An LCD tachistoscope with submillisecond precision
Tachistoscopes allow brief visual stimulation delivery, which is crucial for experiments in which subliminal presentation is required. Up to now, tachistoscopes have had shortcomings with respect to timing accuracy, reliability, and flexibility of use. Here, we present a new and inexpensive two-channel tachistoscope that allows for exposure durations in the submillisecond range with an extremely high timing accuracy. The tachistoscope consists of two standard liquid-crystal display (LCD) monitors of the light-emitting diode (LED) backlight type, a semipermeable mirror, a mounting rack, and an experimental personal computer (PC). The monitors have been modified to provide external access to the LED backlights, which are controlled by the PC via the standard parallel port. Photodiode measurements confirmed reliable operation of the tachistoscope and revealed switching times of 3 μs. Our method may also be of great advantage in single-monitor setups, in which it allows for manipulating the stimulus timing with submillisecond precision in many experimental situations. Where this is not applicable, the monitor can be operated in standard mode by disabling the external backlight control instantaneously.
KeywordsVisual perceptionTachistoscopeSubliminal stimulationLCDLED backlightVisual stimulation device
Traditional mechanical and electrical tachistoscopes are cumbersome to use and have limitations primarily due to their physical components, particularly when brief exposure durations are used (see, e.g., Lancaster & Lomas, 1977; Mollon & Polden, 1978). Electrical tachistoscopes use fluorescent lamps working as shutter components to passively illuminate images consisting of handmade photo prints presented via manual slide-projection systems (e.g., Karlin, 1955; Kupperian & Golin, 1951). These lamps exhibit rather slow and variable response times (up to 20 ms), creating noise and imprecise timing (e.g., Bohlander, 1979; Glaser, 1988; Merikle, 1980; Mollon & Polden, 1978). Tachistoscopes using mechanical shutters (e.g., Deutsch, 1960; Lancaster, Sayer, Scott, & Sutcliffe, 1971) are often failure-prone, with variable asymmetries in rise and fall times of the order of several milliseconds (e.g., Madigan & Johnson, 1991; Naish, 1979).
Presenting complex visual stimuli at ultrashort durations—that is, within the submillisecond range—as opposed to simple flashes of light (e.g., Efron, 1964) in a controlled and efficient manner has not been achieved yet with standard screens and projectors (Bukhari & Kurylo, 2008; Krantz, 2000; Wiens et al., 2004; Wiens & Öhman, 2005). Recently, tachistoscopes have been described that are based on data projectors and liquid-crystal (LC) shutters (Fischmeister et al., 2010) or on LC panels with light-emitting diode (LED) arrays (Thurgood, Patterson, Simpson, & Whitfield, 2010). Although these setups provide millisecond resolution, they rely heavily on expensive or custom-made components.
Here, we present a simple and inexpensive alternative for displaying images with a high presentation quality and timing, at exposure durations starting in the submillisecond range. Our tachistoscope is based on two liquid-crystal display (LCD) monitors of the LED backlight type, which are set around a semipermeable mirror. Both monitors are aligned so as to appear at the same position when viewed through the mirror. As compared to a slide projector tachistoscope with mechanical shutters, the LC panel plays the role of the slide, and the LED backlight the role of the shutter and the projector. While the backlight is off (shutter closed), the image can be changed on the LC panel without being visible to the observer. Once the image has been transmitted to the LC panel and the LC pixel cells have settled to a steady state, the backlight can be switched on (shutter open) to instantly make the image visible.
Method and materials
We first explain the working principle of an LCD, as this information is closely related to how our tachistoscope operates. An LCD consists of a white backlight that illuminates, from behind, an array of LC pixel cells. These pixel cells serve as light valves whose transmission factors can be programmed individually. A colored filter is present in front of each pixel cell, only letting red, green, or blue light pass. A single image pixel is formed by three such pixels (red, green, and blue). The LC pixel cells and the colored filters are mechanically sealed and are here referred to as LC panel. The LC panel works independently of the backlight.
The backlight, which does not have a pixel structure, illuminates all of the pixels of the LC panel at once, even when a black screen is presented.1 The brightness of the backlight can be controlled by the user through the monitor settings (usually the brightness control). The monitor electronics actually vary the brightness by digitally switching the backlight on and off repeatedly at a frequency of around 200 Hz. The on–off ratio determines the average brightness. This principle is called pulse width modulation (PWM) and is exactly what is used in our tachistoscope application, apart from that the frequency and on–off ratio take rather extreme values. This PWM principle is not only used with LED backlights, but also with cold cathode fluorescence lamp (CCFL) backlights. However, the switching characteristics of LEDs are what make LED backlights the technology of choice for the tachistoscope application.
In a standard LCD monitor, the pixel values in the LC panel are the only parameters that change over time and under computer control. This is a rather slow process, for two reasons. First, if a pixel cell has received a new value, it needs some time to settle to this new value. This time is characterized by the so-called reaction time. Second, the pixel array cannot be programmed at once, but is updated pixel-wise, from the top left of the screen to the bottom right. Once a pixel has received a new value, it autonomously keeps its value until the next update or refresh cycle. This update process is usually visible, at least in a physical sense, because the LC panel is illuminated by the backlight at all times. In our tachistoscope application this is different, as the backlight can be switched off during the update process and be switched on again at any time after all of the pixel cells have sufficiently settled to their new values.2
As compared to a traditional tachistoscope based on slide projectors, the LC panel plays the role of the slide, whereas the LED backlight plays the role of either a switchable light bulb or a mechanical shutter. As the slow changing of the slide in a traditional tachistoscope is hidden by the faster switching shutter, the relatively slow changing of the pixel states in the LC panel is hidden by the fast-switching LED backlight. Switching the LED backlight can be precisely timed and takes just a few microseconds, whereas switching traditional mechanical shutters or LC shutters takes a few milliseconds. Moreover, the switching characteristics are almost identical for both switching directions (i.e., on–off and off–on), thereby minimizing the period during which there is uncertainty about which of the images is being displayed. Having two monitors (i.e., two channels) allows for presenting a stimulus over or embedded in an arbitrary background.
LED backlight control
Here is an example of typical program timing. After the value of a particular pixel P has been altered by the program via the graphics command, it takes one refresh cycle, at maximum, until the new pixel value is processed by the output stage of the graphics card and transmitted to the monitor (16.7 ms, assuming a monitor refresh frequency of 60 Hz). If the monitor does not perform any time-consuming image data processing, the corresponding LC pixel P will be updated at the same time that the value for pixel P arrives at the monitor. Assuming that the LC pixel maximally needs 50 ms to settle to the new value within an acceptable error range, the program must not switch on the LED backlight earlier than 16.7 + 50 ms ≈ 67 ms after having issued the last graphics command. Otherwise, the luminance of the presented pixels might be inaccurate.
In this section, we present measurements acquired with a photodiode (Thorlabs PDA36A-EC), which was connected to a PC oscilloscope (Pico Technology PicoScope4224). When it was needed, the control signal for the LED backlight was connected to the oscilloscope as well. The amplifier of the photodiode was set to a gain of 30 dB, which provided a high enough bandwidth (785 kHz) to accurately capture the steep slopes of the luminance signal. The photodiode was equipped with a lens (Pentax 50mm-F1.4) and a lens spacer, which allowed for measuring a field of approximately 4 × 4 cm at a distance of 70 cm. The luminance was calibrated using a luminance meter (Minolta LS100).
LCD reaction time
As can be seen in Fig. 5, several refresh cycles were needed for the pixel values to settle, especially for changes from low to high values. Only in the very special case of switching to the maximum luminance was the reaction time as short as is found in the monitor’s specifications (≈ 5 ms). Otherwise—that is, as soon as the contrast setting at the monitor was decreased or a gray-to-gray switch took place—the reaction time increased dramatically; this is the usual case, as the luminance will normally be adjusted by the user during the monitor calibration procedure, either by lowering the contrast setting or by limiting the range of pixel values used in the program.
The reaction times for colored stimuli are the same as for white. This is because white, like any other color that can be presented on the LCD screen, is made of a combination of the three primary colors (i.e., red, green, and blue) created by the colored filters in front of the switching LC pixel cells. Therefore, if white has settled, so have the primary colors.
LED switching characteristics
Means and standard deviations of the backlight switching times (in microseconds) as measured from ten luminance curves for each monitor
Backlight Switched On
Backlight Switched Off
Theoretically, such low timing values suggest stimulus durations potentially as short as 20 μs and stimulus frequencies of 25 kHz (assuming an on–off ratio of 1). However, although we think that such extreme values can be safely used, we did not test whether this would affect the LEDs and the LED driver. Thus, strictly speaking, only frequencies smaller than the standard PWM frequency, which is around 200 Hz, can be considered safe.
We measured a longer delay when switching off the backlight as compared to switching it on. This means that stimulus durations were longer than programmed and that a new image (on one monitor) would become visible more quickly than the old image (on the other monitor) would become invisible. However, the difference in the delays was just 10 μs (Table 1), which, if not tolerable or accounted for otherwise, can easily be compensated for by delaying the LED control signal when the backlight is switched on, but not when it is switched off. To this end, the LED control signal was fed through a small and passive analog compensation circuit (Fig. 7). This circuit decreased the slope just of the falling edge of the signal, and thus increased the time needed for the signal to reach the threshold of digital input at the monitor.
Means and standard deviations of the backlight switching times (in microseconds) as measured from ten luminance curves for each monitor, with the compensation circuit in place
Backlight Switched On
Backlight Switched Off
In order to evaluate the predictability and stability of the stimulus luminance under more realistic experimental conditions, we measured the relative error in luminous stimulus energy for 20 different stimulus durations between 200 μs and 20 ms. The interstimulus intervals were chosen randomly on a logarithmic time scale, between 500 ms and 5 s. For each stimulus presentation, we recorded the LED control signal and the luminance signal, computed the integral of the luminance signal, divided this integral by the expected stimulus energy, and subtracted 1, providing us with the relative error in stimulus energy. We preferred to base the error measure on energy rather than on luminance, because at the short presentation times used here, the visual system is more sensitive to the stimulus energy than to a luminance modulation.
The expected stimulus energy was calculated as the stimulus duration multiplied by the steady-state luminance, where the steady-state luminance was measured after an LED warm-up phase of 15 s. In our setup, we used a standard PC with a parallel port for generating the LED control signal, which, however, was only accurate to a few microseconds. Therefore, we used the measured width of the LED control signal pulse as the stimulus duration rather than the programmed value when calculating the expected stimulus energy. Moreover, the compensation circuit described earlier (see Fig. 7) was in place during the measurements, which makes a difference especially for very short presentation times. For example, if a programmed 200 μs stimulus turns out to be a mere 10 μs longer because of the differences in the switching delays (see Table 1), this would result in a systematic additional error of 10/200 = 5 %.
The errors shown in Fig. 10 are all positive: That is, the stimulus energies were systematically higher than the energies calculated from the respective stimulus duration and the steady-state luminance. This was to be expected, given that the luminous output was found to be higher for cold LEDs (see Fig. 9) and that the LEDs must have been rather cold also during the pulsed stimulus measurements, as they were switched on only once in a while and only for very brief periods of time. The variability of the measured errors—that is, the repeatability error—was nearly independent of the stimulus duration and, overall, very small (SDavg = 0.2 %).
Tachistoscopes have been widely used in vision research for experiments necessitating extremely brief visual stimulus presentations. Despite technological improvements made over the years, the available tachistoscopes remain expensive and depend strongly on customized components. Yet exposure times are still limited to the millisecond range, and timing accuracies are comparably low.
Here we have presented an inexpensive two-channel tachistoscope offering exposure durations well below one millisecond at a high precision of timing. We measured rise and fall times of the luminance output of about 3 μs, which allows for extremely short exposure durations. Such short rise and fall times greatly reduce timing uncertainties, not only while switching between the two monitors, but also with respect to the onset and duration of the stimuli. In our application, the precision of stimulus timing was only limited by the PC that controlled the tachistoscope. Despite the short rise and fall times, we found asymmetries in the switching delays of about 10 μs. We showed that these asymmetries, which cause minor switching artifacts, can be minimized by adding a simple compensation circuit to the LED control. Finally, we found the luminance of a cold LED backlight to be about 4 % higher than after it warmed up. This bias could be accounted for when calibrating the monitor or by using the two monitors in a way that avoids cooling off the LED backlights. More importantly, we found the repeatability error regarding the stimulus energy to be very low across different stimulus durations (SDavg = 0.2 %).
Other tachistoscopes adopting LCD technology have been proposed recently. Fischmeister et al. (2010) described a three-channel setup based on LCD projectors and LC shutters. LC shutters rely on the same technology as LCDs and have, at least in principle, the same shortcomings. However, the physical dimensions and the sheer number of pixels in an LCD monitor make it much more difficult to achieve fast reaction times in an LCD as compared to a single and comparatively large LC shutter. Nevertheless, the authors measured rise and fall times of approximately 1 ms and 0.25 ms (10 %–90 %), limiting exposure durations to about 2 ms, an order of magnitude longer than is possible with our approach. Moreover, as these LC shutters do not fully block the light when closed, two shutters per projector had to be mounted serially, further increasing the complexity of the setup and adding to the bill of materials. The setup showed a fairly high variance in output luminance, even when exposure time was fixed—which, however, could be attributed to peculiarities of the projector light sources, and might be resolved by using different projectors. Proper selection of the imaging device is also an issue with our setup, because LED backlights might exhibit irregular switching characteristics when they are switched off for extended periods of time (see Appendix A).
Another problem arises from semipermeable mirrors, as they cause color shifts that depend on the angle of light passing through or being reflected by the mirror (see Appendix B). Projector setups are an advantage in this respect, because they allow for the use of smaller mirrors with correspondingly higher optical quality. Moreover, due to the large throw ratio (i.e., the ratio between projection distance and screen width), the range of angles at which the light passes through or is reflected by the mirror is considerably smaller than in our compact LCD monitor setup. Moreover, these angles do not depend on viewing position, because the projection screen scatters the light. Other advantages of projector setups include a potentially high luminance output and a fairly free choice of screen size. Lately, projectors using LEDs as light sources have become available and provide sufficient luminous output power for the screen sizes typically used in visual experiments, especially in functional magnetic resonance imaging (fMRI) environments, where projector setups are commonly used for visual stimulus presentation. Such projectors would be particularly suited for modifications of the LED electronics and to be used in a tachistoscopic setup, much as we did with our standard LCD monitors.
Thurgood et al. (2010) described a one-channel tachistoscope consisting of an LC panel illuminated by a custom-made LED array, allowing for exposure durations of 1 ms and above. Although the net output luminance was not reported, it can be assumed to be much higher than the luminance output of the off-the-shelf LCD monitors that we used. However, the advantage of higher luminance output comes at the cost of having to build an LED backlight and properly integrating it with an LC panel so as to get a uniformly illuminated screen. More importantly, a one-channel setup limits the scope of application, as it only allows for brief stimulus presentations over a pitch-black background or, vice versa, a pitch-black stimulus over any background. When these limitations do not play a role, our method can be used in a one-monitor setup as well, while being simpler, cheaper, and faster than the setup described by Thurgood et al. Moreover, if the suggested multiplexer circuit is installed, the monitor can be operated in the backlight switching mode or as a standard stimulus screen. Opting for a two-channel solution with a semipermeable mirror, however, adds more flexibility, as it allows for presenting stimuli over any desired background or, alternatively, presenting two independent stimuli (over a pitch-black background), as would be required for visual masking or priming experiments, for example. In two-stimulus paradigms, the high timing resolution provided by LED backlight switching is not only beneficial for defining short exposure durations, but also for having almost unlimited control over stimulus onset times.
The development of a tachistoscope is technically demanding, and none of the available setups is perfect. Nevertheless, our approach is comparatively inexpensive, as it relies on just two standard LCD monitors, a semipermeable mirror, a mounting rack, and a PC. Even though the monitors have to be modified, the modifications can be kept simple and do not require costly hardware. Therefore, our tachistoscope makes a good compromise between what is technically achievable and what is economically reasonable.
This is not true when special techniques are applied, including dynamic contrast, flashed backlight, scanning backlight, and so forth.
This description of LCD monitors refers to standard consumer products and cannot be accurate for all different types and variations of LCD monitors.
We used MATLAB (The MathWorks, Natick, MA) with the Psychophysics Toolbox extensions (Brainard, 1997) to program the device.
Some parallel-port implementations work with a logic high level of 3.3 V, at least for the data lines. Others work with 5 V.
We thank Stefan Hudson for his invaluable ideas and advice, Alexis Hervais-Adelman for comments on an earlier version of the manuscript, and Iwan Roy for assistance in the building of the tachistoscope. This study was supported by the SNF Grant No. 320030-132967.