For the foreseeable future, anyone creating an HDR production will also be required to create a standard dynamic range one alongside it. For single-camera drama, a colorist might create either of those first, and use it as a basis for the other. In either case, the color grade will generally be revised - if only slightly - for the other output. That alone is sometimes an unwelcome expense, especially given the need to satisfy several different distribution systems for both cinema and television.

People creating material for a live broadcast face related issues. There is a huge variety of distribution platforms both in broadcast and online, some of which might require changes as fundamental as reframing to suit picture shapes which have never existed as broadcast formats. It’s perhaps no surprise to find that this area has been a particularly early adopter of machine learning and AI.

Either way, the broadcast sector is no stranger to a need to finish the same production in more than one way. The main concern of production personnel is how HDR and SDR can be created side by side with minimum additional effort. HDR can affect not only vision engineers but plausibly everyone from camera operators, lighting directors and production designers to directors, at least in small part. HDR is not solely a technical enterprise; like any form of photography, it involves some opinion and interpretation.

In some ways, HDR production could be said to make the vision engineer’s job slightly easier, since the HDR display theoretically requires less stringent control over the contrast in an image.  At the same time, viewer comfort often means that the extremes of HDR contrast need to be used carefully, and the need to create a parallel SDR broadcast may impose further requirements. What’s crucial, though, remains the engineer’s understanding that the subject of - say - an open air event are the sunlit participants, not the shadowed background, and the image must be engineered with that in mind.

Humans (and perhaps future AI) know this instinctively. Digital electronics - less so.

Common Standards

The issue of simultaneous HDR-SDR broadcasting was anticipated early, when the BBC and NHK began collaborating on the technology which would become hybrid log-gamma, HLG, in 2015. It was demonstrated to SMPTE later that year and quickly adopted by key companies in acquisition, distribution and home electronics. As the name suggests, HLG combines two techniques, with the intention that the resulting signal works well for both HDR and SDR displays.

Gamma encoding is a familiar term to people involved in digital imaging. The term arises from the use of the Greek letter gamma, γ, to represent a value used to relate signal level to light level. The output of a standard dynamic range camera operating according to the ITU’s Recommendation BT.709, for instance, should properly output a signal which is proportional to the light level raised to gamma, where gamma is equal to 0.45. The encoding is therefore exponential, or roughly exponential, given some adjustments made to the standard to better accommodate the characteristics of real-world equipment.

This approach has worked well for decades, but it cannot represent the large contrast range of HDR images in a reasonable range of digital values. As such, HLG uses an exponential encoding only for the darker areas of the image, with a logarithmic representation for the brighter areas. The result is a signal capable of representing, in photographic terms, perhaps 16 stops of dynamic range. 16 stops is more than most cameras, but the real benefit of HLG is that it looks good on both new HDR-capable and traditional SDR displays.

It is a very convenient approach, and HLG has seen a lot of use in live broadcast for that reason. The problem with HLG, if there is one, is that it does not make the best possible use of the signal. Some digital values represent brightness levels which are so close together that there is no visible difference. This compromises the range of values it can represent - stretch the range any further and quantization noise (banding) might become visible, though practical situations would not normally reveal it.

The alternative is PQ, the perceptual quantizer. It uses more careful mathematics to ensure that the apparent difference in brightness between two adjacent digital values is the same. This makes better use of data space and means that PQ can represent up to 28 stops of brightness in the same data space.

Crucially, though, PQ (and the formats based on it, including HDR10 and Dolby Vision) are not backward-compatible with standard dynamic range displays. That requires broadcasters to produce two entirely separate output images. Commercial interests have promoted PQ (or PQ-based formats such as HDR10 and Dolby Vision) over HLG under some circumstances and this has led to its greater uptake in certain situations; in general, HLG is more common in live broadcast and PQ more common for single-camera content, though the opposite is quite possible.

Standards & Subjectivism

A complete analysis of how the standards work is beyond the scope of this discussion, but there are some differences in how they are written which complicate conversions between them.

HLG is scene referred, meaning that the signal level is related to the amount of light coming out of the scene. The contrast in the scene is likely more than any practical monitor can present to the viewer and the monitor is responsible for producing a reasonable image with reference to the amount of contrast it can achieve.

PQ is display referred, with the signal level representing the amount of light coming out of a theoretical display. The theoretical display used to define PQ has more contrast than any real display and all real displays will down convert PQ signals according to own capabilities.

Technologically, conversions from PQ or HLG to SDR pictures are trivial. The problem, though, is that a subjective interpretation - an opinion - is involved. That opinion must rely on an understanding of what the image contains, as in the example of a sunlit sports match in front of deeply shadowed stadium seating.

Where To Begin

Common solutions are sometimes counter-intuitive. Some workflows are designed to have a vision engineer work mostly on the standard dynamic range production, adjusting contrast, exposure and color to suit a standard dynamic range broadcast. This may seem strange: invariably, where we want two different outputs from a process, we create the higher-capability output, then derive the lower-capability output from it. For instance, we would not usually create an HD signal then try to upscale it to UHD, high quality AI up scalers notwithstanding.

Producing HDR pictures based on SDR vision engineering, though, is not quite the same thing. Both the SDR and HDR outputs are ultimately converted from camera image data which may not innately be either HDR or SDR. Those conversions inevitably involve opinion but this approach is predicated on the idea that image data produced by a camera engineered to produce an SDR result is likely to be consistent enough, given the restrictions of SDR, that the same image data also converts straightforwardly to an HDR result. Most of the opinion is already embedded in the decisions made during the creation of the SDR version.

The vision engineer is still likely to make reference to HDR monitors and test and measurement devices in order to ensure this approach is working. The specifics of the conversion will vary from manufacturer to manufacturer, broadcaster to broadcaster and even job to job; often, lookup tables are involved.

The alternative - and often more intuitive - approach is to engineer an HDR image and derive the SDR image from it. Often (not always) this might involve an actual down conversion of the HDR output to SDR. At the time of writing, the details of doing that, and the devices involved, were somewhat situation-specific, but the world’s best-known manufacturers offer various approaches to doing so either in an equipment rack or perhaps even in the vision mixer itself.

Automating The Opinion

The technical and creative challenges are the same regardless. Converting HDR to SDR is strictly speaking a down conversion that does not need to generate new information. The problem is that the process is somewhat similar to the idea of implementing an automatic exposure function, and broadcast television would not usually use automatic exposure.

Strictly, the difference between SDR and HDR is about contrast, not solely brightness, but the issues are broadly comparable. The early solution of doubling up staff to allow two separate human beings to look at two separate outputs has budgetary implications that the industry was never willing to sustain. As such, HDR to SDR conversion devices often adapt their behavior based on picture content. It is a field which currently attracts a lot of innovation around careful algorithm design - and that innovation seems likely to continue into the future.