Dynamic Backlights for Automotive LCDs
LCDs in vehicles need to demonstrate high luminance and resolution while also saving on power usage – two features that may seem mutually exclusive. One solution is a local-dimming matrix backlight, which can significantly improve the visual quality of automotive LCDs while also conserving power. This approach, combined with emerging micro/miniLED technology, should help make industrial implementation of higher quality automotive LCDs more feasible.
by Chihao Xu, Maxim Schmidt, Torsten Lahr, and Markus Weber
Displays are installed in virtually all late-model automobiles and have become a major differentiation factor for vehicle manufacturers. Tough requirements exist for automotive display applications. These include long lifetimes and wide temperature ranges, as well as daylight readability. High visual performance, including good black levels for nighttime operations, is also very desirable. In addition, automotive display content consists mostly of artificial HMI (human-machine interface) images with high contrast. Their share of bright area is usually low. These requirements have led to liquid-crystal displays being the overwhelmingly dominant display technology used in cars today.
However, the performance of current automotive LCDs, ubiquitous as they are, is substantially lower than ideal. The high luminance required for daylight operation causes high power consumption, which in turn can affect interior design due to the requisite thermal management constraints. Emerging active-matrix OLED (AMOLED) technology offers potentially superior contrast and black levels. But certain aspects of OLED technology, including lifetime, maximum luminance, and mechanical ruggedization, are still limitations to deployment in automotive applications. If these limitations are resolved, then AMOLED technology could represent a new option for automotive displays. In the meantime, improvements in LCDs related to reduced power and higher visual quality are important topics for research and development.
Dimming LED Backlights
It is a straightforward concept to enhance LCD performance by dimming the LED backlight. This means that the LEDs are dimmed while the pixel gray values are increased to maximum transmission. This way, the power consumption of the LED backlight, which represents 80 to 90 percent of the power consumption of the entire LCD module, is reduced. Visual quality can be enhanced as well, as explained below.
The first approach is the introduction of global dimming, meaning that the backlight is no longer constantly driven, but varies according to image content. For global dimming, all LED devices are dimmed by the same magnitude. This approach is easy to realize and can deliver good results when natural images are displayed. For HMI images, however, global dimming generates very little benefit because the maximum gray value is always included in an HMI image. For example, Fig. 1 demonstrates a globally dimmed image with a power saving rate of 40 percent. The peak luminance, a critical value for daylight readability, is reduced in the same manner. In addition, the brilliance of the displayed image deteriorates.
Fig. 1: At left is an undimmed image. At right, after global dimming, degradation of image quality and brilliance is apparent, especially in the areas shown within the red circles.
The next step is the introduction of local dimming. Several LED units, either as a string with several LEDs in series or as a single LED, are separately controlled. This way, an adoption of the backlight distribution to the image is possible. For example, the backlight for the bright parts may be brighter than the one for the dark parts. The black level and the contrast ratio can be improved, since the light leakage is proportional to the dimmed backlight. The image in a dark environment, e.g., during night operation, will appear more vivid. If the black level is sufficiently low, e.g., < 0.01 nits, the display will effectively blend into the rest of the interior without a noticeable background glow.
Current state-of-the-art local dimming algorithms were developed for TV applications. Figure 2 shows a benchmark between an OLED and an LCD TV with local dimming.1 The x-axis is the size of the white window displayed. LCDs can produce a higher luminance than OLED. However, the local dimming algorithm fails to meet the objective for white (1,000 nits) for small windows. This is a demerit for TVs and unacceptable for automotive displays.
Fig. 2: This benchmark chart shows the difference in terms of luminance between an OLED TV and an LCD TV with local dimming.
The sorted sector covering (SSC) local-dimming algorithm was developed to substantially meet the luminance given by the image.2 This is also indicated by the word “covering.” The algorithm has two tasks – one for the determination of the LED values and one for the gray-value adaptation of every pixel. Since an image usually contains millions of pixels, the algorithm needs to be highly efficient, so that a reasonable hardware implementation is feasible. In Fig. 3, the SSC process flow is depicted.
Fig. 3: An SSC-based local dimming process flow is shown from left to right.
The input image data are processed in two parallel paths. In the upper path, the image data are condensed, yielding a gray image of much lower resolution. However, this condensed image still has a much higher number of pixels than LEDs and is the input for the determination of LED pulse-width modulation (PWM) values. This way, the complexity of the local dimming problem is drastically downsized. The foundation of the local dimming process is the SSC optimization core, which considers the crosstalk between LEDs. This is described by means of the light-spread function (LSF) of LEDs. The proper consideration of crosstalk, which is unique for every LCD model, is essential for power saving and black levels. The output is a solution for LED-PWM values, which are used for the LED driver.
The described optimization problem solved by the SSC algorithm is shown in the equation below. The LSF Al (i,j), describes the contribution of the LED l to a certain pixel (i,j) and x(l) is the PWM value belonging to the LED l. The basic constraint is to ensure sufficient backlighting for a certain condensation pixel c(i,j). The cost function of the optimization is the sum of all LED values, which is proportional to the power consumption of the whole backlight unit (BLU). The optimization is based on a greedy algorithm and executed in an iterative approach to properly consider the LED crosstalk. The power consumption as the cost function is minimized.
Since the brightness of each LED is individual, the backlight is no longer uniform, but unique for each image. The transmission of every pixel needs to be adapted, which is executed as shown in the lower path of Fig. 3 and is called pixel compensation. Pixel compensation can be organized in a pipeline that complies with the input of the image data. In this way, the SSC local-dimming algorithm can be realized by a few 100K gates, which can be implemented in a field-programmable gate array (FPGA) or on an application-specific integrated circuit (ASIC) at modest cost. In Fig. 4, an image is displayed without (left) and with (right) local dimming. It is obvious that for the image on the right, the contrast ratio gets much higher and the black level is significantly better.
Fig. 4: A captured image appears without (left) and with (right) local dimming.
The advantages of local dimming are, of course, welcome. However, artifacts may appear. Some are known from TV applications. These include halos, clipping, flickering, etc. For automotive applications, further problems arise due to the HMI content. Human viewers have certain expectations for HMI images with regard to uniform area, smooth luminance/color transition, etc. Deviations due to local dimming may be detected much more easily than in the case of natural images. In addition to perceivable deviations of the images displayed, the brilliance of an image may deteriorate, which is usually a differentiating factor in HMI design.
One way to avoid possible artifacts is to employ a “conservative” type of local dimming. This means that a large margin is reserved, e.g., for the LED values. For dark images/scenes in movies, for which the maximum gray value may be just a small percentage of the full scale, significant improvement will still be achieved. In the case of HMI images, this approach will negate the positive effects of local dimming. Specific methods are needed that suppress the possible visual artifacts and deliver substantial advantages.
The peak luminance of a high-contrast image is a key parameter of the image quality and assures the readability of the image under daylight conditions. Just a fraction of the peak luminance is reproduced by most local-dimming algorithms implemented on TV sets, which is unacceptable for automotive applications. An example of such an insufficient performance is depicted in Fig. 2. This problem is solved by the adaptation of the SSC local-dimming algorithm by setting a specific condensation function. One example is demonstrated later on in Fig. 9.
Figure 5 shows two photographs of an image displayed on an LCD with 16 LEDs at the lower edge. For a better illustration of light leakage, the photographs were overexposed. For the upper photograph, the power-saving rate is 72 percent, but halo is apparent. In order to mitigate the halo, a spatial LED filter was inserted into the SSC optimization core. The lower photograph shows the resulting image, in which the halo artifact is effectively suppressed. Thanks to the dimmed backlight, the black mura is much weaker. The power saving rate is 53 percent.
Fig. 5: The upper image, which is displayed on an LCD using LED edge lighting, shows a halo artifact. The lower image, which is displayed on an LCD with an added spatial LED filter, has a somewhat mitigated halo.
At the present time, most automotive LCDs work with edge-lit backlighting. Local dimming can generate advantages for typical contents displayed in the center console. For the instrument cluster displaying content such as the images in Fig. 9 and Fig. 10, the advantages of local dimming are modest. The light-spread function of an edge-LED rather appears as a stripe, either vertical or horizontal, and does not fit into the circular structure of the content.
Therefore, direct-lit backlighting may be considered as the next step to improve display performance and enhance the value of the display. The LEDs are placed behind the panel in a matrix configuration. Figure 6 shows a photograph of such a backlighting unit (BLU). Each LED device is individually controllable. The number of LED devices may be in the range of a few hundred. A high luminance like 1,500 nits can be realized, so that good daylight readability is achieved. While the local-dimming technology will produce the advantages mentioned above, the disadvantages of a matrix LED BLU are higher costs and a thicker module. In the section titled “Further Development with MiniLEDs” in this article, a technology that may reduce these problems will be described.
Fig. 6: This matrix backlight unit features 29 × 11 LEDs.
A matrix LED BLU is by far the most feasible solution to significantly increase the visual quality of LCDs. The effectiveness of local dimming is not just a matter of algorithm, but also a matter of BLU design. A specific adaptation of the BLU for local dimming may leverage the local-dimming technology.
In order to limit the cost, the number of LEDs should be kept low. On the other hand, a significant improvement for typical images on the instrument cluster has to be achieved. In other words, the aim is to achieve a good trade-off between the performance and the cost for this specific application.
Figure 7 shows the light-spread functions of two adjacent LEDs. The LSF of one LED can be modeled by Gaussian distributions. A characteristic value for the BLU design is the ratio between FWHM (full width half maximum) of LSF and the LED pitch. If the ratio is too low, the uniformity of the backlight will be poor. In the opposite case, the uniformity is high, but the LSF is unnecessarily broad, so the potential of local dimming cannot be fully realized. Figure 8 shows two measured full-on backlight distributions between a few adjacent LEDs for two different ratios. The ratio for the left image is 1:1. The variation of the backlight luminance is 1:2 percent. It leads to visible non-uniformity. The ratio for the right image is 1:5. The backlight variation is 2 percent and is perceived as uniform.
Fig. 7: Above, two adjacent LSFs are shown with the definition of FWHM-to-LED pitch ratio.
Fig. 8: These close-ups show backlight distribution for two BLUs with FWHM-to-LED pitch ratios of 1:1 (left) and 1:5 (right).
According to the authors’ investigations and experiments, a value of 1:5 is reasonable and can produce a sufficiently uniform backlight. For comparison, the FWHM-to-pitch ratio in Fig. 7 is 1:96, which is unnecessarily high. In order to achieve a proper ratio, several measures can be taken. The two key parameters are the number of the LEDs and the thickness of the LC module. Further parameters are the radiation characteristic of the LED and the diffuser film, which may, however, impair the black level in the remote region.
With a higher number of LEDs, the complexity of the local-dimming algorithm gets higher, while the real-time processing requirement remains unchanged. A hardware implementation for a few hundred LEDs may comprise 1 Mbit of SRAM and 400K NAND gate equivalents. The SSC local-dimming algorithm has been amended,3 so that the results are still close to the optimum, while the hardware costs have only moderately increased.
For test purposes, a representative image is displayed and measured. The image is displayed on an LCD with a BLU containing 29 × 11 LEDs in undimmed and dimmed mode. For each mode, the displayed image and the corresponding backlight distribution with full-on LCD are measured by using an image photometer (ELDIM’s U-Master). The results are shown in Fig. 9. The peak luminance of the dimmed image is slightly lower than that of the undimmed image (-18 percent), as the contribution of the strongly dimmed LEDs is now missing. The black level is improved by a factor of five. The power saving rate for this image is 39.7 percent.
Fig. 9: Luminance measurements of an image (left) and the corresponding backlight (right) are compared in undimmed (upper) and dimmed (lower) mode.
Further Development with MiniLEDs
In order to further achieve a better visual quality, particularly for HMIs, and to compete with AMOLED in terms of visual quality, a higher LED resolution for the BLU is required. On the other hand, the costs must not drastically increase. Recent developments in microLED technology offer a new perspective. In contrast to a display with millions of pixels, a BLU with a few thousand LEDs already has a high resolution for this application. The assembly process of a few thousand LEDs on a substrate is much easier to set up than that of millions. In order to achieve a sufficiently high backlight luminance like 10K nits, the area of such an LED device has to be much larger than a microLED pixel. Thus, instead of microLED, the term miniLED has been introduced for this application. The area utilization of a miniLED is significantly higher than that of a microLED. Also, a passive-matrix operation of the matrix miniLED is feasible and reasonable; for example, by applying multiline addressing and multiple scanning. Such an arrangement can realize high luminance for high dynamic range (HDR) and is at the same time thin. The daylight readability, even for novel interior designs or convertible cars, can be guaranteed. The two issues, i.e. the high-power consumption and the limited contrast of the LCD, can be solved by local dimming.
In Fig. 10, a source image and three simulated backlights with 16 × 6, 29 × 11, and 80 × 30 LEDs are shown. It is obvious that with a higher LED resolution, matching between the image and the backlight improves. For a very high resolution like 80 × 30, a very good representation of the image is achieved. This means that the visual quality will be excellent. The LED power-saving rates are -29 percent, -50 percent, and -70 percent, respectively. Also, a very high luminance can be produced without problems such as image sticking and other lifetime issues. The thermal management will be simpler due to the low power consumption, and overheating can effectively be avoided.
Fig. 10: The backlight distribution for a gauge display (upper left) appears after the application of local dimming with increasing numbers of LEDs. At left, 96 LEDs are used. At upper right, 319, and at lower right, 2,400.
During the night, the brilliance of an image will provide a very strong impression to the driver, which is not just a matter of contrast ratio, but depends also on the size of the black area. One measurable quantity is the share of black pixels in the image, as introduced in an upcoming IMID paper by two of the authors and their colleagues.4 For a very high resolution like 80 × 30 LEDs, it can be expected that an OLED-like visual quality will be achieved.
The authors can also predict that a high-resolution matrix BLU such as a miniLED BLU may allow LCDs to achieve near-ideal performance and fulfill automotive requirements under daylight conditions. The cost should be in a reasonable range, and an industrial implementation is realistic. Furthermore, such a BLU will allow new designs such as curved and non-rectangular displays. New HMIs and interior design features will become feasible as a result.
1J. Wang, H. Wang, X. Li, Q. Sun, K. Jia, and S. Zhang, “Key Subject Evaluation Factors of HDR Image Quality Based on LCD and OLED Comparison,” Proceedings of IDW 2017, 1032–1035, 2017.
2M. Albrecht, A. Karrenbauer, and C. Xu, “Sorted Sector Covering Combined with Image Condensation – An Efficient Method for Local Dimming of Direct-Lit and Edge-Lit LCDs,” IEICE Transactions on Electronics 93(11), 1556-1563, 2010.
3M. Schmidt, M. Grüning, D. Schäfer, and C. Xu, “Efficient Modeling of LED Crosstalk of a Matrix Backlight Unit,” Proceedings of IDW 2017, 1457–1460, 2017.
4M. Schmidt, C. Xu, M. Grüning, J. Ritter, and A. Hudak, “Design and Evaluation of High Resolution Matrix Backlight for Excellent Local-Dimming Results and Uniformity of LCDs,” submission to IMID 2018. •
Maxim Schmidt has B.Sc. and M.Sc. degrees in computer and communications technology from Saarland University. He works at the Institute of Microelectronics (LME) as a research assistant. His Ph.D. thesis focuses on dimming algorithms for LCDs.
Torsten Lahr has a Dipl.-Ing. (FH) degree in electrical and electronic engineering from the University of Applied Sciences of Bingen. He is lead technical expert for display signal architecture at Continental Automotive.
Markus Weber has a Dr.rer.nat. degree in applied physics from Technical University of Darmstadt. He is in charge of automotive display technologies with Continental AG, Business Unit Instrumentation and Driver HMI.