Project Results

Methods for traceable and validated measurement of temporal light modulation (Objectives 1 and 2)

Based on the models given in IEC TR 63158:2018 and IEC TR 61547-1:2020, uncertainty components, which affect TLM quantities for flicker and the stroboscopic effect, have been identified. To propagate uncertainties, from the time domain to PstLM and SVM, models have been built. Using these models, sensitivity coefficients for uncertainty propagation have been determined for various waveforms. This uncertainty analysis will be used for the calibration of TLM measurement devices. Further investigation into the models revealed shortcomings of the current definitions as well as of reference implementations of TLA metrics. In addition, the improved models have been implemented in a luminous flux measurement setup which has been used to measure a large number of light sources for TLM.

For validation of implementations of TLM models a dataset, containing discretised mathematically generated waveforms, named “MetTLM TLM waveform set 1”, has been generated and validated. The dataset is accompanied by a report guiding and exemplifying how the dataset can be used by laboratories to achieve measurement uncertainties below 0.05 on PstLM and SVM. The dataset and report have been released to the MetTLM community on Zenodo, an open access repository.

Typical performance of measurement devices can be expressed in quality indices, which characterise how a physical effect influences the instrument’s reading. For TLM measurement devices, quality indices have been defined for frequency response and dynamic range of signal. An LED-based facility has been built with the aim of characterising TLM measurement devices, which will be used to assess dynamic range. A laser-based facility has been realised, and procedures to measure the frequency response of TLM measurement devices have been tested. The frequency response of various commercially available TLM measurement devices has been characterised and compared against the developed TLM models, for flicker and the stroboscopic effect. An approach for a quality index for frequency response has been tested and further developed. Quality indices can be used for instrument classification, helping prospective instrument buyers selecting suitable TLM measurement equipment.

To validate the traceable TLM measurement methods developed in Objective 1, an interlaboratory comparison has been carried out. Artefacts (four waveforms of a TLA box and 7 lamps, of which 4 are also measured in the IEA 4E SSL Annex comparison) were selected and aged prior to the comparison. The eight participants measured the artefacts using their facilities, and data was analysed according to the “Guidelines for CCPR Key Comparison Report Preparation”. These results show that overall the participants are in consensus about lamps entering the EU market, regardless the differences in measured value. The results of the TLA box are consistent between participants, likely due to elimination of the effect of external power supplies, making this kind of source ideal for validation of measurements against the Ecodesign regulations.

Novel methods for TLM measurement (Objective 3)

To probe individual light sources inside complex field scenes a luminance TLM meter based on a fast photocurrent flicker-meter, including an anti-aliasing filter, and a luminance photometer head was set up. This luminance TLM meter has been verified inside the lab and thereafter used in field measurements. To enabling a multi-channel TLM meter with synchronized acquisition a trigger extension for this fast photocurrent meter was initiated in this project. Using an adopted software development kit a multi-channel configuration has been set up, verified and used to demonstrate the advantages of parallel TLM measurements. First results have been disseminated in a conference presentation and a training session. Both the luminance TLM meter and the multi-channel TLM meter are now available as commercial products.

In laboratory-based measurements, a set of three TLM luminance sources with patterned transmissive filters have been used to generate luminance contrast patterns which are then measured by using cameras. Doing so, limitations identified regarding the sampling theorem, resulting from the charge accumulating principle as used in most pixel-based detectors, have been addressed. The linearity of the TLM luminance source in constant luminance mode and during transient operation (regarding the actual TLM waveform compared to the nominal one) was investigated which revealed issues regarding modulation depth (offset) and small deviations resulting from internal decay time constants of the electrical circuit of the TLM luminance source, in addition to the well-known droop effects attributed to the included LED strains itself. This information is a prerequisite for facilitating these sources in a characterization of TLM measurement devices.

An installation of TLM and temporal colour modulation from smart lighting products (RGB and tunable white LED lamps) was assembled for visualizing TLM by the rolling shutter of camera sensors with evidence of limitations regarding reliable measurements and assessments of TLM metrics. These examples also demonstrated the needs and advantages regarding measurement of TLM and temporal colour modulation, i.e. regarding spectral mismatch of flicker meters.

An accurate method for the measurement of temporal colour modulation was developed by setting up a 4-channel tristimulus detector head and four of the fast photocurrent meters, coupled by a common trigger signal. This creates a unique tristimulus-TLM meter that allows high-speed measurement of the tristimulus waveforms and temporal evolution of colour coordinates and demonstrated the advantages regarding measurement of TLM and temporal color modulation. For a measurement of faster modulations, i.e. with components above 10 kHz, a photocurrent meter with a low-pass filter of 100 kHz was used, which can also be used in a triggered mode to enable a synchronized acquisition for multi-channel measurement of RGB LEDs at short PWM duty cycles of just a few 10 µs.

In an experimental study, conducted in an environment illuminated with multiple light sources, image sequences at frame rates of 8 kHz and 4 kHz have been taken with RGB cameras. For each colour channel of the cameras, (namely, red, green, and blue) the TLM waveforms have been extracted for a region of interest marked in the image sequences. The results reveal the operation principle of tuneable white LED-based lamps, which consist of various types of white LEDs or RGB-LEDs. The study underlined the need to evaluate TLM by (multi-)spectral and spatially resolved measurements. Vivid examples have been obtained by imaging TLM measurements of field scenes: a Christmas tree with different fairy lights; car headlights and daytime running lights; road lighting; a car dashboard with head-up display; E27 socketed LED-based lamps providing RGBW and tuneable white light. Heat maps of SVM have been generated for relevant TLM metrics.

A commercial high-speed RGB camera was used for multi-spectral TLM imaging measurements in lab-based and field scenes using real time sampling and equivalent time sampling. For multi- and hyper-spectral TLM measurement, a hyperspectral camera was used to measure LED luminaires in office scenes.

Measurements taken with an imaging luminance measurement device (ILMD), on different lamps and luminaires, demonstrated the feasibility of TLM measurements with such devices.  Results from this feasibility demonstration resulted in an improvement of the TLM measurement modes: The use of the ILMD to implement a TLM imaging measurement was hindered by issues which had been reported to the manufacturer and were fixed in an revised version of the control software and the device-internal firmware. This solution was successfully verified during the project but not yet picked up by means of demonstrating TLM imaging measurement by an ILMD. Instead, industrial machine-vision cameras (monochrome and RGB) were used to implement and demonstrate also an Equivalent-Time-Sampling (ETS) mode for spatially resolved TLM measurements well above the aliasing frequency. The industrial cameras lack a V(λ) matching but their lenses allow a change of the aperture (in contrast to many ILMDs that use a fixed aperture) to adjust the device responsivity to the luminance level of the scene. The implemented ETS measurement mode successfully demonstrated the possibility to determine the waveforms of TLM inside complex scenes with cameras. Because the parameter range for the integration time is limited by the required frequency resolution, the adaptation to the luminance of the scene is mainly done by adjusting the lens aperture and applying neutral density filters. The main issue that limits the TLM measurement by cameras is the low dynamic range compared to traditional systems, i.e. using a photometer head and a fast photocurrent meter. For scenes with sufficiently low dynamic range contrast, this measurements can be executed automatically. The method of generating the phase shift for the ETS measurement, i.e. by synchronizing it to the mains frequency by a delay trigger or by a continuous sampling, determines whether the waveform measurement is done for the full scene at once or sequentially for recognized TLM regions. The latter requires to analyze the scene for finding regions showing TLM, grouping them by waveform properties to virtual luminaires, and determine the individual base frequency (modulation period) of these regions in a scene image under the condition, that the measured signal is not band-limited, which complicates the signal analysis. The ETS mode was successfully demonstrated by waveforms for typical light sources, i.e. of many hundred Hz and PWM duty cycles above 10%, which present significant TLA. Using the ETS mode with an industrial RGB camera also the measurement of temporal colour modulation from from smart lighting products (RGB and tunable white LED lamps) was successfully demonstrated.

Also, the impact of TLM on ILMD measurements of the average luminance and measurements of spectral irradiance by array-spectroradiometers was demonstrated. Errors as encountered during luminance measurement for glare assessment from artificial light at night caused by high-intensity discharge lamps (HID-lamps), or pulse-width-modulated LEDs have been studied. In addition, the possibility of using conventional cameras that provide a high frame rate mode of up to 1000 Hz, such as compact cameras or smartphone cameras dedicated for slow motion recordings, were investigated. In contrast to these, photos obtained with a long integration time of i.e. 0.05 s or 0.1 s captured during a camera pan can give visual evidence for the phantom array effect. As such cameras are widely used, this is expected to increase the uptake of results.

Model for the visibility of the phantom array effect (Objective 4)

Based on an initial literature review, five psychophysical experiments were designed to study the effect of temporal frequency, colour of the light source, saccade amplitude and velocity, and ambient illumination on the visibility of the phantom array effect. The experimental protocols for experiments 1-4 were approved by the Ethical Review Board (ERB) at Eindhoven University of Technology, and for experiment 5, by the Swedish Ethical Review Authority. All five experiments used a two-interval forced-choice (2IFC) task for the observers, in which observers need to indicate in which of the two sequentially presented stimuli the phantom array effect is visible to them. Changing the modulation depth in the pair of stimuli in combination with the QUEST+ method (a Bayesian adaptive psychometric testing method), enables adaptive collection of data, thus reducing the number of perceptual experiments needed. By doing so, the visibility threshold of the phantom array effect could be determined for the various lighting conditions.

In experiments 2, 3 and 4, a narrow slit white light source was used. Experiment 2 focused on modelling the temporal contrast sensitivity function to the phantom array effect. In this experiment 22 participants were included, and 10 different frequencies were tested. The resulting sensitivity as a function of frequency, averaged over all participants shows that the sensitivity is clearly higher at the medium frequencies, with the maximum at 600 Hz and the sensitivities are substantially lower at the two far ends of the measured frequency range.

Experiments 3 and 4 focused on the effect of saccade-related characteristics (i.e., saccade amplitude and velocity) and the effect of ambient illumination on the visibility of the phantom array effect. The use of eye-tracking technology helped us understand to what extent the differences in the saccade speeds can explain the individual differences in perception. At first, the effect of saccade amplitude on the visibility of the phantom array effect for one light condition (i.e., in the dark) was investigated and the visibility threshold was determined at seven temporal modulation frequencies. Secondly, experiments were performed in office lighting conditions for a subset of the participants. Results show that the sensitivity is frequency dependent, and that the ambient light level has a substantial effect on the visibility of the Phantom Array effect. It is much more difficult to observe when the contrast is lower.

In experiment 5 the results of the previous experiments were verified in real-life context. One of the most common situations when the Phantom Array effect is reported is when viewing temporally modulated taillights of a car in low-light conditions. Therefore, a setup was constructed in the laboratory simulating viewing conditions closely resembling the real-life situation when driving behind a car with modulated taillights. The set-up used taillights from a Volvo XC60. To mimic real-world conditions, with pulse-width modulated taillights, a waveform generator and an amplifier were used to generate square waves with 50% duty cycle. The light output was controlled via a LabVIEW program. A total of 20 participants were included in the test. For this experiment, instead of using an adaptive procedure, predefined levels of MDs for each frequency were used. The settings were selected based on pilot experiments. Data showed individual variations in the threshold values for the visibility of the phantom array effect, but the frequency dependence was consistent among most participants.

A psychometric function (the Weibull cumulative distribution function) was used to fit the data across all included frequencies. Based on the visibility threshold determined at each frequency, a visibility threshold curve could be obtained showing the similar band pass shape as the more controlled experiments 1-3 albeit with a slightly higher peak frequency and a flatter appearance. This may be partially attributed to the use of square wave modulation instead of sine wave modulation. In this experiment, we utilized two light sources with a curved design, whereas experiment 1-3 used one single source (a narrow slit). Additionally, no chin rest was used in this experiment, allowing the observers to move their eyes more freely. Consequently, this brings new insights into the visibility of the phantom array in real-life situations. With increased knowledge, more informed decisions about limit values in different settings can be made, thereby improving the safety, comfort, and overall quality of LED-based lighting systems.

A description of the setup (for experiments 1 and 2) and methods were presented at the CIE Expert Tutorial and Symposium on the Measurement of Temporal Light Modulation in Athens, Greece, October 2022. The results of experiment 1 was presented at the CIE 2023 conference in Ljubljana, Slovenia, September 2023 and the results from experiment 5 will be presented at the IES conference in New York, USA, August 2024. The results from experiments 2-5 are drafted in three separate scientific papers that will be submitted for publication in peer-reviewed journals.