Liquid crystal displays (LCD) are currently replacing the previously dominant cathode ray tubes (CRT) in most vision science applications. While the properties of the CRT technology are widely known among vision scientists, the photometric and temporal properties of LCDs are unfamiliar to many practitioners. We provide the essential theory, present measurements to assess the temporal properties of different LCD panel types, and identify the main determinants of the photometric output. Our measurements demonstrate that the specifications of the manufacturers are insufficient for proper display selection and control for most purposes. Furthermore, we show how several novel display technologies developed to improve fast transitions or the appearance of moving objects may be accompanied by side–effects in some areas of vision research. Finally, we unveil a number of surprising technical deficiencies. The use of LCDs may cause problems in several areas in vision science. Aside from the well–known issue of motion blur, the main problems are the lack of reliable and precise onsets and offsets of displayed stimuli, several undesirable and uncontrolled components of the photometric output, and input lags which make LCDs problematic for real–time applications. As a result, LCDs require extensive individual measurements prior to applications in vision science.
Funding: TE and TGT have been supported by Max Planck Society. TE has been supported by National Institutes of Health grant R01 EY018664. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Copyright: © Elze, Tanner. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
In many fields of experimental and clinical vision science where display devices are used, the accurate characterisation of the display output including its temporal properties is crucial for reliable measurements or diagnoses. There are several challenges of display technology for applications in vision research and clinical vision. In ophthalmology, for instance, clinical tests rely on precise presentations of visual objects for diagnostic purposes. In visual psychophysics, a number of experimental paradigms, such as rapid serial visual presentation, visual masking, or priming, require short presentations of visual stimuli with precise onsets, offsets, and precise interstimulus intervals. In certain eye tracking applications, the display needs to be updated rapidly depending on the observers’ current gaze position (gaze–contingency paradigm), which requires an immediate processing of the input signal. In the visual neurosciences, the photometric properties of the display output play an essential role if neuronal responses to visual stimuli are recorded and analyzed, and erroneous assumptions about the stimulus signal may lead to data analysis errors and possibly to incorrect experimental conclusions about the visual system. For some computational models violations of assumptions about the input signal shape may completely invalidate the modelling.
In all these fields, cathode ray tube (CRT) monitors have long been the dominant display devices. There is a large amount of literature about the temporal properties of CRTs [1]–[6], and many practitioners in the fields of vision science are familiar with this technology. While in recent years these CRT devices have been largely replaced by liquid crystal displays (LCD) the photometric and temporal properties of the latter are very little known outside the engineering community.
In this paper we provide extensive measurements and analysis of the temporal properties of LCDs. We identify the main determinants of the LCD output signals and discuss possible effects of the temporal dynamics in vision science applications.
In the first part we give an overview of the LCD technology and summarize recent findings. In the second part we present the results of extensive measurements of LCD signals focussing on two different aspects. First, we illustrate the main determinants of the temporal signals and their variability over different monitor models and different LCD technologies. Second, we unveil deficiencies of the LCD technology which are not mentioned in the manufacturers’ specifications but may be of high relevance for applications in vision research. We demonstrate several cases where incomplete, if not deceptive, manufacturers’ specifications might mislead practitioners in visual psychophysics and neuroscience to misapplications of the respective monitors. In fields of medical research where accurate temporal signals are required, such technical artifacts could render experimental results or medical diagnoses invalid.
This study does not claim that the discussed problems would affect all experiments or monitors but it does point out potential pitfalls that should be taken into account for proper scientific studies with LCD monitors. Ideally, the effect of the temporal properties on the results should be evaluated for every experiment or task. In practice a LCD monitor used for many experiments should be charaterized at least once to check which of the discussed problems may occur. The knowledge of the constraints of LCD technology may also rule out the application of LCD to certain experiments a priori.
The following sections introduce the main determinants of the temporal LCD signal, possible issues and optimizations and enhancement found in some monitors to improve the LCD output.
The duration of a user controlled luminance transition, i.e. a luminance change of a pixel from one frame to the other operated by the graphics adapter signal, is called response time (RT). According to the ISO 9241-305 standard, response times are to be measured between the 10% and the 90% level of the luminance transition.
Response times are commonly considered as the primary characterization of the temporal signal of LCDs, and many previous studies about applications of LCDs in vision science restrict their discussions of dynamic aspects to effects of the LC response [7], [17], [18].
Several issues related to response times are known from previous studies. The first issue concerns the great variability of response times. Response times vary not only over different monitor models but also between different transitions on the same monitor [7], [17], [19]. A decade ago, Suzuki and colleagues [19] compared the response times of four different LC display modes (TN, MVA, TN with DCC [see next section], and IPS) by measuring LC cells. For TN mode they obtained an average RT of 30 ms without DCC and less than 10 ms with DCC, for MVA mode an average RT of 20 ms, and for different IPS modes averages between 20 ms and 40 ms but considerably smaller variances compared to the former three modes. As LCD technology has advanced rapidly since their study, we have performed similar RT measurements with more modern LCD panels.
In contrast to Suzuki et al. we measured monitors instead of isolated LC cells, as the cells are just one determining factor. Additional control electronics, backlight and other components also strongly influence the display quality and performance. We follow their measurements but with four modern LCD panels of different types, study the response time variability and reveal further issues related to response times which have not yet been studied but which may be relevant for applications in visual psychophysics and neuroscience.
The second issue concerns the calculation of response times. It has been previously shown that the method for response time estimation suggested by the ISO 9241-305 standard is subject to substantial errors, and alternative methods have been proposed [7], [8].
Furthermore, response times are usually estimated with monitor settings that may be optimal for signal analysis and minimizing response times, but do not reflect typical working conditions, such as color calibration with reduced brightness. We address this issue and compare LCD signals of monitors with manufacturer default settings to signals after luminance calibration.
In addition to the computer–driven signal transitions, the temporal LCD signal is influenced by the modulations of the backlight. The two most popular backlight technologies are cold cathode fluorescent lamps (CCFL) and light emitting diodes (LED). For both of them, backlight luminance is controlled by pulse width modulation (PWM) which results in a dominant backlight frequency . The darker the backlight, the higher is the amplitude of the modulation at frequency . A high amplitude not only complicates the determination of response times but might also cause lower frequency modulations (beats) if is close to the refresh rate. As will be shown later on in this work, even at maximum backlight luminance many monitors show a considerable amplitude. This may be due to technical limits for overheating protection or ergonomic constraints.
Furthermore, the luminance of the backlight is usually neither temporally stable (especially in the first hour) nor spatially homogeneous over the display unless monitors have special compensation methods built in.
In addition to these signal components which are shared by all LCDs, manufacturers may apply special technologies to optimize the LC response with respect to visual effects.
The most popular such technology is dynamic capacitance compensation (DCC). For rising transitions, DCC briefly applies a voltage which is higher than necessary for reaching the target luminance level, which is called overdrive, whereas for falling transitions, the voltage is turned off for a short period at frame start, which is known as undershoot (see, for instance, [20], chap. 4.9.3).
Some PVA monitors apply an additional pre–tilt voltage to LCs during the frame preceding the luminance transition. This technology, also known as DCC II [21], aims not only to further reduce the transition times but also to avoid black spots on the pixels during the transition which result from the random tilting of LC molecules in the center area by a vertically applied electric field.
Fig. 4 illustrates the different DCC types and their effects on the output signal.
PPT
PowerPoint slide
PNG
larger image
TIFF
original image
Download:
Figure 4.
Schematic of the different types of dynamic capacitance compensation (DCC).https://doi.org/10.1371/journal.pone.0044048.g004
Advanced DCC (A-DCC, see [22]) introduces further response optimizations. It necessitates two independent lookup tables to address the transition preceding the current frame and the transition following the current frame. A-DCC balances rising and falling transitions to achieve symmetric response times and introduces special optimizations to pure black–white transitions and to dynamical transitions like moving lines.
We will show later on in this work that improper DCC might introduce severe visible artifacts. Note that DCC is accompanied by an inevitable input lag (see Discussion).
The voltage applied to each subpixel controls the transparency and therefore the luminance (see Fig. 1). Although the polarity of this voltage does not matter, if only a positive or a negative voltage is applied, the crystal may be decomposed and thus be permanently damaged. Therefore, the polarization of every single subpixel switches in alternating frames. As a result, each single subpixel of a liquid crystal panel oscillates with half the frequency of the refresh rate [11], [23].
Such a modulation with half the refresh rate could be lower than the critical flicker frequency for humans [24]. The reason why it is not perceived nevertheless is that LCD panels invert the polarity of their single dots in a spatially anti–phasic manner for neighboring pixels so that the oscillations cancel out.
There are several possible patterns for neighboring LCD dots to oscillate in phase or anti–phasically. These patterns are called inversion schemes.
When natural images or standard desktop elements are displayed on a monitor the occurence of such a pixel pattern is quite unlikely. However, in applications with artificial stimuli the display image may exactly match the inversion scheme. If a displayed pattern happens to switch off all antiphasic dots, clearly noticeable and undesirable low frequency flicker (half the refresh rate, therefore in most cases 30 Hz) would be perceived. For row inversion, a technique frequently used in notebook LCDs where neighboring rows of points are inverted, even a simple horizontal line is sufficient to elicit this effect. Due to problems in the manufacturing process, voltages may not completely cancel out anti–phasically which would also result in perceivable flicker [11].
Motion blur [14]–[16] is a well–known and unavoidable side effect of sample–and–hold displays, including LCDs. On such displays the stimulus can only be updated framewise and stays visible (at least) up to the end of each frame. As a result, an observer tracking a moving stimulus on a standard LCD will perceive a streaking and smearing of the edges of the visual object. The amount of motion blur is determined by the frame rate and the hold time [25] of the display.
Motion blur has been investigated thouroughly because it impairs the perceived quality of dynamical presentations such as video sequences not only in vision science but also in the consumer market in general. Therefore, we refer to the large body of existing literature considering modeling [26] as well as characteristics and assessment [14]–[16], [25], [27], [28] of motion blur.
Several technologies have been developed to reduce motion blur on LCDs (see [29], Chapter 6.5.2). One method is to switch the whole backlight on and off during each frame (blinking/flashing backlight). However, this method ignores the line updating time difference between top and bottom of the monitor. The scanning backlight technology overcomes this problem by vertically separating the monitor into discrete areas and flashing the backlight of each area from top to bottom according to the respective time of the display update. A third method is the insertion of black data after the beginning of each frame, which, however, requires very fast response times.
Among the monitors measured in our study, only the NEC 24WMGX monitor addressed the motion blur issue by its optional motion picture mode (see above).
We determined the strength of the perceived motion blur by applying a motion blur model to our transition signal measurements.
It is widely known that for LCD panels the light emission is more or less viewing angle dependent [6]. These issues differ between the different LCD technologies. As in most vision science experiments observers look perpendicular at the monitor, we do not cover viewing angle dependencies in this work.
Another aspect affecting the temporal signals, which is not covered by this work, are high-contrast mechanismns present in some LCDs such as local dimming. Local dimming refers to a family of technologies of LED backlight monitors in which parts of the display area can be dimmed or turned off in order to produce very deep black levels. However, as each backlight LED usually covers an area of substantially more than one pixel, local dimming has been reported to impair small bright objects on larger dark backgrounds.
We generally discourage the application of such contrast enhancement technologies for vision science experiments as long as it is not fully clear what impact they have in a certain experimental condition.
The focus of this study is a description of features and artifacts of the LCD technology which are supposed to be relevant for psychophysical and neuroscientific experiments in general. A wide range of different monitor technologies and determinants of the temporal signal are compared. Three recent studies [30]–[32] approach the topic from the opposite side by focussing on well defined psychophysical requirements which they relate to only a few aspects on one or two LCD panels. In the following, we will briefly review these works and compare their approaches and results to the present study.
Kihara and colleagues [30] compare the performance in three psychophysical experiments which were performed on one LCD and two CRT devices, respectively. They statistically analyze the experimental results, fail to find significant differences for most of the conditions, and conclude that the three displays elicited similar performance profiles.
While experimental comparisons of different display technologies clearly may have merit, we have two objections with their appraoch. First, the authors apply null hypothesis significance testing (NHST) and start with the null hypothesis of equality of performance on the three display devices. In the NHST approach, the null hypothesis can only be rejected but never be proven [33], [34]. Therefore, being unable to reject the null hypothesis and to conclude from this that there are no performance differences over the three monitors is a logical fallacy.
Second, even if the authors could have shown an equality of performance over the different displays, the generalizability of their results to other experimental paradigms remains unclear. The practical implications of their study are therefore limited.
Wang and Nikolić [31] compared one CRT monitor and two different LCD panels, an old and a new model, with respect to both their spatial and temporal properties. The authors report that for the new LCD monitor the level of accuracy of timing and intensity was comparable, if not better to the benchmark CRT monitor, while the old LCD panel had a number of issues with respect to accuracy.
While their conclusions are generally in agreement with our study, we would like to discuss a few methodological differences. First, as a minor issue, although the authors measured a considerable Hz ripple for the old LCD device, this finding is not interpreted as backlight pulse width modulation and hence not discussed in the context of the LCD technology. The reader might attribute this ripple to a deficiency of that specific old monitor and erroneously conclude that it is not present anymore in newer LCD panels. Instead, we show that backlight pulse width modulation is a prominent topic for many LCD devices, independent of age and LCD technology, and propose to disentangle this optometric signal component of the luminance transition in order to appropriately characterize the temporal behavior.
Second, the authors propose an idiosyncratic definition of stimulus duration which is used to measure the temporal precision. The established model to specify onset and offset effects, liquid crystal response time, which is proposed by the ISO display metrology standard, is not even mentioned, which makes it difficult to compare their results with existing studies. While there may be good reasons for novel definitions of stimulus durations, their study would have clearly benefited from a comparison with standard approaches.
Third, the authors measure these temporal components only for black white transitions, although these transitions have frequently shown to be fastest over all luminance levels (a result which we generally approve in the present work). Their reports of stimulus duration times should therefore be considered as a lower bound over all possible luminance transitions. Wang and Nikolić indirectly demostrate this variability over different transitions by showing effects of the luminance in the preceding frame on the luminance of the successive frame. However, they measure these effects by randomly permuting all 256 shades of gray (in our notation 0 rgb to 255 rgb) in a sequence of frames and repeat that procedure 100 times. This way, they randomly draw 100 times 256 specific transtions from the total of 65,280 possible transitions in each block. In their quite general, graphical analysis of the data they do not consider the single transitions separately for rising or falling transitions or depending on the distance between lower and upper level. An additional systematic presentation of response times between those levels suggested by the ISO and the VESA measurement standards would have been useful in order to compare the results with existing studies.
The study by Lagroix and colleagues [32] also analyses temporal properties. The authors investigate psychophysical estimates of visible persistence of stimuli immediately after their assumed disappearance on the display device. In their experiments, observers performed forced choice tasks on these stimuli, where a shutter controlled that the stimulus could not be seen during the period when it was (intendedly) displayed. They compared performance using a CRT and an LCD monitor. While there was considerable visible persistence on the CRT for white stimuli on black background, the authors did not find any perceptual persistence on the LCD panel.
The authors measured response times between three distinct luminance levels (10 65, 25 165, and 0 255, respectively), applying a method following the recommendations of the ISO display metrology standard. Due to proper DCC, all transitions occured in less than 5 ms on their LCD monitor. The authors conclude that LCD monitors using the DCC technology are superior with respect to visual persistence effects compared to CRT monitors. Their work makes an important contribution by showing that small response times due to proper DCC correlate with the lack of visual persistence and therefore eliminate a potentially serious artifact of CRT monitors in vision science experiments.
Our study, however, demonstrates a number of artifacts due to improper DCC with some substantial effects on the luminance transition signal, such as luminance stepping or substantial overshoots. It remains important future work to study these artifacts with experimental paradigms as developed by Lagroix and colleagues, as it is likely that some of the artifacts presented in this work have considerable impacts on visual persistence.
For more information Graphic LCD Manufacturer, please get in touch with us!