Re: [Discuss-gnuradio] HDTV

[Eric Blossom]

On Fri, May 27, 2005 at 01:08:29AM -0400, Ilia Mirkin wrote:
> On Thu, 2005-05-26 at 21:41 -0700, John E. Don Carlos wrote:
> > What is the best way to get hdtv going?
> 
> I don't know if anyone's done anything on this topic lately, but this is
> what I found from my experience on this:
> 
> There is code in gnuradio-0.9 (hint: you don't have to make install
> gnuradio-0.9 -- the code will work out of the compiled source tree)
> which will allegedly take a 20MS/s (short samples) file, where the
> signal is centered around 5.75MHz (as is the case with the TVRX board
> assuming it uses the Microtune 3x7702), and produce a file containing
> the ATSC-encoded samples.

Right.  The non-ported code is also in the 2.x CVS tree.

> The problem is that the USRP can't send 20MS/s over the USB bus, nor can
> it sample at 20MS/s without using an external reference. I think that
> support for the USRP in gnuradio-0.9 is... lacking. Another problem is
> that the code isn't particularly clever -- it needs a rather clean
> signal in order to decode it properly -- all of my attempts ended up
> with 100% errors while my pcHDTV card got the signal on that channel
> just fine.

FYI, my guess is that the equalizer is the weak part.

There's nothing special about the 20MS/sec.  That's just what the
Measurement Computing PCI DAS 4020/12 A/D board put out.  It's what we
had to work with, so we used it.  16MS/sec real baseband should fine
with a couple of tweaks.

> What I did was take the USRP, record the ATSC signal to a file centered
> at 0MHz using DDC (make sure to downconvert in the right direction)
> (compex floats), which allows you to use 8MS/s and the USB bus seems to
> be able to keep up with that, mostly. Then you can write a script to
> convert that to a file of shorts with the signal centered at 5.75MHz
> (e.g. by multiplying it by a complex sine wave).

There's no reason to put it back at 5.75MHz.  Just change the
constand that defines the location that the FPLL is looking for the
pilot tone. 

Since the USRP is currently lacking the half-band filter, the signal
wouldn't have been flat enough (given our equalizer) and will have
aliasing that could cause a problem.  The fix is to make the half-band
filter work in the USRP.  Voluteers and patches welcome.  The guts of
the code are done and in CVS.  The state machine to control it still
needs to be written.

> Now you have a file that the gnuradio-0.9 code is willing to deal with,
> and if your signal is clean enough then you should get an mpeg file.
> This is where I hit a dead end, and thus wasn't really willing to try to
> convert the old code to new code since I had no way of knowing if it
> would work or not.

There are at least three independent existence proofs that the old
code works.  Me, John Gilmore and the folks at UIUC.  A bit of
searching through the archives and/or google will turn up links.

Eric