Of course Julian Dyer [in 200907 MMDigest] is right that recovering
the punch matrix is an important step in creating a high-quality
emulation of a music roll, contrary to Stanford's stated position.
However, I take issue with Julian's assertion that punch matrix
reconstruction is complex and time consuming.  In fact, it can done
simply and quickly, and there is no reason not to do it if possible.

Not all music rolls lend themselves to punch matrix reconstruction.
The deciding factor is the degree of phase noise, which is a measure
of the regularity of the row advance.  If the advance is highly
uniform, it is always possible to reconstruct the matrix, but if it
is highly irregular, reconstruction becomes impossible.

I have found that the best rolls in this regard are Welte-Mignon
DeLuxe rolls.  In contrast, it is never possible to reconstruct the
punch matrix of Welte-Mignon T-100 rolls because of their fine row
advance (on the order of 0.50 mm).  Ampico rolls manufactured before
the 1932 merger with Aeolian present no difficulties, but post-merger
rolls made on the converted Aeolian perforators are problematic because
they contain many short rows sprinkled throughout.

Coarse-advance Duo-Art and Artrio-Angelus rolls (254 rows/foot) were
perforated on Aeolian equipment and do not present a problem, but
later fine-advance rolls (381 rows/foot) apparently made on the same
machinery are difficult to work with, and the punch matrix cannot
always be recovered.

However, much of the benefit of punch matrix reconstruction can be
had for emulations, even if reconstruction is not possible.  This is
a consequence of the two-step process required for reconstruction.

In the first step we find the static skew (overall tilt of the punch
line), the dynamic skew (variation of the tilt along the length of the
roll), and scatter (channel-to-channel variations across the width of
the roll).  After determining the errors we remove them, yielding an
improved roll image.  It is only in the absence of these errors that
we can assign a row number to each perforation, that is, reconstruct
the punch matrix.

The first step brings about a large improvement in the quality of
the scan image.  Given this improved image, the second step may or
may not be able to assign rows correctly, but for the purpose of
emulation row numbers are not necessary, only event displacements.
If the displacements produced by the scanner are sufficiently accurate,
the first step provides most of the gain in the quality of emulations;
the additional improvement due to the second step is much less.

I invented punch matrix reconstruction in the early 1990s, and I have
made the program suite that I developed for that purpose freely
available since the mid-1990s.  I have continued to make advances in
determining and removing errors in the roll and scanner (the goal of
Step 1) between that time and this, in response to what I and others
have learned from scanning many rolls of various types over the
decades.

The current versions of my programs are available free of charge to
all interested parties, as they have been all along.  These programs
are in daily use by a number of scanner owners, who rely on them for
punch matrix reconstruction.

Wayne Stahnke
Reno, Nevada
wayne@live-performance.com.geentroep [delete ".geentroep" to reply]