CORDIS Archive
This page has been archived. It will no longer be updated.
MPEG2 compresses by removing redundancy, both within individual TV frames and between successive frames. Individual frames are split into blocks, each of which is then processed with a DCT (Discrete Cosine Transform). This, in effect, converts each block into a set of coefficients that define its spatial frequency components. Additional compression is achieved by discarding coefficients with a small value (i.e. whose contribution to the overall picture is small).
Further compression is available by exploiting the fact that the change in picture content between one frame and the next is usually quite small. Therefore, compression can be achieved by transmitting only the changes that occur between one frame and the next. In fact, the algorithm predicts what subsequent frames will contain, and then transmits error signals that indicate how the 'real' picture deviates from the predicted picture.
To gain an idea of the efficiency of MPEG2 compression, consider the following. Recommendation CCIR 601 specifies a scheme for digitizing component-quality (i.e. luminance and colour information transmitted separately) television signals at 216 Mbits/sec. An HDTV signal digitized in this way would require about 1 Gbits/sec. After compression, the respective bit rates are 10 Mbits/sec and 40 Mbits/sec. Still further reduction in bit rate is possible if picture quality can be sacrificed; but even a 5 Mbits/sec compressed PAL signal compares well with a VHS videocassette.
The nature of MPEG is that the signals cannot be mixed easily. With ordinary (non-compressed) video signals, it is a straightforward matter to cut between one picture source and another, accurate to the resolution of individual frames. With a compressed picture, however, the signal that makes up a particular frame will probably have components of previous frames, and the operation becomes non-trivial.
Three existing approaches to mixing MPEG2 video signals all have their shortcomings. The first, Naive Cascading, fully decompresses the signal, mixes it with other sources and then re-compresses it. The disadvantage is that in real-life broadcasting, the signal would probably be mixed several times, and each process degrades the picture quality.
The second method, Restricted MPEG2, uses a limited subset of MPEG2 encoding, and this allows something approaching frame-accurate editing to be achieved. The penalty is that it is incompatible with the mainstream MPEG2 encoding, and is therefore only really suitable for closed systems. The third method, Bitstream Splicing, makes the cut between one picture source and another at precisely the right moment, carefully avoiding the creation of processing artefacts. This has the disadvantage that frame-accurate editing is not possible, neither are gradual cross-fades between pictures.
The ATLANTIC approach is to preserve the side information, and to pass it on to the re-encoding process. In this way, the encoding process does not have to spend precious processing power re-creating the side information, and picture fidelity is preserved. Meanwhile, the de-compressed video signal is passed on to a video switch and processed in the usual way.
The side information used in this way is called a mole, because is 'burrows' through the studio equipment automatically emerging at the re-encoder. The distortion inherent in this process is very small; mathematically, the distortion is calculated to be of the order of 0.0002dB, whereas multiple naive cascading has a distortion of typically 5dB.
By Adrian Rawlings, Euronet Associates Ltd, 25.08.1997