It is well-known that the current TV production workflow involves several encoding and decoding cycles throughout its lifetime, from the content capture to user visualization. Typically, the content from news event coverage (ENG or EFP) is commonly recorded into high quality production formats such as DNxHD, ProRes, AVC-I, or XDCAM HD, or it is live transmitted to TV studios through contribution ENG networks, using different compression standard such as MPEG-2, H.264 and in middle term the emergent HEVC.
Probably, the content will be re-encoded in the editing and post-production process, and encoded one further time in a different compression format in order to be stored on video servers for playout. The last re-encoding stage is applied to content at the broadcasting process, according to DVB standards.
This multi-generation process causes quality losses which degrade the content in a progressive way, due to the use of intrinsic lossy compression schemes, but also due to side factor including colour space conversion, 4:2:0 chroma subsampling, image resizing, and different GOP and compression data rates. Consequently content quality at TV facilities should be kept as high as possible in order to achieve an appropriate quality at the end user.
One of the most widespread practices is to use the 4:2:2 chroma subsampling for contribution and editing processes, which ensure high colour fidelity along the entire production process. On the other hand, the content distribution stage uses lower colour resolution by means of the 4:2:0 chroma subsampling format, since no more digital processing is commonly applied. Figure 1 depicts a general TV production scenario with two concatenated contributions services using the 4:2:2 chroma subsampling format, and the delivery and broadcasting services through DTH, DTT or OTT networks supporting the 4:2:0 chroma subsampling.
With the aim to quantify the quality losses of several multi-generation stages, we have run some simulation with three re-encoding cycles using the “High 422″ profile of H.264/AVC. Figure 2 shows the quality results (average luminance and chrominance) of the three generations obtained for a wide range of bit rate over a HD content. It can be note that the quality is decreased in each re-encoding, especially for the usual contribution HD bit rates, from 10Mbps to 30Mbps.
Can be observed that for the first generation (1G) a quality of 36dB (PSNR) is obtained encoding the content with a bit rate of 17Mbps, while to achieve the same quality for the third generation (3G), the bit rate should be increased up to 24Mbps, which means to increase a 40% of network bandwidth. That is the reason why broadcasters and network operators should keep in mind the encoder concatenation losses in their contribution and transmission architectures, in order achieve the best content quality.
Is the 4:2:0 format the more efficient than 4:2:2 for contribution services?
It is widespread the idea of 4:2:0 subsampling can be more efficient that 4:2:2 chroma subsampling has been settle, for high bandwidth constraints scenarios, such as HD newsgathering or low quality contributions. That is argued on the premise that de lower 4:2:0 resolution requires lower compressed bit rate, which is not wrong, but it is not also false that the lower 4:2:0 resolution decreases the colour quality.
Colour subsampling could be beneficial if not any more process is applied to the content, but in the real scenario the decoded sequence is converted to 4:2:2 to be transported inside the TV facilities by SDI/HD-SDI interfaces, allowing to be processes through video mixers, DVE generators, logo inserter and any other broadcast devices which works only in the 4:2:2 domain.
In spite of the recently approval of international recommendations as SMPTE RP2050, defining the perfect reconstruction filters for 4:2:0/4:2:2 conversion, the high frequencies losses caused by 4:2:2 to 4:2:0 downscaling, cannot be avoided over the colour components.
With the aim to analyze the 4:2:0 performance in a real multigeneration scenario where the encoder-decoder is interconnected by 4:2:2 interface, we have repeated the three concatenation cycles, using the 4:2:0 profiles of H.264/AVC and HEVC.
Figure 4 depicts the HEVC results for the third generation with the “Main10” profile, showing the luminance performance on the left side and the “V” colour component performance on the right side. It can be observed that for the 4:2:0 luminance component shows a slightly gain compared to 4:2:2, due to the lower resolution of 4:2:0 chroma allow to the rate control allocates the saving bits from colour components to the luminance component.
Nevertheless the quality of the colour components for the 4:2:2 format is 1dB higher than the 4:2:0 format. This evidences the colour quality penalty of 4:2:0 contributions.
Concerning to H.264 encoding using the “High” profile, figure 5 shows the colour performance (U component) for the three generations, reporting a progressive degradation of 4:2:0 format compared with the 4:2:2 chroma subsampling. Even though the colour quality losses for the first generation are near 1.5 dB, the quality goes down dramatically nearly to 3dB for the third generation.
This paper examined the impact of the cascaded encoding-decoding process in professional contribution services in terms of quality losses, for the H.264 and HEVC standards. The results show that 4:2:0 encoding achieves worst quality performance compared with 4:2:2 format, for both standards.
The colour degradation is over 1 dB for the first generation, and it could increase nearly 3 dB for the third generation when H.264 compression is used. Moreover, HEVC 4:2:0 degradation keeps constant to 1dB for any generation.
Thus, network operators and broadcasters have to be aware of these multigeneration issues in the real scenarios in order to reduce the quality losses, and being aware the colour quality degradation that 4:2:0 encoding can cause.
Sapec R+D Departament