[inforoots] Recovering data from old media (Was: "The digital Dark Age")

[Mike Albaugh]

 Dick Weaver wrote:

> As I understand it the significant problem with most magnetic tape is 
> the binder (tapes without a binder, "metal evaporated", have other 
> problems).  See, for example,
> 
> http://www.clir.org/pubs/reports/pub54/2what_wrong.html

 Which is indeed a nice summary, but toward the end my spidey-sense
tingled a bit. "Analog" magnetic recording is not as "smooth" as
the article implies. The audio/video is still "sampled", it's just not
quantized. (PAM Time-division Multiplexing of telephone calls
is another example). A nit, but It left me wondering what else
had been glossed over.

Non geek/pedants can skip the rest.

An audio tape-recorder has what is mis-named a "bias"
oscillator. The purpose of this is essentaily the same as
the dither sometimes added to high-resolution A->D convertors
to trade sample-rate for more resolution. As the analog signal
increases in amplitude, the sum of it and the bias more quickly
(in each bias cycle) reaches the point where the domains
are flipped (there are no "partially-flipped" domains). 

On playback, the changes (and there must be changes, to
generate any signal at all in the head) are essentially integrated.
So, an "up", followed by a long pause, followed by a "down"
becomes a large positive signal. That same "up" followed by
a short pause and a "down" is a smaller (or negative, once
DC-corrected) signal. This sampling becomes relevant when
you notice that it, like typical A->D conversion, is subject to
aliasing. Your bias has to be at least twice the highest
frequency present in your signal (NOTE: not "the highest
frequency of interest". a 63 KHz signal will become
painfully "of interest" in a system with 128KHz bias) :-)

These concerns also show up in digital recording. GCR, for example,
was designed to ensure enough transitions to allow (per track)
clock recovery. The head can't "see" constant magnetization
(A Hall sensor could, but we're talking normal coils here :-)
IBM's "Zero Modulation" went further, to eliminate the DC
component in the signal, which was AC coupled. Takes a
bit more processing, "looking backward and forward in time"
to do, than simply summing a 128KHz signal with your mic :-)

 Mike, geek/pedant