The new Cunning Plan *
Background:
Real world communications cables depart from the ideal in they
attenuation of the signal entering one end, and exiting the other, and
they disperse the signal, i.e. the frequency dependent distortion of the
signal exiting the cable. When communications pairs are in close
proximity to each other, they exhibit another departure from ideal in
that they couple energy from one to another by virtue of departures
from perfect mechanical symmetry. If the cross-talk occurs
at the end of the cable that is being driven, it is called
'near-end'... if it occurs at the load end, it is called
'far-end'.
Near-end cross-talk coupling ratios in the range of -30 dB to -70dB are
produced by various manufacturers. The weakness of the coupling
reflects the measure of quality control of symmetry and uniformity
during manufacture.
Until now we have been able only to measure near-end and far-end
crosstalk in 0.7mm (wire diameter) Ericsson quad cable in a Lab
environment. In the lab, the entire length of quad is wound on a
spool. The assumption was that so many turns of wire wound on the
spool will have an effect on the measurements because some systematic
magnetic or capacitive coupling exists which manifests itself as
excessive cross-talk. In an attempt to prove this assertion,
it will be helpful to make cross-talk measurements on a cable
in-situ.
The Ericsson quad from which the String 18 cable was built consists of
0.7mm diameter wire, surrounded by a foam dielectric layer with a solid
outer skin. Four of these insulated wires are twisted around a solid
core a few tenths of a mm in diameter, with an over-all black jacket.
Some of the quads in the main cable spiral around the high
strength central core in a clock-wise direction on the inner layer, and
the remainder spiral around on top of them (along with some service
pairs). This bundle is confined within an over-all jacket of
durable, flexible plastic.
The in-situ measurement differs from the lab measurement in that the
quad lays adjacent to a different quad instead of itself displaced by
some fraction of a meter.
Some communications performance measurements (Jan 30, 2001) have already been made with
cables of AMANDA string 18 using the DAQTB test boards which were
installed in January 2001. The near-end cross-talk was
found to be less than 1mV, and consistent with the noise level of the
input amplifiers of the oscilloscope. In this case,
two of the four wires in a quad (e.g. the A-C pair) were driven
by the communications modulator to a level of a volt or so. The
neighbor was the B-D pair in the same quad.
Likewise, the DOM at the far end of the cable communicating on the A-C
pair was caused to generate modulation, while the neighbor pair (B-D)
was monitored. Far-end cross-talk also appeared to be below
the electronics noise-floor.
Objective this time around:
Concern has been expressed that the high pressure (3000 psi to 7000
psi) compressing the cable during freeze-back and even after the ice
'relaxes' will disrupt, for the DOMs themselves, the excellent
symmetry, resulting in a deminishment of the performance exhibited by
the Ericsson cable. (This skepticism is usually expressed by
Ericsson's competitors when considering their own bid on cable
manufacture for future deployments, because the Ericsson cable utilizes
a foam dielectric construction which the high pressure will cause to
collapse). AMANDA analog modules utilizing both 0.7mm quad and
0.9mm quad, however, are something of an existence proof that
near-end and far-end cross-talk are not greatly affected by the
compression.
The hardware in the ice is flexible enough to be reconfigured so that
both near-end and far-end cross-talk measurement data can be collected
into chunks 4 megabyte long.
This is how, subject to agreement of all concerned:
Two new experimental 'configurations' will be needed to cover all the
bases. Both depend upon altering the firmware in a string 18 DOM
so that, when triggered, the communications ADC data is written into
look-back memory until it is full.
The first configuration requires that one of the DOMCOM board FPGAs be
reprogrammed to generate a repetitive data pattern as soon as it is
configured. This is the source of far-end cross-talk signal that
the DOM in the ice will capture, where that DOM may be sharing the same
quad, or another quad in the cable. Intra-quad and extra-quad
far-end cross-talk can be measured.
The second configuration requires that one of the DOMs have its FPGA
programmed to generate a repetitive data pattern as soon as it is
configured. This is the source of near-end cross-talk signal
that the DOM in the ice will capture, where that DOM may be sharing the
same quad, or another quad in the cable. Intra-quad and extra-quad
near-end cross-talk can be measured.
The infrastructure already exists to read back a 4 megabyte block of
data from the look-back buffer.
The DOM can be put into a mode where it captures communications data in
response to writing a bit in the FPGA's API, or the DOM can acquire
communications data immediately upon the completion of FPGA programming.
The Data Pattern:
The look-back buffer can hold 4 Megabytes of communications data, which
can be written at a rate of 16.8 megasamples per second (the local
oscillator frequency). Each baud period is 24 clock cycles long.
350 kbaud communications alternates baud-period windows with
'quite periods' of 24 clock cycles. A 350 kbaud pattern
256 ASCII characters long (the complete alphabet) could be
repeated/collected 34 times into the 4 megabytes of look-back memory.
The writing of 8-bit data is justified because the cross-talk
(near-end, or far-end) is already known to be at least 24 dB
below the signal received in a pair. 6-bit data would be
adequate!
Analysis:
The data block could be cut into contiguous blocks (10*48*256) =
122880 samples long, which would then be added, vector wise, to each
other, to produce a noise-suppressed, average waveform.
Since the least count of the ADC is 2 mV, averaging 32 times
improves the effective sensitivity by a factor of about 5,
after taking the affects of averaging into account.
The peak-to-peak amplitude of that waveform is the measure of the
cross-talk induced into the neighbor.
Refinements:
It will, in my view, be useful to capture cross-talk waveforms for AMI modulation waveforms packed more closely together,
i.e. 1/2 baud period spacing, and immediately following each other.
Likewise, it will be useful to capture the received waveform on the
main communications pair for higher data rates. The upper 8 bits
of the data would have to be stored, which implies a second variant on
the comm-logger plan. This second variant could be actuated by an
API bit.
Higher modulation rates can be incorporated into the transmit pattern
generator. For instance, the packet can contain 256 bytes at 350
kbaud, 256 bytes at 467 kbaud, and 256 bytes at 700 kbaud.
The pattern would repeat every 276480 bytes, or 15 times in 4
megabytes.
If, however, this longer pattern were written to the full 4,718,592
bytes of look-back memory, the pattern would repeat 17 times. The
effective sensitivity of the communications receiver would be improved
by a factor of 4, after accounting for statistical affects.
Glossary
AMI Alternate Mark Inversion modulation, a form
of RZ (Return to Zero) modulation, characterized by pulses of equal
width of alternate polarity, one following immediately after the other,
and inverting polarity, from one baud pattern to the next.
Miscellanea
* Apologies to Rowan Atkinson... "Cunning plan"
is a term used by Baldrick in various episodes of Black Adder
gtp April 17, 2003