Hi Eric, I just saw your cool gnuradio presentation and heard you talking about a not yet realtime ATSC transmitter. I am a telecommunication engineering student in italy at Pisa University (Italy) at the moment, and it is quite likely that my final thesis (to be started in about a year from now) will be a DVB-T transmitter. So I'm slowly approaching the subject, and starting to figure out some preliminary choices and possibilities... My first questions then are : ..Do you think it is possible to reach real time, considering hardware improvements since the video was recorded? ..How could I test a non real time transmitter? (I think I understand not real time receiver but how did you test the transmitter?) Best Regards
on 2007-03-05 09:18
on 2007-03-05 09:52
Vincenzo, I did some experiments with DAB which is, (due to differential modulation), simpler than DVB. On the PHY level I managed to get "almost" (about 60%) real-time performance on my P4 with GNU Radio and prerecorded samples. This was only PHY (even without channel coding!), and the synchronization algorithm used was pretty shaky. Channel coding eats up a lot of resources as well, look at the throughput figures for gr-trellis. DAB has a bandwidth of 1,5 MHz, DVB more than 5 MHz. All in all considered, my guess is that 5-8 years from now it might be possible to do these kind of bandwidths on a general purpose processor. But this is just my unjustified guess. :-) When starting to work on DVB, you'll have to write a transmitter chain and a receiver separately, to you can test both. Just look at the GNU Radio examples. I would suggest to limit your final thesis to the PHY layer first - OFDM synchronization by itself is a real challenge. What I noticed is that, due to the way transmission protocols are designed, there is not "one solutions fits all" approach. You would have to tailor available OFDM code to the specific needs of DVB (DAB). A good way to start might be the new GNU Radio OFDM code available from trunk. A good way to "get a feeling" for real data might be to look at it in Matlab or Octave. You could test signal processing algorithms in such an enviroment as well. Best regards, Jens > improvements since the video was recorded? > > ..How could I test a non real time transmitter? (I think I > understand not real time receiver but how did you test the > transmitter?) > > > Best Regards > -- > Vincenzo P.
on 2007-03-05 16:28
On Mon, Mar 05, 2007 at 08:18:06AM +0100, Vincenzo P. wrote: > My first questions then are : > > ..Do you think it is possible to reach real time, considering hardware > improvements since the video was recorded? I could be. I'm unfamiliar with the DVB-T spec, so can't really offer an opinion. Transmitters are always easier than receivers. One of my goals is a port of GNU Radio to the Cell processor. I expect that we'll be able to realize about 30 GFLOPS on that platform, which is about 10x what we can get out of a commodity PC today. On the Cell and with some optimization work on the ATSC tx and rx, I suspect that both can be made to run in real time. > ..How could I test a non real time transmitter? (I think I > understand not real time receiver but how did you test the > transmitter?) Write the generated samples to a file, then "play it back" in real time from the disk. With the ATSC work, we wrote the transmitter and receiver at the same time, then looped them back at compatible places in the pipeline. > Best Regards > -- > Vincenzo P. Eric