The research, led by Lvyun Chen and colleagues at Peking University, tackles a deceptively hard problem: can a single fixed set of five LED channels, driven only by time-varying pulse-width modulation, reproduce the photoreceptor-level signature of real daylight across an entire day? The team measured outdoor spectral irradiance on a clear summer day in Beijing, recording 27 full spectra from 05:30 to 18:30 at half-hour intervals. Solar elevation swung from roughly 6.7 degrees at dawn to 72.9 degrees at noon, driving the familiar warm-to-cool-to-warm arc of natural light, from red-rich sunrise spectra to a blue-enriched midday distribution approaching the CIE D65 standard illuminant.

Those 27 measured spectra became the optimization targets. The simulated light source comprised five channels—blue, cyan, green, yellow, and red—each modeled with a parameterized spectral power distribution whose bandwidth is not a free variable. Drawing on the physics of III-nitride and III-phosphide semiconductors, the authors tied each channel’s full width at half maximum to its peak wavelength through an empirical relation, so specifying a peak wavelength fixes the entire spectral profile. The yellow channel, constrained by the well-known green-yellow gap in direct semiconductor emission, was modeled as a phosphor-converted emitter with a broader, independently specified bandwidth, mirroring commercial practice.

The heart of the framework is a bi-level genetic algorithm. The outer layer searches for one common set of five peak wavelengths that works for the whole day, encoding candidate wavelength combinations as chromosomes and scoring each by the total accumulated error across all 27 time points. The inner layer, for every candidate wavelength set and every sampling time, optimizes the five PWM duty cycles that determine how the channels mix. Hard constraints guard visual quality throughout: correlated color temperature between 2000 and 8000 kelvin, a TM-30 color fidelity index Rf above 85, and a gamut index Rg above 95. Candidates that cannot meet these bounds at any time point receive a heavy penalty, steering the search toward physically and perceptually acceptable solutions.

Crucially, the team ran the entire optimization twice with two different objective functions. The first, a melanopic-centric model, minimized only the deviation of the melanopic daylight efficacy ratio—the standard CIE S 026 metric centered on ipRGC stimulation—from the daylight target. The second, a five-receptor integration model, minimized weighted deviations across all five α-opic daylight efficacy ratios, covering S-cones, M-cones, L-cones, rods, and ipRGCs. The weights were not arbitrary: they came from a statistical synthesis of 19 laboratory studies of human non-visual responses, giving the ipRGCs and rods the largest influence while cones contributed smaller but non-negligible terms.

The two models produced visibly different LED configurations. The five-receptor model shifted its channels toward longer wavelengths, with the green channel red-shifting by 23 nanometers and the cyan channel by 13 nanometers relative to the melanopic-centric design, and its blue channel moved 8 nanometers toward the S-cone sensitivity peak near 440 nanometers. The red channel settled at 636 nanometers, where L-cone sensitivity remains substantial. The optimized spectra were also broader and more evenly spaced, with inter-peak gaps of roughly 45 to 53 nanometers, producing smoother combined spectra that the authors link to balanced visual and non-visual performance.

When the synthesized spectra were compared against the measured daylight, the difference between the two strategies became stark. The five-receptor model tracked all five α-opic daylight efficacy ratios with relative errors mostly within 2 percent and never exceeding 5 percent across the full day. The melanopic-centric model, by contrast, matched its own target well but left large deficits elsewhere: S-cone stimulation fell 15 to 25 percent below the daylight reference, and rod-weighted quantities deviated noticeably during the morning and evening transitions. This matters biologically, because studies have shown that S-cone input can counteract melanopsin-driven melatonin suppression, that polychromatic melatonin suppression is overpredicted by melanopsin-only models, and that cones dominate the early phase of light exposure before ipRGCs take over.

Color quality told a similar story. Evaluated with the ANSI/IES TM-30 framework, the five-receptor design raised the mean color fidelity index by about 1.14 percent and reduced the variance of its fluctuations compared with the melanopic-centric result, cutting comprehensive color deviation by 9.40 percent overall. Its chromaticity points stayed within roughly ±0.0054 of the Planckian locus, tighter than the ±0.01 dispersion of the melanopic-centric model, which was free to wander because Duv was not an explicit constraint. The optimized color temperature profiles did swing more steeply than real daylight near dawn and dusk—up to about 1660 kelvin per hour rising and 1978 kelvin per hour falling—but the authors note that slow, interpolated transitions on the timescale of human chromatic adaptation would render such changes essentially imperceptible.

The framework arrives amid rising concern about circadian misalignment, which has been linked to sleep disorders, chronic insomnia, seasonal affective disorder, metabolic dysregulation, and elevated cancer risk. Meeting circadian lighting recommendations by simply boosting luminaire output can raise lighting energy use by 10 to 100 percent in offices and classrooms, so spectral tuning that delivers the right photoreceptor stimulation per unit of photopic output offers a potential efficiency dividend. By matching daylight’s full five-receptor signature rather than a single scalar, the approach aims to reproduce not just the clock-setting signal but the entire ensemble of retinal inputs that evolution tuned to the solar cycle.

The authors are careful about scope. The results rest on 27 spectra from one clear summer day at a single Beijing location, and the optimization is entirely numerical; the reported peak wavelengths and PWM schedules are candidate design parameters, not validated hardware settings. Moving to a practical luminaire will require building a five-channel prototype, calibrating radiant output, and testing thermal stability, spectral drift, driver nonlinearity, and channel crosstalk, along with human-subject studies of visual comfort and physiology. Still, as a design-stage tool, the bi-level framework offers luminaire developers a rigorous way to select LED channels and generate time-resolved control setpoints for daylight-responsive lighting in offices, classrooms, and homes—and it makes a compelling case that healthy lighting should speak to all five photoreceptors, not just one.

Subject of Research: Multi-objective optimization of five-channel LED spectra for time-resolved matching of daylight's five CIE S 026 photoreceptor-weighted circadian metrics.

Article Title: Multi-objective optimized five-channel LED simulation system using melanopic-centric model and the five-receptor integration model

Article References: Chen, L., Chen, Z., Zou, Q., Chen, X., Song, L., Dang, W., Lv, X., Zhang, H., Dong, B., Huang, F., Hu, L., Kang, X., Chen, W., Wang, Q., Tong, Y., & Shen, B. (2026). Multi-objective optimized five-channel LED simulation system using melanopic-centric model and the five-receptor integration model. Results in Optics, 25, Article 101153. https://doi.org/10.1016/j.rio.2026.101153

Image Credits: AI Generated

DOI: 10.1016/j.rio.2026.101153

Keywords: circadian lighting, five-channel LED, melanopic DER, CIE S 026, α-opic metrics, genetic algorithm, daylight matching, ipRGC, TM-30 color rendition, pulse-width modulation, non-visual effects of light, tunable white lighting