The engineering standards of the digital video signal

What a digital video signal actually is

A digital video signal is a sequence of sampled images, each described as a rectangular array of picture elements, or pixels. Every pixel carries one sample of brightness and two samples of colour difference. At the moment of capture, a sensor converts incident light into electrical charge; that charge is then read, scaled and quantised. The result is a grid of integers whose values rise and fall across the frame. Nothing about that grid is intrinsically a picture: it becomes a picture only because a display reverses the process, turning numbers back into light at the correct position and time.

The study of the digital video signal therefore sits at a meeting point of three disciplines. The first is colourimetry, which asks how a physical spectrum of light is mapped onto three numbers and how those numbers are exchanged between capture, storage and display. The second is sampling theory, which asks how finely a continuous scene must be measured in space and in time before the measurement loses information. The third is information theory, which asks how few bits are needed to describe the resulting grid without the viewer noticing that anything was discarded.

Compression standards are the formal answer to the third question. Groups working under ISO/IEC MPEG and the ITU-T study groups published a long family of specifications, from MPEG-1 and MPEG-2 through H.264 AVC, H.265 HEVC and the newer coding schemes such as AOMedia AV1. Each of these standards defines a syntax for a bitstream, a set of decoding operations, and a model of how a decoder should reconstruct pictures. The standards deliberately leave the encoding side open, so that any manufacturer may build an encoder that produces a conforming bitstream, and any decoder that follows the specification will reproduce an acceptable picture.

Reading a bitstream as a layered description

A conforming bitstream is usually described in layers. At the top, a sequence layer groups the pictures that share the same basic parameters, such as picture size and colour arrangement. Below it, a group-of-pictures layer marks the boundaries of independently decodable runs. Below that, each picture is split into a slice structure, and each slice is divided into units of block-level data. The lowest layers hold the transformed and quantised coefficients that describe small patches of the image, together with motion vectors that describe where those patches came from in an earlier reference picture.

This layered arrangement matters because it lets a decoder start work at fixed entry points. A broadcast receiver may join a stream in the middle, wait for the next independently coded picture, and begin decoding from there. A storage system may skip to an arbitrary point in a long recording and begin playback without reading everything that came before. The price of this convenience is that the independent pictures are large, while the pictures that depend on them are small. The whole art of bitrate control is the art of balancing those two forces.

The standards at a glance

The table below places four widely documented coding standards side by side. It compares the block structure each one uses, the colour precision it can carry, and the broad efficiency that published studies attribute to each family. The figures are representative values drawn from the public specifications and from peer-reviewed measurements, not fixed limits.

Colour, sampling and the limits of the eye

Human vision is far more sensitive to fine brightness detail than to fine colour detail. A reference that ignores this fact wastes bandwidth. The standard remedy is to sample brightness, called luma, at full resolution while sampling the two colour difference channels, called chroma, at a lower resolution. The common arrangements are written 4:4:4, 4:2:2 and 4:2:0, and they differ only in how many chroma samples accompany each group of four luma samples. A later sheet of this reference draws each grid cell by cell so that the ratio becomes visible rather than merely stated.

Sampling theory sets a second limit. A signal that varies faster than half the sampling frequency cannot be represented faithfully; instead it folds back and appears as a slower pattern that was never present. This is the aliasing effect, and every capture stage must guard against it with an optical or electrical low-pass stage. The same principle governs motion: a temporal sample rate that is too low makes a rotating wheel appear to turn backwards, and the remedy is again a limit on the detail the signal is allowed to carry.

Why the mathematics of bitrate matters

Bits are the currency of a video system, and every design decision spends them. A practitioner who wants to choose a picture size, a frame rate and a quality target needs a way to estimate the resulting data volume before committing to it. That estimate can be built from a few honest quantities: the number of pixels per picture, the number of pictures per second, the number of bits spent on each pixel, and the factor by which chroma subsampling reduces the raw sample count. Multiplying those quantities gives a raw figure that is then reduced by the efficiency of the compression standard in use.

Volume = pixels × pictures per second × bits per pixel ÷ compression factor.

That single line, unpacked carefully, is the substance of Sheet 04. It explains why a 4K frame carries four times the pixels of a 1080p frame, why doubling the frame rate doubles the raw data, and why a modern standard can afford to spend more bits per pixel on a difficult scene without exceeding a fixed channel.

An open excerpt from the common questions

Is compression the same as damage?

Compression is a controlled approximation. A lossless scheme preserves every sample exactly; a lossy scheme discards detail that the eye is unlikely to miss and records what remains. The discarded detail is real, but it is chosen so that it lies, as far as possible, below the threshold of perception. The skill of a well-designed encoder is to place the approximation where it will not be seen.

Does a higher bitrate always mean a better picture?

No. Bitrate is a budget, not a guarantee. A stream that spends its bits on the wrong regions, or that uses a poor motion model, can look worse than a smaller stream that spends its bits wisely. Quality is a property of the whole encoding decision, not of the size of the file alone.

The remaining questions, together with the full glossary and the editorial method, are collected in Sheet 05 of this reference.