This is very good info. Thank you. I think with more modern analysis
(and/or a decent editor like Adobe Audition) it's pretty easy to see
oppertunities to clean up the data using some normalization and maybe a
compressor/expander. So if I were to take a recording of a tape
program and reproduce it as a 100% square wave tone, then I should be
able to play it back through the apple successfully despite how much
"different" it might appear to sound, right?
Though the signal is overall distributed noise, I did find solid
frequency bands at 797.3 (the inital tone). 2070, 4140 and 8280 hz. I
couldn't find a recognizable band at 1045 though. Doesn't matter too
much. Andy's approxomations were pretty much right on. :-) Filtering
these to produce only those frequencies screws up the data because, as
you've pointed out, the rounded signal falls below the 70% threshold
more often and as such the data is misinterpreted. However, the peaks
are at the right spacing so the information is still there, but the
apple doesn't look at peaks, it looks at threshold. This is beginning
to make a lot more sense.
How does the checksum logic work?