1. Introduction

Today, in many lighting applications LEDs are becoming the preferred means of illumination. There are a few applications, however, where gas-discharge lamps, like UHP and Xenon lamps, are still superior. Especially arc lamps can provide a small source with high brightness and are used in applications like digital projection [1], automotive lighting and special spot lights, e.g. for arena lighting. The main drawbacks of these lamps are a short life time, the use of mercury (in some cases) and limited lumen maintenance over life-time (ageing) as well as catastrophic failure. Conventionally, it is not possible to use LEDs in applications where a high brightness is required, i.e. the étendue of the light source must be small. Naively one could think that it would be possible to combine the light of many small LEDs into one bright source. However, because of étendue conservation it is impossible to combine the light in a small area, while maintaining a low angular divergence. The resulting source may have high optical power, but not high radiance (i.e. power per surface area per solid angle). Below we will discuss how this étendue conservation can be circumvented in a well-designed LED-phosphor combination that acts as a concentrating light source. Most white LED sources nowadays make use of blue LEDs combined with phosphors. These sources, however, have been designed to obtain the desired color properties with high (lumen/Watt) efficacy and not to obtain high brightness.

In this paper, we will first address fundamental considerations of making high-brightness sources and argue that this can be done by luminescent concentration. Next, we will show that it is possible to realize such a source in practice and discuss experimental results. These will be compared to modeling results, which give insight in the attainable performance. Finally, we will discuss the limitations and prospects of these sources.

2. Fundamentals

When a device (e.g. a lens) is used to combine light from several incoming light sources (or an extended light source), the radiance (power per unit solid angle per unit projected area) $L_{\text{out}}$ of the combined outgoing light is restricted. The law of conservation of étendue [2] implies that the maximum attainable radiance concentration is ${(L_{\text{out}}/L_{\text{in}})}_{\max }={(n_{\text{out}}/n_{\text{in}})}^2$, where $L_{\text{in}}$ is the radiance of the incoming light source(s) and $n_{\text{in}}$ and $n_{\text{out}}$ are the refractive indices, respectively, of the media of the incoming light source(s) and of the combined outgoing light. This restriction is greatly relieved in case of a luminescent device. In a luminescent concentrator, short-wavelength light is converted into longer-wavelength light, which is guided towards the edges. If the energy difference (Stokes shift) between the incident short-wavelength and emitted long-wavelength light is $\Delta E=E_{\text{in}}-E_{\text{out}}$, the maximum attainable radiance concentration equals ${(L_{\text{out}}/L_{\text{in}})}_{\max }={(n_{\text{out}}/n_{\text{in}})}^2{(E_{\text{out}}/E_{\text{in}})}^3\exp (\Delta E/\text{k}T)$, where T is the temperature of the concentrator [3,4]. In such a device, étendue conservation seems to be violated, but the second law of thermodynamics is not. The crux is that the heat $\Delta E$ generated in the luminescence process can be exploited to lower the entropy of the light by an amount $\Delta E/T$, concentrating the light. The underlying physics is that each luminescent site can be regarded as an independent new source. This phenomenon is exploited in luminescent solar concentrators (LSCs), in which sunlight is converted into longer-wavelength light which then is guided to small photovoltaic cells [5–7]. In practice, the concentration of such LSCs is restricted (see Section 6), since their performance is hampered by severe losses. In the case of a luminescent concentrating light source, partly the same but also different loss mechanisms play a role, as will be discussed below.

3. Constituting parts of a luminescent light source

Figure 1(a) shows schematically the essential parts of a luminescent light source [8]. The light from several blue LEDs is coupled into a rectangular rod-shaped transparent (i.e., non-scattering) luminescent material. To couple as much blue light as possible into the luminescent rod, the LEDs are placed as close as possible to it. Most of the incident light is absorbed and converted into light of longer wavelength, which is emitted in all directions. Part of this emitted light escapes from the four long edges of the rod; the greater part is internally guided by total internal reflection (TIR) towards the small edges of the rod. At one of these, a mirror is placed to direct all light to the other small edge, where it is extracted by a suitable structure or device. In the figure, a compound parabolic concentrator (CPC) is used [9].

 

Fig. 1 (a) Schematics of luminescent light source: light emitted from blue LEDs is absorbed in a luminescent rod. Most of the emitted light is guided by total internal reflection to an edge of the rod, where it is extracted, e.g. using a CPC. At the other edge, a mirror is placed. (b) Mechanical drawing of the module. Through the rectangular aperture at the front, the exit window of the CPC is visible, which is mounted on the luminescent rod.

Download Full Size | PDF

Crucial for the application is a proper choice of the luminescent material. In the case of LSCs, plastics filled with organic fluorescent dyes are widely applied [6]. This is not the best choice for a high-brightness light source, for which a material is required that can withstand high temperatures and fluxes. An obvious choice is cerium-doped yttrium aluminium garnet (YAG:Ce), a material widely used for lighting applications. However, the light needs to be guided over large distances without being scattered. For that reason we will use highly transparent YAG:Ce rods. The concentration of Ce is chosen such that nearly all incident blue light is absorbed. This can be easily achieved since the incident spectrum from the blue LEDs has a very limited wavelength range. This is a difference with solar applications, where only a limited part of the incident spectrum can be covered by the absorption spectrum of the luminescent material [7].

In both applications, the difference between the energy of the incident and luminescent light is lost. In an LSC this conversion loss is not an additional loss in efficiency, since this amount of energy would be lost in the solar cell otherwise. In a luminescent light source, it is one of the main loss mechanisms. Moreover, the heat generated inside the luminescent rod due to this conversion loss has to be removed, as will be discussed below.

Similarities with the LSC application are the presence of losses due to quantum efficiency (QE), guiding (i.e, escape of light not in TIR) and reabsorption of the converted light. Fortunately, the QE of YAG:Ce can be 95% or more. Since YAG has a high refractive index (n = 1.83), the critical angle for TIR is only θ_c = 33° and the guiding loss, ½ (1- cos θ_c) [7,8], is limited to 8% per rod side. Because of the rod-like geometry, escape occurs from four sides. The converted light that is guided inside the rod can be reabsorbed since YAG:Ce has a broad absorption spectrum that overlaps with the emission spectrum (see Fig. 2). Because of the high QE, most reabsorbed light will be reemitted, but at higher wavelengths. Again, part of the reemitted light will escape from one of the four sides. After one or two reabsorption events, the wavelength is so much shifted that no further reabsorption takes place.

 

Fig. 2 Absorption coefficient (purple line) of YAG:0.23% Ce and emission spectra of powder (red); see Section 4. Also the output spectrum measured from the edge of a YAG:0.23% Ce rod (green, Section 4) and simulated with as input the powder emission spectrum (dashed blue line, Section 5) are shown.

Download Full Size | PDF

Without special measures, the extraction efficiency would be only 8% from the small exit edge; with a mirror at the other side this would amount to 16% extraction efficiency. This can be further enhanced by use of a suitable extraction structure. Refractive extraction structures (like microlenses or small pyramids) or scattering structures on the exit edge may be used [8], but the best extraction efficiencies were obtained using a CPC optic. We use a rectangular CPC [9] with an exit half-angle divergence of 34°, of which the entrance facet fits the exit edge of the luminescent rod. In theory [2], close to 100% of the light reaching the exit facet of such a three-dimensional CPC may be extracted. Actually, the CPC has a two-fold function. Besides extracting the light, it also shapes the light beam to a well-defined angular divergence and exit area. This gives the possibility to design the high-brightness source such that it suits a particular application. In the following Sections, practical numbers will be discussed as well as more realistic theoretical estimates, based on the used geometry and materials.

4. Experiments

A light module has been made like sketched in Fig. 1(a) and shown in Fig. 1(b) with the following constituents. We use a rod of YAG:0.23% Ce of dimensions 52 × 1.9 × 1.2 mm³. A rectangular CPC, made of glass with refractive index 1.52, is attached to the rod in optical contact. Its entrance and exit facets have dimensions of 1.9 × 1.2 mm² and 5.2 × 3.3 mm², respectively. At the rod end, a mirror is used with a reflectivity of > 97% (Alanod). At the largest sides of the rod, 56 LEDs (Lumileds LUXEON Z-ES Royal Blue [10]) of 2 mm² surface area per LED, emitting at 445 nm, are accommodated at close distance. They are operated at 3.4 V and 1.7 A maximum, at which current the output is 1.5 W of blue light (radiance = 0.24 W/mm²/sr). The LED boards are attached to copper cooling blocks. Also the rod is held by two copper cooling blocks, while avoiding optical contact. The temperature of the rods is maintained below 100° C.

The rod material is characterized in the following way. The absorption coefficient (Fig. 2, purple line) is obtained from a transmission spectrum (Perkin Elmer Lambda 950 UV-VIS spectrometer) measured using a slice of YAG:Ce of 0.3 mm thickness. The emission spectrum (Fig. 2, red line) is measured for a thin layer of YAG:Ce powder (using an Edinburgh Instruments FLS980 Spectrometer). It is assumed that this is the intrinsic emission spectrum without the effect of reabsorption. The QE of the material is 95%, measured in an integrating sphere using excitation at 450 nm.

The module performance is assessed in an integrating sphere (Instrument Systems). Using a fitting flange, the CPC exit end is just located inside the integrating sphere to measure the output flux and spectrum. The measured output spectrum is shown in Fig. 2 (green line) and has a lumen equivalent of 480 lm/W_opt. The optical power from the blue LEDs versus electrical power is measured in a separate measurement. The determined output power as a function of incident optical power is shown in Fig. 3(a). The slope of this curve can be considered as the optical efficiency of the system, i.e. the fraction of optical input power that is converted to useful output power. It is 0.220 at low input power and slightly decreases to about 0.215 at high input power. Figure 3(a) shows that the total optical output can reach close to 18 W (8500 lm) at an input optical power of 84 W. With an étendue of 17 mm² sr, this corresponds to a radiance of 1.1 W/(mm² sr) and a luminance of 500 cd/mm².

 

Fig. 3 (a) Measured output optical power and luminance vs incident optical power. (b) Relative optical output (normalized to that at 52 mm) vs rod length: measured (red) and simulated (blue, see Section 5). The dashed line indicates the behaviour if the relation would be linear.

Download Full Size | PDF

We also have assessed the scalability of the concept as a function of rod length. To this end, we have made rods of several lengths and measured the optical output (at low input power). The number of LEDs used is increased linearly with rod length as well. The results are shown in Fig. 3(b). It is seen that the output increases more or less linearly with rod length. Although one expects that at longer rod lengths the slope decreases because of scattering and reabsorption (blue line, see Section 5), no decrease is seen within the experimental accuracy for the available rod lengths. Hence, the concept is scalable to a large extent in both optical input power and rod length.

5. Simulations

Simulations are performed in a manner very much alike that used for luminescent solar concentrators [7] with LightTools ray-tracing software [11]. The luminescent rod material is characterized by a wavelength-dependent refractive index [12] and absorption coefficient (see Fig. 2). The emission spectrum is approximated by that of a fine powder (see Fig. 2) and QE = 0.95. It is assumed that the volume scatter of the luminescent material is described by the Henyey-Greenstein model [11] with a scatter length of 500 mm, whereas surface scatter is neglected. The rod dimensions are 52 × 1.9 × 1.2 mm³ unless stated otherwise.

As discussed above, a rectangular CPC with an exit half-angle divergence of 34° is used of which the small window fits the exit edge of the luminescent rod. The refractive index of the CPC is 1.52. The CPC is in optical contact with the rod. At the rod end, a 98% reflective mirror is placed (with a small air gap). The LEDs are approximated by (52 × 1.9 mm²) rectangular light sources along the wide sides of the rod. The LED spectrum is that measured for the used LUXEON Z-ES LEDs (essentially a Gaussian with peak wavelength 445 nm and FWHM 20 nm) and the directionality is taken to be Lambertian [10]. The reflectivity of the LED sources is taken to be 50% (which has to be considered as an average between that of the LED dyes and the printed circuit boards). The reflectivity of the copper blocks surrounding the other two long rod sides (assumed to be coated with a mirror-like layer) is taken 90%.

It should be noted that the reflectivity of the surrounding copper blocks and LEDs has only a minor influence on the output of the module. The main effect is that the blue light from the LEDs that does not reach the rod directly can reach the rod via the reflective bodies. Also part of the converted light that escapes the rod could in principle be directed towards the rod front facet. However, because of étendue conservation, this part should not be more than the ratio between the areas of the front facet and the rod sides, unless further conversion takes place (cf. Section 2). Indeed, reabsorption and reemission will occur (cf. Section 3), but only for a small fraction of the converted light for which the absorption coefficient is appreciable. These effects are accounted for in the simulations (where it is noted that this kind of ray-trace simulations may erroneously violate étendue conservation if all absorption is set to zero).

We simulated the output spectrum, as well as the optical efficiency of the system. The simulated shape of the output spectrum (Fig. 2, blue dashed line) agrees very well with the measured one (Fig. 2, green line). The optical efficiency, defined as the fraction of optical input power that is converted to useful output power (i.e. leaving the CPC), is found to be 0.26. The losses in the rod due to conversion, guiding and extraction into glass amount to an efficiency of ca. 0.35. Smaller loss contributions are due to limited incoupling and collimation plus extraction into air. The total optical module efficiency of 0.26 found by simulation is somewhat larger than the experimental value of 0.22 (Fig. 3). A possible explanation is surface scatter, which is neglected in the simulation. This might be in agreement with the slight overestimation of the red shift in the simulation, which indicates that the path length in the rod possibly is somewhat smaller than simulated.

We also simulated the output as a function of rod length. The results are shown in Fig. 3(b) (blue line). The relative optical output vs. rod length is slightly declining at high rod lengths, because of higher loss due to scatter and reabsorption. In the measurements, the experimental uncertainties are too large to see this effect for the available rod lengths.

6. Discussion and conclusion

We have shown that a luminescent concentrator YAG:Ce rod enables a high-brightness source with an étendue of 17 mm² sr and a luminous flux of 8500 lm and a luminance (‘brightness’) of 500 cd/mm² using 56 high-power pump LEDs at 330 W electrical input. This translates to an efficacy of 26 lm/W. The luminance of this source is more than an order of magnitude higher than that of LEDs in the green wavelength range, but still half an order of magnitude below that of commonly used arc-based lamps. The efficiency in terms of optical input power converted to useful output power is 0.22. The theoretical modeling prediction for this efficiency is 0.26. A possible explanation for this deviation is surface scatter in the rod.

Although LSCs are different devices, it is interesting to make a comparison with them. It was found before [13] that a plate-shaped LSC filled with an organic luminescent dye can have an optical collection probability (outgoing converted photons / incident photons) of 0.19 and a concentration ( = ratio between outgoing and incoming irradiance in terms of photons per steradian per unit area) of 1.9 (for dimensions 100 × 100 × 3 mm³). In terms of photons, the number for the luminescent source translates to an optical collection probability of 0.28. The concentration in terms of photons is 8 if the ratio is taken of the source radiance (1.1 W/mm²/sr) and the irradiance of the rod sides (0.13 W/mm²/sr) . These numbers cannot be directly compared, since in an LSC the incident solar spectrum is not completely absorbed (efficiency 0.3). For the luminescent source, only blue light is absorbed, with an efficiency close to 1. In both cases, the conversion loss is appreciable, but as argued above this number does not enter in the LSC efficiency.

To assess the concentration of the luminescent source, it is more elucidating to consider the ratio of the outgoing radiance (1.1 W/mm²/sr) and the incoming radiance (0.24 W/mm²/sr) from the LEDs. This concentration factor is 4.5. So, despite its limited optical efficiency the irradiance has increased significantly. This number can be increased even more, since the concept of a luminescent source is scalable to a high extent. As a function of length, no significant decline of efficiency has been noticed.

The presently achieved flux and luminance already approach those of an arc-based lamp (approx. 20000 lm, 3000 cd/mm²) up to half an order of magnitude. Of course, the usability of the mentioned luminous flux depends on the application. To make white light with a desired color temperature, a certain amount of blue light needs to be mixed in. For digital projection purposes, green light (approximately 500 < λ < 600 nm) and red light (approximately 600 < λ < 700 nm) may be needed separately. The spectrum of YAG:Ce is favorable for neither. It is well known [14] that the spectrum can be blue shifted by substituting a larger ion (e.g. Sc or Ga) for Al or a smaller ion (e.g. Lu) for Y and red shifted by substituting a larger ion (like Gd or La) for Y. In this way, the composition may be tuned to obtain a desired spectrum, suiting a particular application.

In conclusion, we have introduced a scalable concept for LED-based luminescent light sources. In practice, these can already be used as mid-segment projector sources, for instance. Because of the scalability, there is a prospect for higher brightness.