Spencer Chase wrote in Digest 961128 that he is designing a  roll
scanning and punching system, and that he plans to punch the new
rolls "at the same step size as the originals to maximize speed and
accuracy."  In Digest 961204 Spencer wrote:

> I would like to scan rolls with a degree of precision that
> captures the information as accurately as it was punched.

That's great, Spencer, and as Robin Pratt would say, "Let's do it right!"

Several people have made music roll copying systems in the last 40
years that I've been following this topic.  Most of the time the recuts
were okay for orchestrions, and passable for 88-note pianos, but the
crude methods just couldn't maintain the precision needed to copy
reproducing rolls.  The Ampico recuts by Orville Cooper, using a single
punch with a single-channel "reader", were miserable!  Then Larry Givens
started producing "microfine" recuts (more than 60 steps per inch,
asynchronous) and the results were generally okay.  _But not perfect._

A couple of years ago I visited Siegfried Wendel's museum in Rudesheim
am Rhein, and he also showed me his music roll copying machine.  Siegfried
has a problem:  because the instruments in the museum are played several
times daily, he cannot keep original rolls on the machines, and so he
must make copies.  His semi-automatic single-punch machine works not
much better than Cooper's did:  it's okay for orchestrion rolls, but
not at all satisfactory for the reproducing piano rolls.

Siegfried is a fine engineer, and a canny businessman.  He might decide
that it isn't worth spending _his_ time to fix the process, and instead
he might engage a smart machine builder to get the job done.  So he
goes to a clever machinist friend, who has worked with Siegfried
previously, and who says, "Ja, Siegfried, but you are a difficult
customer to satisfy.  How shall we both be satisfied that I have
done the job properly?"

These guys, the machinist and the customer, aren't making machines
and copying rolls for fun -- they are both spending time and money
on a serious job.  They know very well that the task must be completely
specified, and that both parties must agree upon the conditions which
define that the job has been done acceptably.  You can be sure that
they will "do it right".

My illustration above is fictional, but it is probably so close to the
truth that Siegfried will wince when hears of it!  My illustration
following is also fictional, but I think it illustrates how an
engineer/businessman like Siegfried would proceed.

The task:

The task is to replicate a standard 88-note music roll, or perhaps the
"98-channel" versions such as Ampico or Welte Licensee.  That the data
is music is inconsequential: the task is only to duplicate the pattern
of holes which are punched in a paper roll.  A skilled machinist and
machine builder, who has no knowledge about the _use_ of the paper
roll, will need the information discussed below.

The most important aspect is the definition of an acceptable product,
so that the machinist will be paid by the satisfied customer!  The
customer describes the problem to the machinist, and they write a job
specification which is agreeable to both.

Acceptance Criteria:  The replica is declared acceptable if

1.  The pattern and number of punched (overlapping) holes matches the
original; this will be verified by laying the copy over the original
and inspecting visually.

2.  The dimensions are within the specified tolerance.

Facts:

1.  The spacing of the holes on the short axis is constant, 9 per inch,
and the centers of the holes lie upon a straight line which is
perpendicular to the long axis.

2.  The spacing of the holes on the long axis can be from 240 to 400 or
more holes per foot, but it is constant for a given roll.

3.  All holes are the same diameter.

4.  The paper is 11.25 inches wide; the length is indeterminate.

After some discussion these inspection tolerances are agreed upon:

1. Hole diameter: +/- 0.002 inch¶
2. Paper width and short axis (note axis) dimensions:
+/- 0.012 inch non-cumulative¶
3. Long axis: 0.030 inch in 10 inches, therefore 0.3 percent.¶
4. Skew: The hole centers will lie within +/- 0.004 inches of the true
perpendicular line.

The customer leaves and the machinist begins designing.  It will be
easy to make a paper transport for his computer-controlled punching
machine -- no problem.  But how shall he "read the data" of the
original old paper roll?

After some thinking, the machinist builds a simple roll transport with a
"reader head" consisting of 98 photodiodes which read the short-axis
data.  The paper roll wraps around a precision measuring roller which
registers the distance along the long-axis each 0.001 inch.  (For this,
the machinist uses a shaft encoder transducer similar to the devices on
his numerical-control milling machines.)

The on/off signals from the photodiodes are sampled every 0.001 inch
(0.25 mm); each _change_ in the state of the channel is sent to the
computer, accompanied by the distance measurement output from the
measuring roller transducer.

The computer program is told beforehand that the original roll was
punched at approximately 360 steps per foot, and so it looks for
changes at roughly these intervals.  "Clusters" of change events which
occur together are grouped together (a "chord").  After grouping the
events in this fashion the computer processes the data again, to
determine the actual steps per foot of the original roll, and finally
the computer assigns the clusters to discrete output distances at 360
steps per foot.  It is a _synchronized_ copying process.

The machinist punches a copy of the roll from the data file created by
the reader machine.  He and the customer verify that the new copy meets
the Acceptance Criteria, and everyone is happy.  No musical parameters
were needed -- they simply copied the holes.

Epilogue -- Questions and Answers:

  Q:  "Why did you require hole-for-hole replication?"
  A:  "Because we know that if it matches hole-for-hole, we did the
job perfectly."
  Q:  "Is there an easier method?"
  A:  "We couldn't think of another method which would yield
acceptable results and also be _measureable_.  We had to have
a simple and conclusive acceptance method."
  Q:  "What was the most difficult task?"
  A:  "Defining the acceptance criteria!"

Robbie Rhodes