[dtvmcdonald]

As to actually converting the digital signal to a form that is equivalent to
a Windows .bmp file for each frame (assuming a progressive format) ... its
not easy! In fact, its probably impossible since there are "gotchas" in
the public description of the signal format intentionally
put there to make it impossible without paying tribute.

I'm talking ATSC signals here.

There are two parts: first you have to digitize the analog signal.
This is easy. Its normally done at the IF frequency with subsampling
(i.e. if the IF is 44 mHz you digitize at say 15 or 16 MHz sample rate).
This is done in quadrature so the resulting "positive" and "negative"
frequencies can be distinguished. Note that this is a heterodyne that
leaves the carrier (ATSC has a carrier!) at a couple of MHz.

Everything else is done in software.

Next, and hardest, you have to take the signal
and de-ghost it. The descriptions of how the signal works in the
public documents are good enough for this. What's hard is that
mathematically it is a difficult problem. Normally its done using
adaptive Weiner filters but each implementation is highly proprietary
and there is "magic" involved. I own a couple of patents that
show how to do it using Fourier transforms and adaption in the frequency domain.

Next you have to demodulate (the signal is digital, but its not yet baseband).
To do this you have to phase lock the carrier. This usually is trivial ...
BUT ... though the carrier was transmitted IT MAY NO LONGER BE THERE ...
there could be a ghost at exactly 0dB level and exactly 180 degrees phase
which cancels. This is guaranteed to happen sometimes, it called Murphy's Law.
Fortunately there is a trivial method that regenerates the carrier even if missing!
(I should add this ATSC is vestigial sideband AM ... if it were quadrature AM
as used by cable this would be no easier ... in fact the difference, to the software,
is trivial). Now you have baseband.

Conceptually these two steps are separate and done in this order ... but in
practice they are inside a feedback loop that makes the de-ghosting easier.

Once you do that, you simply take the (already digital probably 32 bit floating point) signal and
"quantize" it into the 8 levels of ATSC (well, its a bit more complicated
because of sync pulses that are not at those levels.) You then have an
error-corrected MP3 with meta-data (e.g. frequency mapping and program guide).
The rest, using to error correction to get
plain MP3 ... is straightforward ... but this is where
there are ambiguities in the spec.

As to using a commercial box at a fixed frequency with a pre-converter
this simply cannot be done, as the boxes presuppose regular frequencies
and use digital information in the signals to "remap" channels, and this
remapping can't be turned off. Try it and the box gets confused. That
is true for all the boxes I'm familiar with ... there is no reason at all a box can't
be built that does not do it.