Skip to content

Reconstruction and output

Before comparing precision, read the picture priorities: clipping or missing source information is not repaired by choosing a larger output bit depth.

At a glance

Internal precision, buffer layout, HDMI transport and panel precision are different properties. An RGB 8-bit transport flag can describe a carrier for higher-precision Dolby Vision samples.

Reconstruction

  1. Decode BL and EL and pair frames by presentation time.
  2. Resample EL onto the required component grids.
  3. Reshape BL using polynomial or MMR parameters; reconstruct the residual through inverse NLQ.
  4. Combine the contributions with defined shifts, rounding and clipping.
  5. Convert or pack into the output's chroma, bit-depth and metadata representation.
  6. Apply display mapping and panel processing to produce visible light.

NLQ means non-linear quantisation. Its inverse recovers residual values from coded EL values. MMR is multivariate regression involving components and cross-products. Operation order and sample positions affect the result.2

FEL reconstruction uses the composer parameters even when the display can reproduce the source without tone mapping. It reconstructs the encoded picture, not a guaranteed lossless copy of the original studio master. The two-layer arrangement preserves an HDR10-compatible base picture; it is a compatibility choice rather than proof that two layers are inherently better than a single higher-precision encode.2

10-bit dithering

Reducing a 12-bit result to 10-bit reduces the available code values. Dither distributes quantisation error spatially and/or over time, which can make banding less noticeable. It does not retain every original 12-bit value in every output pixel.

For a native 10-bit panel, a higher-precision picture eventually has to be represented using the panel's available levels. Dither can reduce visible contouring, but is not mandatory for every image or a guarantee that banding disappears. Spatial patterns, temporal variation and noise shaping have different trade-offs; panel driving and frame-rate control also matter.

Where the reduction happens matters. Dither added before lossy encoding can be altered by compression or cost bitrate. Dither applied after decoding avoids that encoding step, but the TV may still rescale, convert colours or map brightness afterwards. An uncompressed HDMI link does not make subsequent processing transparent. Keeping higher precision until a later reduction avoids an early loss of code values, without guaranteeing that the TV has the better algorithm.

Repeated quantisation is not automatically invisible. A second reduction can add error, and later filtering or nonlinear mapping can change the first dither pattern. A “maximum doubling of noise” is not a general bound: adding independent, equal-power errors without intervening transformations is a particular mathematical model, not a description of every HDR pipeline.1 There is no established universal visibility threshold for these player/TV combinations; see the proposed viewing tests.

The inspected AetherEngine default path rounds or dithers to 10-bit. An opt-in fullPrecision path places additional information in normally unused P010 bits, but its documentation explicitly says the display path is unvalidated. That is not evidence of 12-bit HDMI output from Apple TV.3

Whether mapping happens before or after a precision reduction can matter. Report the order, not just a bit-depth number. Also avoid “every TV ultimately becomes 10-bit” as a universal claim: panel design, processing precision and temporal dithering vary by model.

A newer Xiaomi S905X5M report describes spatial dithering in Dolby Vision output. This illustrates why a 12-bit carrier alone cannot establish full-precision reconstruction. Its exact internal pipeline and the visibility of the reported pattern remain unverified here.

HDMI tunnelling

CB1 describes 12-bit 4:2:2 component samples packed into an RGB8 carrier. An analyser may therefore report RGB/8-bit without the picture having been reduced to ordinary 8-bit RGB. The bit budget is consistent: two 12-bit 4:2:2 pixels and two RGB8 pixels each use 48 bits. Actual byte packing and metadata still need correct interpretation.4

Capture equipment must preserve carrier bytes. Colour conversion, range remapping or chroma filtering applied to the carrier as if it were conventional RGB can invalidate a comparison.

TV-led and player-led

TV-led: the display performs display mapping. Player-led: the source device performs that mapping. FEL composition is a separate operation. TV-led does not mean the TV receives and decodes the compressed Blu-ray EL.5

Tone mapping needs a suitable description of the target display. The TV can account for its own operating characteristics; a player or external processor needs appropriate target information and configuration. Neither location guarantees the more accurate result. Compare the particular modes on the particular display, including clipping, brightness tracking and any further processing in the TV. A player-led signal still passes through panel processing.

Not every player or display supports both modes. A source-domain capture cannot be compared directly with display-mapped output without accounting for the transformation. See displays without Dolby Vision.

Chroma sampling

4:2:0, 4:2:2 and 4:4:4 describe spatial sampling of colour components, not HDR quality or bit depth. Left/top-left siting, filter coefficients, edge handling and intermediate clipping can change samples before dither. “12-bit output” alone does not qualify the entire pipeline.

Upsampling 4:2:0 estimates missing colour samples; it does not recover all the information discarded by subsampling. Colour conversion and finite-precision calculations can introduce rounding too. A display that needs no brightness or gamut compression still performs these signal and panel conversions.

The code comparison identifies concrete filter and arithmetic differences. The player matrix summarises which paths are documented.

Reviewed 9 November 2026 · Source-reviewed · no local playback measurements

Evidence policy · Sources

  1. David Kress, Analog Devices, Understanding AC Behaviors of High Speed ADCs, 1 January 2019, “Quantization” and “Noise”: noise-model assumptions and addition of uncorrelated contributions. PRIMARY technical explanation; read 9 November 2026. This is general signal theory, not a video visibility test or evidence about a specific TV. ↩

  2. S02, ETSI GS CCM 001 V1.1.1, composer §5.4.2–5.4.3 and informative Annex B. PRIMARY. ↩↩

  3. S03/S05, AetherEngine design, output precision; shader source, dovi_quantize and dovi_full_precision. CODE / DEVELOPER; pinned commit, static inspection only. ↩

  4. S06, CB1 contract, admitted output and reconstruction. DEVELOPER; pinned commit. The RGB8 carrier description is attributed to CB1 here; a general Dolby Vision source for TV-led tunnelling remains TODO. ↩

  5. S14, OSMC Dolby Vision documentation. DEVELOPER; dynamic page. ↩