16 Dec
2007
16 Dec
'07
9:06 p.m.
Robert Nansel wrote:
I've come across a reference to a reverb unit made with a garden
hose, speaker, and microphone. Would something as bone-headed as
this work as a data delay line?
I've read that mercury was used in delay lines because it was a
better impedance match with quartz transducers, but wouldn't water
work nearly as well? Everything would need to be kept at a constant
temperature, and no doubt there would be some dispersion of the
compression waves. It's my understanding that a lot of materials
were tried for delay lines, but that mercury was the "best."
Magnetostrictive delay lines are attractive, though I hear they are
more than a little microphonic. I notice that most magnetostrictive
delay line designs use the transducers to generate torsional waves,
which apparently suffer less from dispersion and have a slower
propagation speed to boot. The pictures I've seen show the
transductive materials spot welded to the delay wire such that they
twist the wire when a magnetic pulse is applied:
http://www.science.uva.nl/faculteit/museum/delayline.html
Does anybody know how well plain nickel wire would work OK, or would
some more exotic like Terfenol-D be required? The delay wire itself
need not be magnetostrictive; it's just the storage medium.
I have three calculators which use magnetostrictive delay lines. Two of
these (same manufacturer, diff models) use torsional waves as you describe
with two bands of MS material going through a double solenoid and spot-welded
tangentially to the delay wire. The other ends of the bands are anchored in
an elastic/rubber material.
The third calc uses compression waves with the delay-wire being MS material
and the ends going through simple solenoids as one would more typically expect.
Although all of these are SSI IC-based calcs, the delay line interfaces are
discrete transistor. The read amps are fairly straightforward multi-stage
RC-coupled amps, 4 stages for the torsional types, 6 stages for the compression.
All these calcs get some hundreds of bits of storage out of the delay lines
over delays in the range of 1mS and clock rates around 1MHz.
The two torsion-mode machines use crystals in the master clock for timing
stability. No temperature compensation for delay is apparent in these
machines, so apparently at those lengths the variability fits within the
crystal tolerances. The machine using the compression-mode line has an RC
astable FF for the master clock but adjusted in a PLL style loop. The other
side of the phase comparator is fed from what appears to be a sync slice (track)
written into the delay line at power-up.
I've sometimes wondered whether things were handled differently in early video
terminals or processors with MS delay lines where the lines were presumably a
fair bit longer. Have to check out the PB250 docs Chuck mentioned.
One of the interesting aspects of the two calcs using the torsion-mode is that
the delay line actually has two write transducers, one at an end and one
part way along the length nearer the read end. Bits are injected at the end
in every second bit-slot, the mid-way write transducer fills those empty bit-slots
to interleave the bits from different registers. It's used to advantage in the
architecture in a manner similar to the way drum machines would have multiple
heads for lower latency of the main registers. The interleaving techniques in
bit-serial machines can be quite novel.