Experimental Characterization of Single-Color Power LEDs Used as Photodetectors

Semiconductor-based light emitting diodes can be used for photon emission as well as for detection of photons. In this paper, we present a fair comparison between off-the-shelf power Light emitting diodes (LEDs) and a silicon photodetector with respect to their spectral, temporal, and spatial properties. The examined LED series features unexpected good sensitivity and distinct optical bandpass characteristic suitable for daylight filtering or color selectivity. Primary application is short range optical underwater communication, but results are generally applicable.

are applied. A gigabit per second (Gbps) visible light communication (VLC) system based on yellow LEDs as receiver is introduced in [13], and, in [14,15], a communication system by which two LED devices are used for two-way visible light communication while also providing illumination. Visible light communication and related tasks-like spectral-based intensity detection for the purpose of high-quality human central lighting-are among the driving forces in LED research [16].
Because LEDs are not intended to be used as photodetectors, manufacturers do not provide specifications about their response function, spectral sensitivity, or dynamic range [8]. For this reason, an experimental characterization of LEDs is necessary when used as photodetectors [17].
Original contributions of this article include: • The wavelength-dependent responsivity (in A/W) is determined for single-color power LEDs when used as photodetectors. Power LEDs are shown to behave quite differently compared to low-power LEDs probed so far.

•
The spectral sensitivity is compared with a typical silicon PD and the theoretical bound. It is shown that the gap with respect to the optimum responsivity is small.

•
The dynamic behavior in terms of rise/fall time and junction capacitance is investigated.

•
The optical field of view (FOV) is compared for emitter and detector mode.

•
The impact of light polarization is tested.
Experimental results are obtained for two off-the-shelf color LED series with flat lensless surfaces. Unlike low-power LEDs studied in previous publications (see e.g., [17]), no plastic lenses needed to be rubbed off and the photosensitive area (necessary for a computation of the responsivity) could be precisely taken from the datasheet. Our main emphasis is on optical underwater applications [18,19] because, in this area, single-color high power LEDs are needed and because optical (colored glass or thin film) filtering for the purpose of ambient light suppression is troublesome. However, the results reported next are universally applicable.
The remainder is organized as follows. In Section 2, the experimental setups under investigation are presented. Numerical results are reported in Section 3. Finally, conclusions are drawn in Section 4.

Description of Experimental Setups
For visible light communications, particularly in underwater applications, single-color power LEDs are the first choice for low-cost and short-range applications. Compared to white LEDs, the market only offers a limited selection of single-colored power and high-power LEDs, respectively. Regarding the suitability as photodetector, multiple-die and converted types are beyond the scope of this contribution. Popular low-cost single-die power LEDs in the 1 mm 2 chip size class include the Osram Golden Dragon series (Osram Semiconductor, Regensburg, Germany) and the Lumileds Luxeon Z color series (Lumileds Holding BV, Schipohl, The Netherlands), see also Figure 1, the latter offered in a rarely found large variety of colors. The few high-power LEDs on the market with larger single dies, for example a 12 mm 2 series offered by Luminus (Luminus Inc., Sunnyvale, CA, USA), are mostly available only in red, green, and blue colors. This generally high-priced segment is not necessarily interesting for a second application as PD. The two series mentioned, Osram Golden Dragon [20] and Lumileds Luxeon Z [21], and the silicon positive intrinsic negative (Si PIN) photodetector Osram SFH 2400 [22] were chosen for direct comparison, and they are specified in Tables 1 and 2. They have the same active area and shape and are all planar types without primary optics, offering a typical FOV of 120 • full width half mean (FWHM). This selection therefore allows comparative measurements to be carried out under repeatable conditions. Experiments were conducted without applying a reverse voltage to the device under test (DUT).

Spectral Measurements
To determine the spectral sensitivity of a photodetector, monochromatic light or light with a small optical bandwidth is needed. In order to maintain this, a light source with a wide spectrum like a halogen bulb can be combined with a monochromator that is separating the wavelength of the light spatially. For our experiments, a grating type monochromator Oriel 77250 (Newport Corporation, Irvine, CA, USA) was used in combination with halogen source Schott KL1500 (Schott AG, Mainz, Germany), see also Figure 2. The externally stabilized light source was operated at 70 W with a usable continuous spectrum of approximately 400 nm to 750 nm. The light of the monochromators' output hitting the 1 by 1 mm active area of the DUT at a distance of 50 mm has an optical bandwidth of a few nanometers. The generated photocurrent is measured directly by a Keithley 6517 electrometer (Keithley, Solon, OH, USA) in the nanoampere range at a monochromator step size of 10 nm.

Temporal Measurements
The bandwidth of a photodetector can be determined by measuring the rise time t rm of the impulse response. Ideally, a fast light source providing a rise time t rs t rm is used for this task. In that case, t rs can be neglected. However, since it is intended to use a relative slow light source employing a power LED, t rs needs to be identified first. Power LEDs including driver are known to have typical bandwidths in the range from a few MHz to tens of MHz. Figure 3 shows the configuration for temporal measurements, comprising a signal generator Rigol DG5072 (Rigol Technologies Inc., Suzhou, China), a TC4452 driver (Microchip Technology Inc., Chandler, AZ, USA) including the LED as DUT, a Thorlabs PDA-10A 150 MHz transimpedance amplifier (TIA) PD module (Thorlabs Inc., Newton, NJ, USA), and an R&S HMO3004 digital oscilloscope (Rohde & Schwarz, Munich, Germany). The oscilloscope is directly providing the 10% to 90% rise time of the measured signal. For measuring the rise time of the DUT, the setup in Figure 4 uses the source introduced in Figure 3. The optical bandwidth of the sourcing LED should overlap well with the band of the DUT. The generated photocurrent is fed into a transimpedance amplifier THS4631 evaluation board (Texas Intruments, Dallas, TX, USA), which offers a gain bandwidth product (GBP) of 210 MHz and is configured with a feedback resistance R f of 47 kΩ and a feedback capacitance C f to accomplish a quality factor of approximately Q = 0.7. The capacitance value C was measured with a Wavetek LCR55 m (Wavetek Corp., San Diego, CA, USA). Using a TIA calculator is very helpful at this point, available online at [23]. The evaluation of the response signal stored by the oscilloscope delivers the rise/fall time and provides an estimate of the achieved quality factor. The source is configured as in Figure 3. The irradiated DUT is connected to a transimpedance amplifier. Its pulse shape is recorded by a digital oscilloscope.

Spatial Measurements
The angle of incidence (AOI) is an important parameter when dealing with optical systems. An optical bench for precise alignment and good repeatability is utilized. For spatial measurements, light from a stabilized LZ4 series LED source (Ledengin, San Jose, CA, USA) in blue, respectively, amber color with wavelength matching the DUTs, is used. The DUT is mounted on a rotatable device, to be able to adjust the angle of incidence, see Figure 5. Such setup is also known as a goniometer. The generated photocurrent is measured precisely with a Keithley 6517 electrometer in the nanoampere range. Measurements were conducted for a green and red Osram LED used as photodetector and for a green Lumileds Z LED to allow for minimal comparison within a series and between series, since no major deviations from the radiation characteristics as LED source are expected. AOIs between 0 • and 80 • taken in 10 • steps were recorded. Based on the measurements with varying AOI, the FOV can be determined. Another test is to check whether the DUT reacts unexpectedly to changes in the direction of polarization. For this purpose, non-polarized light from an LED source with the appropriate wavelength matching the DUTs passing band is filtered through a rotatable Edmund glass polarizer 53344 (Edmund Optics, Barrington, NJ, USA), see Figure 6. By turning the filter by 90 • , the polarization direction can be changed from horizontal to vertical. Due to the effort, the full series was not examined, as DUTs red LEDs of both examined series were chosen exemplary. An amber-colored Ledengin LZ4 LED served as the source, operating in constant current mode under stable conditions. Photocurrent measurements were made directly with a Keithley 6514 electrometer, taken in 10 • rotation increments.

Analysis in the Spectral Domain
Figures 7 and 8 depict the wavelength-dependent responsitivity of single-color power LEDs deployed as PDs. The sensitivity maximum of the DUTs is approximately 50% to 100% of the reference Si PIN PD at the appropriate peak wavelength, which is a comparatively high sensitivity. The theoretical responsitivity bound of an ideal Si PIN PD with a quantum efficiency (QE) of 100% is shown for reference purposes [24]. In Figures 9 and 10, the spectral characteristics are normalized to compare the intensity as LED and the sensitivity as PD for two series of power LEDs in four, respectively, seven colors. The LED spectrum was measured at 300 mA forward current and 25 • C ambient temperature using a BTS 256 Spectrometer (Gigahertz Optik GmbH, Türkenfeld, Germany). Unfortunately, the full spectral bandwidth of blue to green LEDs as PDs can not be shown, as it is limited to wavelengths above 400 nm, due to the confined spectral range of the light source. Nevertheless, the position, the width, and the overlap of the spectral intensity respectively sensitivity can be identified. The spectral overlap of all DUTs is relatively small, which would result in a reduced efficiency if the same LED type would be used as transmitter and receiver. Visually, this effect appears to be more pronounced in the blue-green compared to the yellow-red color range, and the extent is difficult to estimate. Calculations for the yellow-red regime are giving spectral efficiencies of 56% to 63% in relation to imaginary overlapping peaks; for the blue-green range, only poor values of 7% to 18% can be assessed. Looking at pure bandpass filter features like width of the passing band and slope, LEDs deployed as PD can be an alternative to Si PIN PDs combined with a colored glass bandpass filter, see Figure 11 and [18]. Colored glass bandpass filters are only available on the market in the blue-green band, not in the yellow-red band. A thin film filter, however, generally delivers steeper slopes and can be designed over wide ranges with respect to the center wavelength and are much narrower in the passband, but with the disadvantage of a high price. Figure 11 compares the spectral sensitivity of an LED deployed as PD with an Si PIN photodetector combined with a colored glass bandpass filter G08 (Hebo Spezialglas, Aalen, Germany). Investigations of the influence of the temperature on the photodetectors responsivity have not been carried out, since they are known to be small and would not justify a considerable effort for a supposedly insignificant result. Information on the measurement uncertainties is given in Appendix A.    Osram PD SFH2400 and Hebo G08 col. glass filter Figure 11. This diagram compares the spectral sensitivity of an LED deployed as PD with an Si PIN photodetector combined with a colored glass bandpass filter G08.

Analysis of the Temporal Domain
The approximate bandwidth (BW) resulting from rise/fall time measurements can be calculated by This equation holds for signals with approximately the same rise and fall time [25]. This condition is given for the LEDs under investigation. For an experimental determination of the rise time t rd of the detector, the rise time t rs of the source needs to be taken into account unless t rs t rd . The measurable rise time t rm is the geometric addition of t rs and t rd : In the case of an LED used as source and a fast TIA-PD module as detector, with the smallest measured value t rm of 40 ns and a t rd of 2.3 ns for the used 150 MHz TIA-PD module, t rd is negligible and t rm approaches t rs .
In Table 3, the measured values t rm of the LEDs used as emitter are shown, given the setup in Figure 3. The calculated bandwidth is in the expected range.
In Table 4, the measured values for t rm of the two LED series employed as PD are given; the corresponding setup is depicted in Figure 4. The source was matched to the spectral sensitivities, see Table 3 and Figures 7-10. The rise time of the DUTs and the regarding bandwidth are calculated.
In Table 5, the measured junction capacitance of the LEDs and the bandwidth as result of the TIA simulation are tabulated. The simulated bandwidth based on capacitance measurement and the achieved bandwidth for LED employed as PD driving a TIA in a real LED sourced test setup are showing mostly good agreement. The deviation with respect to the determined bandwidths are caused by the difficulty to tune the TIA exactly to Q = 0.7 due to the limited availability of small graded capacitors in the low picofarad range. Both series of LEDs employed as PDs are showing larger capacitances, resulting in longer rise times and lower bandwidths in the blue-green regime compared to the yellow-red regime. Information on the measurement uncertainties is given in Appendix A.

Analysis of the Spatial Domain
The measurement of the relative sensitivity as function of the angle of incidence for the LEDs employed as PD shows no significant deviation from the values given in the datasheet of the particular LED, cf. Figures 12-14.

Conclusions
We experimentally investigated the spectral, temporal, and spatial characteristics of two single-color power LED series employed as PD. The examined LED series feature unexpected good responsitivity, not just compared to low-power LEDs studied in previous publications, but also compared to a common Si PIN photodetector and its theoretical sensitivity bound. The dual-use of the same LED as emitter and detector is possible, but will be accompanied by a relatively poor efficiency due to the small spectral overlap. This overlap is slightly better in the yellow-red regime compared to the blue-green regime. The well-known "green gap" of LEDs, centered around 550 nm, is still present for LEDs used as PDs, but shifted to lower wavelengths of approximately 500 nm, now presenting a "cyan gap" of sensitivity. One main characteristic of LEDs utilized as PD is to offer an inherent optical bandpass characteristic. This feature can be an alternative to the combination of PDs with secondary filters in thin film or colored glass technology. Particularly since colored glass bandpass filters are not available in the yellow-red region, off-the-shelf PD filter combinations are rare and thin film filters are expensive. The junction capacitance of the examined LEDs employed as PDs was found to be one to two orders of magnitude higher as compared to Si PIN PD reference; accordingly, the achievable bandwidth is reduced. The yellow-red regime offers a lower capacitance, thus allowing higher speeds as their counterparts in the blue-green area. The angle of incidence behavior of the LEDs under investigation was discovered to be the same when operated as emitter and detector, and no polarization direction dependence could be observed. LEDs used as PDs can serve as low-cost solutions in visible light communication, daylight filtered, or color-selective applications.

Acknowledgments:
The authors would like to thank Rüdiger Röttgers from the Helmholtz-Zentrum Geesthacht, Germany, for guidance in the selection of optical components and for provision of the monochromator used in our experiments.

Conflicts of Interest:
The authors declare no conflict of interest.

Abbreviations
The following abbreviations are used in this manuscript:

Appendix A
The main uncertainty of measurements in the spectral domain are the responsivity values given in the data sheet of the reference photodetector, produced by the uncertainty of irradiation measurements and variations in the series. Three photodetectors were measured and their mean was used for reference purposes. The influence of the accuracy of the photocurrent measurement, light source stability, mechanical positioning as well as the wavelength uncertainty of the monochromator and spectrometer are significantly lower. As the main source of uncertainty for the measurements in the temporal domain, the configuration of the feedback capacitor and the resulting quality factor can be identified. Timing and capacity values can be determined with relatively high accuracy. In the spatial domain, the measurements are relative photocurrent measurements, which can be performed with small uncertainties. The light source drift was controlled by alternating between zero position and the individual measuring position for each value taken. The manually rotatable positioning is subject to a certain error. Table A1 is giving an overview.