Modulated LED HDR Display Technology Issues

Recently various companies have demonstrated high dynamic range displays that utilize modulated LEDs in order to achieve an improvement in contrast. This paper is intended to highlight deficiencies of the technology in the context of critical image viewing applications such as Broadcast Television, Motion Picture, Computer Graphics and other areas where image accuracy, repeatability and stability are of paramount importance.

eCinema Systems developed and prototyped a similar technology around the middle of 2001. The technology was abandoned when we realized that it had serious flaws when critical image evaluation was the application context. The displays now making the rounds are not very far from the technology independently developed and tested by eCinema (the principles are basic and simple). This paper relates what we learned through our own work as well as what we have confirmed through observation of the various display demonstrations.

Technology Basics

Modulated LED HDR displays combine a low resolution LED array (about 1800 LEDs) with a conventional LCD. By modulating the LED array the display can produce very dark blacks in areas of the picture where this is required.

Conventional LCDs are currently limited to a contrast ratio of about 1,000 to 1. In principle, Modulated LED HDR displays can produce a theoretical (but not practical) infinite contrast ratio due to the fact that LEDs can be turned off in image areas where this is desirable.

Some designs also drive the LED array with data derived from the most significant bits of the incoming image data and apply a compensation algorithm to attempt to control halo effects resulting from the fact that the LED array is of very low resolution when compared to the LCDs native resolution.

In typical applications the LED array consists of one LED per 33 or more native pixels in the picture.

Overview of Key Issues

On first inspection it may seem that the unique approach taken by the Modulated LED design has merit as a technology that can deliver images with contrast ratios that exceed what is attainable with standard LCD displays. While this may very well be the case for non-critical image evaluation applications (such as digital signage), it certainly isn't the case for any application where image accuracy is the key criteria. The technology has problems in the following areas (to be explored in detail as this document is updated):

• Variable shape halo effect significantly distorts high-contrast images
• Difficulty in achieving acceptable levels of flat field uniformity. About a 5% error is what seems to be attainable today. Professional applications can't tolerate a 5% error in shading uniformity.
• Highly complex problem in maintaining calibration of LED array due to the operating characteristics of LEDs
• Extremely limited frequency response (and MTF) due to low resolution LED array
• Halo compensation limited by LCD native contrast ratio
• Temporal tracking problem between the superimposed LED and LCD images
• LED array pitch reduction increases cost and MTBF geometrically and provides small gains for some of the above issues (and loss of performance for some of the others)

The technology is interesting from an engineering standpoint, however, it fails to deliver the key performance milestones required in professional imaging for such fields as Broadcast, Motion Pictures, Computer Graphics, Medical Imaging, Military/Defense and Simulation, among others.

Simulation of HDR display halo effect.

Variable shape halo effect is easy to demonstrate through a simulation.

The panel is limited to 1000 to 1 contrast ratio. Any light source placed behind the panel will, invariably, produce light on the panel's surface that is at most 1000 times dimmer than whatever the incoming light might be. As the LED array is very sparse (about one LED per 35 pixels) high contrast edges are displayed with a very noticeable halo effect around them. And, as the edges or features move, the halo changes shape due to the variable number of LEDs that need to be activated.

An easy example to follow mentally is that of a vertical white line five pixels wide. If this line is aligned exactly with a column of LEDs then those LEDs are activated. You will have a five pixel wide white line with a halo of over 30 pixels on either side. As the line starts to move away from a perfect alignment with the underlying LED array, additional LEDs have to turn on. At one point up to TWO columns of LEDs are necessary in order to light-up the line. Now the line has a halo around it that is twice as wide as it was a few frames earlier.

Rotation and text make it even more interesting because the shape of the halo changes in nearly every frame.

The following image sequence illustrates the process:

This is the desired image. A high contrast frame with a narrow white line on a black "canvas"...

Here's the LED array for this portion of the display...

The following set of LEDs are turned on in the case where the line happens to be in direct alignment with a column of LEDs...

Of course, the LEDs must be diffused in order to provide mixed even illumination....

The resulting image on the face of the LCD has a halo due to the spread of the LED illumination...

If, now, the line moves...

A different set of LEDs must be engaged...

Which produces a different halo shape...

This is what it might look like with motion...

As features move a variable-shape halo of different intensities is created. It become more evident as the image contrast increases and as the viewing environment becomes darker. The display is fundamentally limited to the basic 1,000 to 1 contrast ratio range of the LCD element. In other words, a 1000 candela per square meter white bar would be surrounded by a 1 candela per square meter halo, which easily visible under most non-critical viewing conditions.

The following Flash simulation demonstrates the effect with moving high contrast features. This simulation is for a 1920 x 1080 display with an LED pitch of one LED per 30 image pixels, for a total of 2304 LEDs. This simulation is a good approximation of what one would see in a typical dark viewing environment used in Broadcast, DI or Motion Picture applications.

High resolution file (15MB) here: hdr_halo_high_res.zip

Low resolution file (6MB) here: hdr_halo_low_res.zip

Flat Field Uniformity Issues

Flat field uniformity of "shading" is a measure of how uniformly lit a display might be across its full surface. Typically one would provide the display with a full frame of white, red, green or blue and use a spectroradiometer or similar device to measure luminance at different points along the screen surface. In general terms, for critical image evaluation applications it is desirable to have a shading uniformity that is better than 99%.

LEDs are notorious for drifting in both center wavelength and output intensity with variations in voltage, current and operating temperature. LEDs can be stabilized through various feedback mechanisms. However, it is very difficult to do this well when a large LED array is being driven with a low resolution image. Accurate stabilization would require a luminance sensor per LED and a very complex compensation algorithm (as surrounding LEDs may influence measurements). In practical terms it is impossible to sense each LED individually. Instead, a scheme whereby an area of LEDs are metered can be implemented. This, of course, only works for cases where all LEDs in that zone must operate at the same output level. As image content seldom offers this advantage, the sensing of an area will, without a doubt, introduce errors into the therma/current/voltage compensation calculation for each member of that zone. In other words, the approach simply doesn't work very well.

Modulated LED HDR technology is currently quoted as producing a flat field uniformity of 95%. This is not adequate for critical image evaluation. It means, for example, at an input code value of 1023 (10 bit image system) the resulting on-screen luminance could be equivalent to that generated by code value 1001 elsewhere on the screen. The effect, depending on implementation, would affect other areas of the gray scale to different degrees.

Although this is rather difficult to reproduce on uncalibrated computer monitors where this document will be viewed, the following two images attempt to provide an idea of what a 5% error in luminance would mean.

The number on the RIGHT is the intended image input and resulting gray scale.
The image on the LEFT is what might happen with a 5% error.

The actual luminance produced on screen would be equivalent to that of an entirely different input code value

In short, a flat field uniformity of 95% is simply not adequate for any activity that requires critical image evaluation. Applications such as digital signage are far less critical and can easily tolerate such errors.

more to come...

Click here to return to eCinema Systems main site