[mythtv] Frame numbers in DVB (and perhaps HDTV)

[Mike Wilcox]

Hi,

I'm using DVB-T as my picture source, and have a couple of enduring 
problems, which I have a hunch may be a little connected. I'm gradually 
trying to debug things, and perhaps someone can help shed a bit more 
light...

1st is that, after editting a recorded program, the frame-numbers in the 
cutlist don't match up with where external programs see things. Myth's 
count seems to be too high - perhaps 3-4 minutes worth in a 30 minute 
program (but *not* regular).

The first hint is that the position map in the DB shows a fair number of 
keyframe intervals of 13. My experience here in the UK is that the 
interval is usually 12, but occassionally has I frames closer.

What appears to be happening is that the streaming/state parser in 
dvbrecorder.cpp recognises some extra "picture start" markers, so 
increments the framesWritten counter. I have altered the code, so that 
the framesWritten counter only increments if the "picture start" marker 
exists *and* the picture type (2 bytes later) indicates an I-frame, 
P-frame or B-frame, and creates a warning otherwise.

Now, this works, in that the position map in the DB now seems to get the 
right values.

I don't know *why* the parser is recognising frames incorrectly. Perhaps 
my code will help someone with MPEG2 experience provide a better answer! 
Debug code attached - *not* a patch proposal.

With this patch in place, I still find that the frame numbers in the 
cutlist don't match the external program, but the file offsets (found by 
looking up the frame number in the position map) *do* match. Still 
working this out...


The 2nd problem is only with live TV, not recordings. I find that I 
regularly get the video playing in a jerky manner. Audio sounds fine, 
but the video plays in half-second long jerks, and usually with warnings 
about prebuffering.

A workaround to this problem is to pause the live TV for a second or 
two, then let it run again. Video will then run smoothly for quite a 
while, but will eventually go jerky again. Pausing briefly gets it 
working for a while...

Now, I note that as I pause live TV, the overlaid indication about how 
far behind I am is gradually getting further back in time - and much 
more than the amount I pause the live programme for. As this time 
indication comes from a calculation of frame counters, I'm wondering if 
this is related to the above - Perhaps one side is getting it's frame 
counting wrong, and not providing enough (real) frames to the other for 
decoding. No solution, but it's another area to investigate for the 
known "stuttering" problem.

Cheers,

   Mike

-------------- next part --------------
Index: libs/libmythtv/dvbrecorder.cpp
===================================================================
RCS file: /var/lib/mythcvs/mythtv/libs/libmythtv/dvbrecorder.cpp,v
retrieving revision 1.30
diff -w -c -r1.30 dvbrecorder.cpp
*** libs/libmythtv/dvbrecorder.cpp	30 Mar 2004 06:06:53 -0000	1.30
--- libs/libmythtv/dvbrecorder.cpp	4 Apr 2004 15:43:53 -0000
***************
*** 642,647 ****
--- 642,652 ----
          unsigned int state = 0xFFFFFFFF, v = 0;
          int prvcount = -1;
  
+ 	// MPW
+ 	int findFrameTypeCount = 0;
+ 	int lastIframe = 0;
+ 	int diff;
+ 
          while (bufptr < &buffer[pos] + datalen)
          {
              if (++prvcount < 3)
***************
*** 649,654 ****
--- 654,691 ----
              else
                  v = *bufptr++;
  
+ 	    // MPW
+ 	    if (findFrameTypeCount > 0) {
+ 	      if (findFrameTypeCount == 1) {
+ 		int type = ((v>>3)&0x07);
+ 		switch (type) {
+ 		case 1:
+ 		  //WARNING("  I-Frame");
+ 		  framesWritten++;
+ 
+ 		  diff = framesWritten - lastIframe;
+ 		  if (diff != 12) {
+ 		    WARNING("Frame " << framesWritten << ": IFrame gap of " << diff);
+ 		  }
+ 		  lastIframe = framesWritten;
+ 
+ 		  break;
+ 		case 2:
+ 		  //WARNING("  P-Frame");
+ 		  framesWritten++;
+ 		  break;
+ 		case 3:
+ 		  //WARNING("  B-Frame");
+ 		  framesWritten++;
+ 		  break;
+ 		default:
+ 		  //WARNING("  v = " << v << ", frame type = " << type);
+ 		  break;
+ 		}
+ 	      }
+ 	      findFrameTypeCount--;
+ 	    }
+ 	      
              if (state == 0x000001)
              {
                  state = ((state << 8) | v) & 0xFFFFFF;
***************
*** 658,663 ****
--- 695,703 ----
                  if (state == SEQ_START)
                      wait_for_seqstart = false;
  
+ 		// MPW
+ 		// WARNING("state = " << state);
+ 
                  switch (state)
                  {
                      case GOP_START:
***************
*** 682,688 ****
                      break;
  
                      case PICTURE_START:
!                         framesWritten++;
                      break;
                  }
                  continue;
--- 722,730 ----
                      break;
  
                      case PICTURE_START:
! 		        // MPW
!                         //framesWritten++;
! 			findFrameTypeCount = 2;
                      break;
                  }
                  continue;
***************
*** 690,695 ****
--- 732,743 ----
              state = ((state << 8) | v) & 0xFFFFFF;
          }
  
+         // MPW
+         if (findFrameTypeCount > 0) {
+           WARNING("  Finishing buffer with findFrameTypeCount > 0");
+           framesWritten++;
+         }
+ 
          memcpy(prvpkt, &buffer[len-3], 3);
      }