the trick I use for tuning in FSK packets on AO-10.
Now let us look at a poorer eye-pattern. Bad conditions include jitter
in the zero-crossing times, un-equal amplitudes for the case of two
adjacent data bits being the same vs. different, etc. The following
might represent a very marginal demod:
-- -- --
-------------------------------------------------------
/ ----- \ \ ---- --\-------/--/--------\--\--
/ / \ \ \ \ \ / / / \ \ \
/ / \ \ \ \ \ \ X / / \ \ X /
/ / \ \ \ \ \ XXX / \ XXX
!---------!-------\-\-\-------!------XXXXXX-------!------XXXXXX
\ \ \ / XXX \ / XXX
\ \ \ / / X \ \ / / X \
\ \ \ / / / \ \ \ / / /
\ \ -----/-/------\--\-------/-/------
\ -----------------------------------
-- --
Look for "droop" in the horizontal lines (horizontal lines connecting
two "eyes" occur because of identical adjacent bits); such droop could
be indicative of low-frequency response inadequacies in the demod. Un-
equal amplitudes between single loops and multi-bit loops is usually a
sign of amplitude imbalance between low and high tones or reflects bad
predection filter values. Jitter in zero-crossing times combined with
large amplitude excursions is often a sign of bad group-delay equal-
ization in the modem or the radio. You also want to see "open eyes"
with no zero crossing at the mid-bit time. Of course, on weak signals
with lots of noise, you can expect to see more jitter. When you start
seeing transitions in the middle of the eyes, the signal is too weak to
get a good bit error rate.
I'd suggest that those of you trying these tests first look at the
strong packet signals on a local FM channel. They should approach the
"perfect" picture above. As the signal gets noisier, you will see the
Bit Error Rate performance degrade.
When I reported on the FT726 originally, the eye patterns I observed
were even worse than that depicted above -- in fact the eyes were not
"open" and the pattern was not discernable. With the new filter I
reported on about a week ago, the eye-patterns are acceptably "open".
It is hard to describe this behavior quantitatively, but after you have
looked at a few cases, it becomes obvious.
Hope this treatise helps. Keep me informed on what you find. 73, Tom