Short question: Why can MPEG-4 compress video so much more efficiently than MPEG-2?

[teko77]

Hi!



I am writing a short paper in regard of using alternative
compression-methods for DVB-T and suggest the usage of MPEG-4 instead of MPEG-2.

I have now browsed a ton of webpages, but cannot find a proper answer to the question above. My main concern is that while MPEG-2 uses 4 Mbit/s, MPEG-4 (I am only familiar with divx though) is able to give accomplish similar videoquality at 800 kbit/s. How in earth does it do that? I need one chapter of text in regard of the issue and what I have found so far is, that the I-frames are being used in MPEG-2 much more often than in MPEG-4 and perhaps that explains a bit about the difference in the usage of transmisson capacity.

But please - how would you describe the reasons for the significant difference between MPEG-2:s and MPEG-4:s needs for transmission capacity in order to reach the same quality-level of video on a regular TV-set?

Thank you for any help,



Teko77

[redcane]

MPEG-4 incorporates everything mpeg2 does, plus new advanced techniques that save even more space. Things like motion estimation and object recognition.... I don't know technical details..

Basically mpeg4 is a much newer standard, with a lot of newer ideas, look up how mpeg4 works, then compare to mpeg2, and you'll get a good idea of why mpeg4 works better.

[Zarxrax]

Basically, the algorithms that it uses to compress the video are far more advanced. Also it can't get similar video quality at 800kb/sec, it needs more like 2,000kb/sec.

[hakko504]

First of all, for broadcast you need fullframes transmitted, at least for compatibility. This means that you can't go below SVCD resoluition (480x576) or quality won't be acceptable. Try make DivX at this resolution and see what you get with bitrate ~800kbps.
Secondly, and much more important, is as you say I-frame ratio. In order to change channels reasonably fast you must have I-frames closer than one every second. MPEG-2 broadcast uses one I frame every 12 or 15 frames. DivX/XviD are usually set to one I-frame every 300 frames.
Why do we need to have I-frames so often? The time it takes to switch channels, from one channel on one frequency to a channel on another frequency, is limited by the time it takes to find an I-frame. When a reciever starts to change channel it must first change the analogue frequency. This takes 0.2 s and is the same in analogue TV. In an analogue TV though, the picture will be shown on the TV as soon as it finds the frame synch, usually with the next frame, or 1/25th of a second. Total time =0.2+1/25=0.24 seconds. In a digital reciever, the demodulator must find the digital signal among the analogue noise first. This takes (with reasonable bitrates/DVB-T) less than 0.2 seconds. Then the MPEG-2 decoder must find an I-frame befor it can start decoding the picture. With an I-frame ratio of 15 this takes a maximum of 15/25 s. Total maximum time =0.2 + 1/25 + 0.2 + 15/25 = 1.04 seconds. On average though, frequency switching time is constant, 0.2s. Frame synch in TV is half maximum time = 1/50 = 0.02s. Demodulator alignment is ~0.05s. And, average time finding I-frame is half maximum time, 15/50 = 0.3s. This makes the average time = 0.2 + 0.05 + 0.02 + 0.3 = 0.57 ~= 0.6 seconds. There is a factor of the response time of the STB, but in theory this should be the average time needed to switch channels. This time is already considered to be too long by the average customer, that is used to (near) instant response from the analogue TV's, and therefor the I-frame ratio should be changed towards more I-frames rather than the opposite.
A rather interesting approach would be to have a frame that both is encoded as a I-frame and P-frame.

[int 21h]

Offtopic chitchat.

[teko77]

Ok, thank you for the information. About the bitrate, it seems I stand corrected.




Teko77