[Discuss-gnuradio] Some NTSC Results

[Prateek Dayal]

Dear All,

Following up the small NTSC test program that was mailed to the list,
I have written some code to demodulate NTSC frames. Unfortunately to
do it fast, I implemented these algos in Matlab. I will implement this
in python soon, but I was too excited to share some results and
observations.

Currently, I am able to locate the horizontal and vertical pulses and
build a frame out of the two fields. I have still not done Y/C
separation. Here is the sample output

http://geocities.com/pmd_iitgw/base.jpg

The complex samples that Eric gave me had video carrier at -1.75 Mhz
and audio at 2.75 Mhz. Like the previous examples, the above frame was
generated by bringing the video carrier to 0 Hz, low pass filtering
and then working on the magnitude of this waveform (envelope
detection). Sync pulses were detected by a correlation method which
exploited the difference in duration of horizontal and vertical
pulses. So much for the results. Below are some of my observations and
suggestions and I would be very thankful if you could give me some
feedback.

The NTSC samples contain and audio and a video carrier. We would like
to demodulate both finally. If we perform a downconversion, we slow
down a lot. An alternative would be to  flip the spectrum while
quadrature sampling it and bring the audio carrier to 0 Hz. Then the
video carrier should be at about 4.5 Mhz. Now the FM can be demoulated
directly after low pass filtering and decimation. So we save one
downconversion from 2.75 Mhz --> 0 MHz. Also to demodulate video,
interpolate the samples to 2*input_rate, then do a complex_to_mag and
then decimate the samples back to input_rate. This interpolation is
necessary to avoid aliasing when doing a complex_to_mag. The low pass
filter applied when decimating will give us the NTSC baseband
waveform, which we can sample after sync detection to form the frame.
Thus we save one more downconversion from -1.75 --> 0 Hz. Thus WE DO
NOT HAVE TO PERFORM ANY DOWNCONVERSION to demodulate both audio +
video . I tried this approach on the original samples (without
downconverting or anything) and here is the result

http://geocities.com/pmd_iitgw/not_base.jpg

As you can see, the result is not much different. So it works !!!

Thank you for reading this very long post. Looking foward to your comments,

-- 
Prateek Dayal
B.Tech 4th Year
ECE
IIT Guwahati

www.geocities.com/pmd_iitgw