Decoding the Domesday discs¶

The BBC Domesday Project (1986) is published on PAL LaserVision discs:

the Community disc, sides North and South;
the National disc, sides A and B.

Community North, Community South and National A are CAV sides. Most of their content is single-frame stills (Ordnance Survey maps, photographs, text pages, pictures of data), with some held pictures and motion sequences. National B is a CLV side.

A single capture of a side can be decoded with a Quick Project. The best result comes from capturing several discs of each side, ideally from more than one pressing, and building a project in which each stage deals with one thing the discs or the players did to the signal. This page explains what each of those stages is for and where it goes. The stage reference pages give every parameter.

The project¶

For one side, with several captures:

Capture 1 ─ Frame Map ─ FM Beat Cancel ─┐
Capture 2 ─ Frame Map ─ FM Beat Cancel ─┼─ Source Align ─ Stacker ─ Dropout Correction ─ Picture Band Limit ─ Channel Equalise ─ Video Sink
Capture 3 ─ Frame Map ─ FM Beat Cancel ─┘                                            │
                                                                                     ├─ Audio Channel Map ─ Audio Sink
                                                                                     └─ EFM Decoder Sink

The stages fall into three groups:

Per capture: the stages before Source Align. They undo what went wrong in that capture or what is particular to its disc's master: player skips, and the FM beat.
Stacking: Source Align and the Stacker. They combine the captures, and the repeats of held pictures within each, into one signal with less noise and fewer dropouts.
After stacking: the stages after the Stacker. They work on the combined signal: dropouts no capture could fill, noise above the picture band, the mastering's edge smear, and the chroma decode.

The order matters. Does the order matter? below says which stages must come before which, why, and which can go either way.

To run the same project over new batches of captures from a script, see Stacking multiple discs.

Per capture¶

Source¶

Use one TBC Source (ld-decode .tbc) or CVBS Source (.cvbs) per capture. A stack takes all its captures in one format. See Source stages.

Frame Map: player skips and repeats¶

A player that skipped or repeated tracks while the disc was captured leaves frames missing or doubled. Every capture then has its own frame numbering, and the Stacker would combine different pictures.

The Frame Map's Disc Mapper tool reads the picture numbers (CAV) or time codes (CLV) from the VBI and rewrites the capture's frame list:

repeats are removed;
gaps are padded with blank frames, which the Stacker ignores.

Run it on every capture. See Frame Map and Disc Mapping.

FM Beat Cancel: the herringbone on the masters¶

The Domesday masters recorded a beat between the FM video carrier and the colour signal. It shows as a faint diagonal herringbone in flat saturated colour: the bars, menus and the seas of the maps.

The beat is on the master, so every capture of that master carries the same beat, and stacking keeps it.
Captures of different masters carry beats of independent phase. Once they are blended, one fit removes much less.

So the stage goes on each capture, before the Stacker. Run its FM Beat Calibration tool on each one first: it measures the master's carrier frequency, which the stage needs to within a few kHz. See FM Beat Cancel.

Stacking¶

Source Align: the same picture at the same frame¶

After mapping, the captures still start at different places on the disc. Source Align offsets each one so that output frame N is the same disc picture in every capture. Its Source Alignment Analysis tool works the offsets out from the picture numbers or time codes. When the set holds captures of two or more pressings, it also measures where the set's picture band ends (lumaCutoffMHz), which Picture Band Limit uses later. See Source Align.

Stacker: noise and dropouts¶

The Stacker combines the captures sample by sample. Noise falls, and a dropout on one capture is filled from the others. Settings worth knowing for Domesday:

mode Weighted and group_pressings (both defaults). Captures of one pressing share the noise recorded into its stamper. The Stacker finds which captures share a pressing and counts each pressing about once, so the pressing with the most captures does not outvote the others.
still_frame_stacking. On the CAV sides a still is often held for several frames. With this on, the Stacker also stacks the repeats within each capture. Each repeat was read from a different place on the disc, so it removes the stamper noise that stacking discs of one pressing cannot. On map stills, repeats took about 6.8 dB of noise out, against about 1.3 dB from stacking five captures without them. The Stacker also records how long each picture is held, which the Video Sink's decoder schedule uses (below).
chroma_gain_match (default). The decode's high-frequency correction can leave one capture's chroma stronger than the others' over stretches of a side. This brings every capture to the same colour burst level before combining.

See Stacker.

After stacking¶

Dropout Correction: what stacking could not fill¶

Where every capture had a dropout, or the captures disagreed, the Stacker leaves the samples flagged. Dropout Correction fills them from the lines around them. Its default method, interpolate, keeps map lines and text whole on the Domesday captures, where copying a single line pastes a streak from four lines away. It goes after the Stacker so it only repairs what the other captures could not. See Dropout Correction.

Picture Band Limit: noise above the picture¶

The Domesday pictures end at about 5.4 MHz, but the decode passes everything up to the sample rate's limit. Above the picture there is only disc noise and the ringing after sharp edges.

Picture Band Limit low-passes the luma there and leaves the colour as it is.
Set cutoff_mhz to the lumaCutoffMHz that Source Alignment Analysis reports for a set of two or more pressings. Otherwise keep the PAL default of 5.4 MHz, where the Domesday captures' picture ends.
It goes after stacking and dropout correction, so it filters the finished signal once.

It changes only the active picture: the VBI lines, and the test signals on them, pass through unchanged. See Picture Band Limit.

Channel Equalise: the edge smear¶

The CAV Domesday sides carry an edge smear from mastering: a faint trail after every sharp edge, about a quarter of the step, decaying over a third of a microsecond. It softens map lines and lettering, and it takes about 2 dB off the colour. National B (CLV) does not have it.

Run the stage's Channel Calibration tool first. It measures the smear on the test signals of field 1 line 19 and sets the stage's two values. The Community sides and National A read about 0.24 and 6.1 samples.
Where it reports no smear, leave the stage out.
The stage removes the smear and brings each frame's colour to the level its burst says, so the colour is the same from frame to frame and from side to side.
It goes after stacking and dropout correction, beside Picture Band Limit, in either order.

See Channel Equalise.

Video Sink: decoding the colour¶

The Video Sink separates luma and chroma and writes the video. The Domesday sides mix two kinds of material, which suit different decoders:

Held pictures and motion, where the picture is the same, or changes slowly, over several frames. transform3d separates best here, because it uses the frames either side.
Single-frame stills, where each frame is a different picture: most maps, photographs and text pages. Neighbouring frames are unrelated, so a 3D decoder would mix them. palstill was made for these: a specialist still-picture decoder that decodes each frame from its own two fields as one picture. It gives sharper colour and much less false colour on maps, text and photographs than the other 2D decoders (town map ΔE 3.5 against 4.8 for transform2d on videosynth's test stills).

palstill is meant only for those single-frame stills. It is expected to do worse on moving video and on held pictures, and the schedule below sends those frames to other decoders. It is also slower: about 2.6 times transform2d's decode time.

How far ahead it is depends on how clean the stack is: - Single capture: it is ahead of transform2d on every kind of still. - Low-noise stack of several captures: on dense maps and on photographs with fine luma detail, transform2d can be slightly more accurate, by about 0.2 ΔE. On the other stills, such as text pages and sharp colour edges, palstill stays ahead.

If a side is stacked from many discs, compare the two on a few dense map frames before choosing.

Decoder Type multiple with a Decoder Schedule follows the material. Fed from a Stacker with still_frame_stacking, this schedule decodes the frames the Stacker found held with transform3d and everything else, single-frame stills included, with palstill:

palstill, held:transform3d

palstill assumes the picture does not move between the two fields of a frame. Motion that the Stacker does not report as held (a sequence where every frame is new) is combed, so give such stretches, and film sequences, a range of their own with transform3d: palstill, held:transform3d, 30001-30450:transform3d.

How much that matters depends on the side:

Film (both fields one instant) is not combed by palstill, but it is still moving footage: decode it with transform3d, which uses the related neighbouring frames.
Slow interlaced motion costs it little: a pan of 2 samples a field, about 9 seconds across the picture, leaves it the best of the 2D decoders.
Faster interlaced motion is where it goes wrong. By 8 samples a field it is level with transform2d on average and worse in its worst places, and by 24 samples a field (3 seconds across the picture) it is clearly the worst.

Community North and National A have almost no interlaced motion, so palstill, held:transform3d suits them as it stands. About a third of National B is video motion, so give its moving stretches to transform3d.

Four more points apply:

The line-alternating error. Alternate lines of a Domesday still differ slightly in hue.
Most of it is a turn of the colour mastered into each still: the same on every pressing, different from one still to the next. palstill measures that turn from the frame's own flat colour and corrects it on every frame by itself. There is nothing to set.
A frame with too little flat colour is left uncorrected.
A leak of V into U from the player is not corrected. It is small on the discs measured, but it shows on held colour bars.
Decoder settings. transform_threshold applies to palstill as it does to the Transform decoders, and simple_pal does not. A frame whose burst is under a quarter of the nominal level decodes as monochrome.
Noise reduction. With noise_reduction_mode still_only, it averages only the frames the Stacker found repeated and leaves single-frame stills as decoded. Two maps of one area look alike to its own test, and the Stacker's view keeps them apart.
Chroma level. Set chroma_gain and chroma_phase only after the decoder is chosen. With Channel Equalise in the project, the chroma level is already set from the burst.

See the Chroma decoder guide and Video Sink.

Audio and data¶

Analogue sound. The analogue sound is mono over much of the discs, as the VBI programme status says. An Audio Channel Map with vbi_mono_average, after the Stacker, averages the two channels on those frames. The two FM carriers' noise is independent, so the result is quieter: 1.0–1.3 dB on Community North's mono passages. See Audio Channel Map.
The EFM channel. Where a side's EFM channel carries data rather than digital audio, the EFM Decoder Sink in data mode writes its ECMA-130 sectors. Leave its doubt_erasure_threshold at 0. The Stacker's efm_stacking is best left at Disabled, which passes through the EFM of the capture with the fewest dropouts. See EFM Decoder Sink.

Does the order matter?¶

Yes, for most of the chain. Each stage either needs something an earlier stage provides, or would spoil what a later stage relies on.

Rule	Why
Frame Map before Source Align	Source Alignment Analysis works out one offset per capture. That is only possible once each capture's skips and repeats are gone: an unmapped capture drifts by a frame at every skip, so no single offset lines it up.
Source Align before the Stacker	The Stacker combines the frames with the same frame number from every input. It does not search for the matching picture, so the inputs must already agree on what each frame number is.
FM Beat Cancel on each capture, before the Stacker	The beat comes from the master, and each master's beat has its own phase. On one capture the stage fits one beat. In a stack of captures from different masters the beats are blended, the sum no longer follows any one master's phase, and one fit removes much less.
Dropout Correction after the Stacker	The Stacker fills a dropout with the real picture from the other captures, and needs the dropout flags to know which samples not to use. Dropout Correction replaces flagged samples with an estimate from the lines around them and removes the flags. Placed first, it would hand the Stacker guesses as if they were picture, and the other captures' real samples would only be averaged with them. The Stacker also flags samples the captures disagree on, for Dropout Correction to repair afterwards.
Picture Band Limit after the Stacker and Dropout Correction	Its low-pass filter reaches many samples either side. Before Dropout Correction it would spread each dropout's spike beyond the flagged region, where the repair no longer reaches. Before the Stacker it would also remove the high band the Stacker measures each capture's noise from. Once on the stack, it filters the signal once instead of once per capture.
The Video Sink's `held:` schedule needs the Stacker's `still_frame_stacking`	The Stacker records how long each picture is held, and the schedule reads that record. The stages between them pass it through. Without a Stacker that stacks stills, `held:` finds no held pictures and every frame takes the default decoder.
Audio Channel Map (`vbi_mono_average`) after the Stacker	It reads each frame's VBI programme status, and averages the sound the stack has already combined and aligned.

Some stages can go either way:

FM Beat Cancel and Frame Map can be in either order on a capture's branch. Frame Map only reorders frames, and FM Beat Cancel works one field at a time. What matters is that FM Beat Cancel stays on the capture's own branch.
Picture Band Limit can sit anywhere between Dropout Correction and the sinks.
Channel Equalise sits after the Stacker and Dropout Correction, for the same reasons as Picture Band Limit: its correction also spreads each sample over the next few. Either order with Picture Band Limit gives the same result. Channel Calibration reads the test signals on line 19 at Channel Equalise's input, and Picture Band Limit leaves the VBI lines unchanged, so the tool measures the same smear in either position. Channel Equalise itself filters the whole frame, VBI lines included.

With a single capture there is no Source Align or Stacker. The chain is Source → Frame Map → FM Beat Cancel → Dropout Correction → Picture Band Limit → Channel Equalise → sinks, and the rules above still apply. Add a Stacker on its own with still_frame_stacking on to stack the repeats of held pictures within the capture.

Checking the result¶

Preview. Use the preview on any stage to see what it changed. Select the stage before and after it, and compare the same frame.
Test Card F. Where a side holds a test card, it is a quick check of the decoder and of the colour level.
A short section first. Before a whole side, render a short section with maps, text and a photograph.