Part Two

To continue my submission about pneumatic roll reading using strain-
gauge electronic transducers which I began yesterday, here are some
additional technical points about the proposed setup:

                            MINIATURE STRAIN-GAUGE
                   ___      PRESSURE TRANSDUCER (1),
SHORT             |___|<----TO COMPUTER SCANNER (3)
TUBE           ____| |________
FROM          |    | |    _   |<----RAIL DRILLED
TRACKER_______|____| |_ _| |  |     AS SHOWN
BAR-->________ ________0<========D<-----BLEED AND
         _____|__________| |__|______   AJUSTING
        |    ____________| |_____    |  SCREW (2)
        |   |                    |   |
        |   |<--WINDCHEST OR---->|   |
        |   |  VACUUM CHANNEL    |   |

(1) This transducer is an Omega model PX74-0.8DV, consisting of a
strain gauge array printed on a chip, mounted in a regular TO-5
transistor case with two nipples soldered on.  This model can measure
either pressure or vacuum over a 0 to 0.8 psi range (0 to 23 in. WC),
producing an output of 0 to 50 mV full-scale.  It is cheap ($19.00)
and crude, but with the strain-gauge array 'printed' directly on the
crystal, its response is above the kHz range, so its ability to place
the measured pressure change events accurately in time is limited only
by the capability of the electronic measurement device connected to it.

Being cheap, it has no calibrating resistors built in like a more
expensive pressure transducer, so sample variation from chip to chip
produce error.  However, we want to pick out a single pressure change
event for each note, which is calibrated as described below, so this
effect can be corrected for, across the whole scale.

(2) To accurately scan a music roll, we must detect the exact point in
time where some repeatable attribute of the punched hole passes a fixed
point.  I would propose that point to be where the 'D' shaped lower half
of the perforation (either a single hole or the first hole of a 'slot'
would act the same way) just passed the upper rim of the tracker bar
port.  If a bleed of about the same effective area as this 'half punch'
were used, then the vacuum measured in the 'pouch well' at that instant
would be half the stack vacuum.

By temporarily connecting a test orifice of that area to each tube in
turn, and reading the transducer output, one adjusts each bleed with
the screw to produce the same reading for all the perforations.  If a
12 in WC vacuum supply were used, the instant where the 'D' hole first
appears would read 12 mV for every note.

Since the lower edge of the tracker port is some distance from the
upper edge, the off-going transition of the output at the end of a
slot would be displaced in time, resulting in a lengthening of that
particular note record in the software (the bane of all pneumatic roll
reading schemes).  However, this amount should be constant in this
system, and can be subsequently removed from the record by 'truncation
software'.  Robbie knows all about how to do this.

(3) The hard part is how to read the output from almost 100 transducers
into a computer simultaneously, with sufficient speed to make this
scanning method practical.  There are two ways to go about this:

There are commercial multi-channel analog scanning input systems
available, which connect to a computer serial port.  One choice would be
the OMB-TEMPSCAN-1000 with three TEPMV-32, thirty-two channel millivolt
input expander cards.  But we are talking almost $5000 for this setup,
and even at its 9.6 kHz scan rate, the roll would have to move at a
'tempo' of 0.8 (10 times slower than normal) to preserve the desired
resolution.

More information about this scanner, as well as the transducers,
can be downloaded from:

    http://www.omega.com/

A better idea, if one were smart about electronics using FET's (which
I am not) would be to hand-hone 96 differential voltage trigger
circuits which would switch outputs from 0 to say 5 volts at the point
where the input from the XP-74 crossed a selected value (12 mv at 6
in. WC, for example).  Then one could eliminate the analog scanner
altogether, and use the Octet digital input-to-MIDI device directly.
This would cut the price down to less than 3 Kbucks, or less if someone
already owned an Octet, which is likely to be the case.

Richard Vance