DTG Digital Receiver Implementation Guidelines and Recommended Receiver Reaction to Aspect Ratio Signalling in Digital Video Broadcasting

A Critique

This important document, used to standardise the handling of programme content in UK digital TV broadcasts where the transmitted aspect ratio differs from that of the user's TV, contains significant errors, which amazingly have gone uncorrected in the four years since it was published.

The author first encountered the document shortly after its publication, and immediately noticed a number of idiosyncrasies and apparent editorial errors. Having spotted them so quickly, I naturally assumed that a corrected revision would soon appear; unfortunately this has not been forthcoming, and worryingly it now appears that some manufacturers of digital set-top boxes are blindly following the DTG guidelines, errors and all. As such, I feel it is important to lobby the DTG to make the necessary corrections before any more harm is done.

It is all the sadder because the DTG document is largely superfluous. I won't go into a full explanation of AFDs (Active Format Descriptors - for the uninitiated, these are metadata describing the shape and makeup of the transmitted signal) on which the DTG document focuses, because there are already good explanations on the web, not least the original DVB specification that defines them. Granted, the DVB spec has the limitation that it only describes the transmitted data, and doesn't attempt to standardise the way that set-top boxes should offer to reformat it for the user for given TV shapes; but all that should have been necessary to specify is an algorithmic interpretation of named user preferences. To spell out each individual case, as the DTG document does, produces unnecessary duplication of information, and leaves the opportunity for errors and inconsistencies to be introduced.

For example, before the DTG guidelines were published, I added AFD support to a set-top box I was working on, based purely on the DVB spec and the application of common sense. In the terminology of the DTG document, there was a "centre cut-out" mode for 4:3 TVs, where the picture was upscaled just enough that no black bars were visible (whether the black bars were in the encoded video or due to an aspect ratio misfit), and a "letterbox" mode for 4:3 TVs and a mode for 16:9 TVs, in both of which the picture was upscaled as much as possible without any of the "shoot-and-protect" region being clipped. The result of these simple rules can be compared against the DTG recommendation in table 1 below.

But first, a quick comment on the notation used. The DTG document (unhelpfully, in my opinion) uses a superfically similar, but significantly different set of symbols from the DVB document: for example the central circle is constrained by the shoot-and-protect region in the DVB document, but by the active video region in the DTG document. I have based my tables on DVB notation, as I think they are clearer than the DTG equivalent:

	The smallest rectangle enclosing the three circles is the shoot-and-protect region
	The region containing active pixels
	Parts of the encoded video signal where all the pixels are black
	The limits, after scaling, of the video visible on a 4:3 TV are denoted by the blue line. Any of the black visible in the table will have been generated in the STB
	The limits, after scaling, of the video visible on a 16:9 TV are denoted by the blue line, but because some of the scaling is often done in the TV, a red line may also be present, indicating the limits of the video in the signal between the STB and the TV. Areas outside the red line but inside inside the blue line are forced to black by the TV, but any other areas depicted in black will have been generated in the STB

For the sake of simplifying the table, I have assumed that the set-top box is capable of arbitrary scaling factors, and that the 16:9 TV fully supports all the scaling modes illustrated in the key above.

Table 1: Table derived algorithmically from DVB specification (compare with table as described by DTG guidelines) Table as described by DTG guidelines (compare with table derived algorithmically from DVB specification)

AFD	Layout when coded frame is		Appearance on a 4:3 TV				Appearance on a 16:9 TV
	4:3	16:9	centre cut-out mode		letterbox mode
0 as the coded frame
						1
1 4:3 (centre)
2 16:9 (centre)			2	2			3
3 14:9 (centre)			4
5 4:3 (with shoot and protect 14:9 centre)				5		5
6 16:9 (with shoot and protect 14:9 centre)			4				3
7 16:9 (with shoot and protect 4:3 centre)						6	3

Note how the position and size of visible black bars and the shoot-and-protect areas are independent of the coded frame aspect ratio, but only in the algorithmic version of the table - this is the first hint that the DTG guidelines are in error. Specific issues raised are as follows:

1. The DTG document actually suggests that there be a further user option to choose between 14:9 and 16:9 letterboxing here. There is no standard naming suggested for this further option, which is strange given the efforts made to standardise everything else. I suggest that this could usefully be combined with an extra user option: the current definition of letterbox mode would be renamed "Standard" mode to reflect the fact that it is the factory default setting, and a new "Letterbox" mode introduced, which upscales as much as possible without cliping the active video region (as opposed to the shoot-and-protect region for "Standard" mode). Then, as a special case, 16:9 AFD 0 video could be displayed in 14:9 letterbox and 16:9 letterbox on 4:3 TVs with user option "Standard" and "Letterbox" respectively. This would also mitigate the concerns of those who wish to continue viewing 16:9 AFD 7 content on 4:3 TVs in the current letterbox manner.
2. The fact that AFD 2 video is letterboxed even when the user selects centre cut-out is an extension over the DVB spec, but I have no problem with this because it makes a useful distinction from 16:9 coded video with AFD 0. However, this needs to be special-cased in any algorithm used to determine scaling.
3. The DTG document mentions in these three cases that there can be an option for the STB to upscale the video to fill all the rasters of an anamorphic signal. This is a curious choice, as it is unlikely in most cases that the STB will be able to do a better job of scaling than the TV - many CRT TVs can readjust their geometry, allowing scaling without artefacts. If the idea is to compensate for TVs that don't have a 16:9 letterbox mode, then why is the option not available for the other AFDs, since lack of a 14:9 letterbox mode (for example) is much more common? (In fact, I would be in favour of such a global "16:9 TV without widescreen signalling" option, if only because of the 14:9 letterbox issue - but more than that, there could also usefully be an option to specify that you have a 4:3 TV with anamorphic mode support, to prevent resolution loss for AFD 2 and other deep letterbox cases.)
4. In these cases, the DTG appears to have forgotten to offer a centre cut-out option. They can't cite loss of resolution due to upscaling, because 117% scaling is already suggested for 4:3 coded frame AFD 6 and 133% scaling is already suggested for 4:3 coded frame AFD 7, amongst other places.
5. The WSS signal for these should be 0111 (4:3 with shoot and protect 14:9 centre) rather than 0001 (4:3 full frame). Compare with the case when the coded frame was 4:3, for further evidence.
6. Last but certainly not least, this case is completely wrong. Neither the 14:9 or 16:9 letterbox options suggested by the DTG document is correct; the shoot-and-protect region is in both cases surrounded on all sides by black bars or by unimportant video. This is exacerbated by broadcasters who use this AFD for programmes that contain a mixture of 16:9 full-frame and 4:3 pillarboxed material: in such cases the 4:3 material is presented with black bars on all sides. Notice that page 23 of the document (16:9 frame, AFD 7) is almost a carbon copy of page 17 (16:9 frame, AFD 0); this suggests an editorial error. Logically, this case should, like the 4:3 frame / AFD 7 case, more closely resemble AFD 6 than AFD 0.

A better proposal

By correcting the errors I have identified, whilst retaining the "hard letterbox" interpretation of AFD 2, reevaluating AFD 0 and adding the third user option for 4:3 TVs, we get the following table.

Table 2: The best of both worlds

AFD	Layout when coded frame is		Appearance on a 4:3 TV						Appearance on a 16:9 TV
	4:3	16:9	centre cut-out mode		standard mode		letterbox mode
0 as the coded frame									note
1 4:3 (centre)
2 16:9 (centre; centre cut-out disabled)
3 14:9 (centre)
5 4:3 (with shoot and protect 14:9 centre)
6 16:9 (with shoot and protect 14:9 centre)
7 16:9 (with shoot and protect 4:3 centre)

My treatment of AFD 0 is worthy of further explanation: in recognition that this AFD (or lack of AFD) is used for material of a variety of shapes, I have applied different rules. I have also illustrated AFD 0 without specifying shoot-and-protect areas, to reinforce the ambiguity.

• 4:3 AFD 0 material is, I believe, best transmitted without a WSS signal, and without any scaling in the STB. This is because the set-top box cannot assume that it can do a better job at scaling the video for a 4:3 TV than the head-end, and to include the WSS signal would inhibit black sensing in 16:9 TVs. The diagram above for 4:3 AFD 0 video displayed on a 16:9 TV (marked "note") deliberately omits the position of the TV tube, because it cannot be predicted in the absence of a WSS signal.
• On the other hand, WSS can usefully be deployed for 16:9 AFD 0 material (signals 0001, 1000, 1101, 1110 respectively) because the STB knows it has inserted black bars, or is producing an anamorphic picture, as appropriate. However, in recognition that this is a "best guess" solution, it is still desirable that the user has easy access to the other modes, and that broadcasters are encouraged to use the correct AFD for their content, rather than AFD 0.

I leave the variations required for 4:3 TVs with widescreen signalling, 16:9 TVs without widescreen signalling, and STBs with restricted scaling factors as an exercise for the reader. Another interesting exercise is to design "centre cut-out" and "pillarbox" modes for 16:9 TVs to complement the existing single mode, which is akin to the "standard" mode for 4:3 TVs; however, this is largely an academic exercise because the same effect can nearly always be achieved using the TV remote control.

References

1. ETSI TS 101 154 Digital Video Broadcasting (DVB); Implementation guidelines for the use of Video and Audio Coding in Broadcasting Applications based on the MPEG-2 Transport Stream
   latest edition V1.5.1 (2004-05); available free of charge from the ETSI website
2. Digital Receiver Implementation Guidelines and Recommended Receiver Reaction to Aspect Ratio Signalling in Digital Video Broadcasting
   DTG Implementation and User Group, latest issue 1.2.1 (February 2001); http://www.dtg.org.uk/publications/books/afd.pdf