3.1 Coupling efficiency between LEDs and fibers

Figure 5Definitions of the azimuthal angle Θ, polar angle ϕ and differential solid angle dΩ for the light emission of a small LED segment.

Download

To provide as much light as possible to the sensor node, an efficient coupling into the fiber is required. An LED has generally a Lambertian radiation pattern of its radiant intensity I (power per unit solid angle) as given by

with the azimuthal angle Θ as defined in Fig. 5. The maximum radiant intensity I₀ is determined by the total emitted power P₀. For efficient fiber coupling an optical fiber with a high acceptance angle Θ_max and a large core diameter is required. Θ_max is the maximum angle subtended to the fiber axis of light coupled into the fiber to be guided. Above this angle there is no longer a total reflection inside the fiber anymore, resulting in very high propagation losses. The acceptance angle of a fiber is usually characterized by its numerical aperture NA from basic fiber principles as defined in Eq. (2):

where n_co and n_cl are the refractive index of the fiber core and cladding, respectively. n_a is the refractive index of the surrounding air which is readily neglected in all further calculations by using the common approximation n_a≈1.

There are different types of optical fibers, like single-mode silica glass fibers (SMFs), multi-mode glass fibers (MMFs), hybrid polymer/glass fibers (plastic-clad silica, PCS) and multi-mode polymer optical fibers (POFs; Ziemann et al., 2009). The polymer optical fiber used in the prototype system has a large outer diameter of 1 mm, a core diameter of 980 µm and a high NA of 0.5 (referred to as a 1 mm standard POF). These properties outweigh the disadvantages of this fiber type, like a higher attenuation compared to silica multi-mode and single-mode fibers, if typical sensor link lengths of up to 10 m are taken into account. Due to the low bandwidth of the LEDs themselves (see Sect. 3.5), the limited bandwidth of the POF does not play a significant role for this application as well.

Figure 6Graphic representation of the coupling efficiency regarding the acceptance angle: (a) blue line: Lambertian radiation pattern of the LED I(Θ); (b) green line: normalized effective radiant intensity ${\stackrel{\mathrm{̃}}{I}}_{\mathrm{eff}}$ (Θ) of the LED; (c) hatched area: fiber-coupled power P_F for a POF with NA = 0.5 limited by the corresponding acceptance angle (red line); (d) hatched area: fiber-coupled power for a silica multi-mode fiber (MMF) with NA = 0.22 (red line).

Download

Figure 6 illustrates the impact of the acceptance angle on the angular coupling efficiency η_angle which is defined as the ratio of the fiber-coupled power P_F to the emitted power P₀ of a small LED segment. To calculate P_F the radiant intensity pattern given in Eq. (1) has to be integrated over a solid angle Ω_max of a cone with the acceptance angle of the fiber Θ_max as apex angle:

where Θ is the azimuthal angle and ϕ is the polar angle as shown in Fig. 5. The expression I₀ cosΘ sinΘ represents a polar angle integrated or effective radiant intensity ${\stackrel{\mathrm{̃}}{I}}_{\mathrm{eff}}\left(\mathrm{\Theta }\right)$ shown in Fig. 6b normalized to its maximum, there. From this quantity only those parts within the acceptance angle Θ_max contribute to the fiber-coupled power P_F. For a 1 mm standard POF with a typical NA =0.5 or Θ_max= 30^∘, and a representative silica MMF with NA =0.22 or Θ_max= 12.7^∘, the fiber-coupled power P_F is represented by the hatched areas in Fig. 6c and d. The integration in Eq. (3) can be solved analytically, yielding

Finally, the angular coupling efficiency η_angle is now simply given by

resulting in η_angle= 25 % for the POF and η_angle= 4.8 % for the silica MMF.

Additionally, the total coupling efficiency depends on the areal overlap of the LED with the fiber core. Figure 7 illustrates this for an LED with a chip size of 0.92 × 0.92 mm² as used in this work and a 1 mm standard POF with a core diameter of 980 µm, as well as a silica MMF with a core diameter of 200 µm for comparable structural flexibility. Assuming close-distance butt coupling, this yields an areal coupling efficiency of η_area= 83 % for a POF and η_area= 3.7 % for the MMF, respectively. Finally, the theoretical maximum coupling efficiency is given by the product of angular and areal efficiency. With η_total= 21%, the maximum coupling efficiency to a POF is 2 magnitudes larger as compared to a silica MMF with η_total= 0.18 %. Interestingly, these maximum values given by the LED size and the fiber parameters can not be improved further by any optical system because of the preservation of the etendue of a light beam, which is the product of its cross section and its solid angle of divergence or convergence. Clearly, the use of a POF gives significantly more transmitted light power from LED sources for fiber lengths of up to at least a few tens of meters, before the advantage of less attenuation prefers silica fibers. The break-even length l up to which the POF has its benefits compared to silica fibers depends on the spectral attenuation coefficient α of the fibers and, thus, on the wavelength of the LEDs as shown in Eq. (6):

For the measured LEDs these break-even lengths between a representative silica multi-mode fiber (core diameter: 200 µm; NA=0.22) and a standard polymer fiber (core diameter: 980 µm; NA=0.5) are shown in Table 1.

Table 1Maximum lengths for which the POF benefits over silica fibers. Attenuation data POF: Toray PFU-CD1001-22E (α_POF); attenuation data MMF: Thorlabs FG200LEA (α_MMF).

Download Print Version | Download XLSX

Figure 7Graphic representation of the coupling efficiency of a 0.92 × 0.92 mm² LED to a fiber regarding the core diameter. Left: core ∅=980 µm. Right: core ∅= 200 µm.

Download

Table 2LED total power, fiber-coupled power into a standard 1 mm POF (NA = 0.5) and coupling efficiency of the measured LEDs (Wächter, 2017).

Download Print Version | Download XLSX

Since these calculations do not include losses due to limited fiber end-face preparation quality, attenuation of the fiber, Fresnel reflections on the fiber's end-faces, and a non-zero distance between LED and coupled fiber, an experimental verification is required. To calculate the coupling efficiency η, the optical power P_N of the LED at a typical current was measured and compared with the measured fiber-coupled optical power P_C after a reference length of 1 m of a POF. Results for five LEDs from the same series made by Cree with different emission wavelengths are shown in Table 2. The measured coupling efficiency is close to the theoretical limit of 21 % calculated above, with even higher measured values which might come from a more directed radiation pattern and from tunneling modes inside the fiber, which are still present after a short fiber length as well, even if they are non-guided rays (Snyder and Love, 2000). The coupling efficiency is almost the same for all LEDs, but due to the higher optical power of the royal blue LED it provides up to 4 times higher fiber-coupled power compared to the other LEDs.

3.2 Mutual responsivity

Figure 8Measurement setup for the determination of the spectral responsivity of the LEDs.

Download

Another important parameter for powering the sensor nodes is the sensitivity of the LEDs to their own emitted light spectrum when operated as photo detectors. This so-called mutual responsivity of the LEDs gives an indication of the maximum current available for any connected sensor electronics. To determine the mutual responsivity, the spectrally resolved radiant power Φ(λ) of the LED has to be measured at first, using a broadband spectroradiometer. This is compared with the spectral responsivity R(λ) of the LED as photo detector. This is measured by using a halogen lamp and a monochromator for narrow-band illumination with adjustable wavelengths. Then the light is guided to the receiving LED using a POF. To enhance the signal-to-noise ratio, an optical chopper was used to modulate the light and a lock-in amplifier is used to detect the LED photo current as shown in Fig. 8. The photo current for different wavelengths can be compared with a calibrated detector in the same setup as a reference, yielding the spectral responsivity of the LED. The results for the spectral responsivity R(λ) and the spectrally resolved radiant power Φ(λ) for the different LEDs are shown in Fig. 9. These two traces give a range of mutual absorption of the LED as depicted by the hatched areas, where photons generated from one LED are converted to charge carriers and, thus, a photo current in another LED of the same type.

Figure 9Spectral responsivity and spectrally resolved radiant power of the different LEDs compared to a quantum efficiency of 60, 80 and 100 %. The hatched area gives a visual indication of the range of mutual absorption calculated by Eq. (7). (a) Royal blue LED; (b) blue LED; (c) green LED; (d) red-orange LED; (e) red LED.

Download

This measurements indicates two important conclusions. First, the peak responsivity of an LED, designed only for light emission, can be surprisingly high. For the red LED, the maximum value for the responsivity is about 0.36 A W⁻¹ at 580 nm. This corresponds to a quantum efficiency (QE) of 77 %. This is not much worse than the best commercially available dedicated converter components (HAMAMATSU PHOTONICS K. K., 2012). Second, a reasonable amount of one LED's emission can be converted back into electrical energy using another LED of the same type. The hatched ares in Fig. 9 give a visual comparison for the different LEDs, and calculated numbers are shown in Table 3. As a figure of merit we calculate the mutual responsivity R_S defined as the ratio between this mutual absorption and the integrated radiant power as described in Eq. (7):

where R(λ) is the spectral responsivity of the LED as photodetector and Φ(λ) is the spectrally resolved radiant power of the LED as a transmitter.

The results for the mutual responsivity of the different LEDs are shown in Table 3. The mutual responsivity is highest for the red and red-orange LED, being larger by an order of magnitude compared to the green-blue types. In principle, this was to be expected, as for the same quantum efficiency and the same spectral overlap more current is generated at longer wavelengths with lower photon energies and, thus, more photons at a given optical power to produce electron/hole pairs. As shown later in Sect. 3.4 this advantage of red or orange LEDs is not strictly valid if the maximum achievable electrical power is discussed due to the higher voltage generated and the higher optical power available from blue LEDs.

Table 3Mutual responsivity R_S of the LEDs under test.

Download Print Version | Download XLSX

3.3 Short circuit current and open-circuit voltage

For the electrical design of the electronics of the sensor nodes as well as for the calculation of their maximum possible power consumption, the LED's maximum short circuit current I_SC and open-circuit voltage U_OC give a good indication of the maximum transmitted electrical power. A rough estimation of the short circuit current I_SCe is given by the product of the mutual responsivity R_S and the fiber-coupled power P_C of the LEDs (Table 4). For a measurement two LEDs of the same type are connected by a POF of 1 m in length, aligned for maximum coupling. The sending LEDs are operated at their nominal currents. The measured currents are noticeable lower than the theoretically estimated values, as the estimation does not take into account the limited coupling efficiency at the receiving LED.

Table 4Estimated and measured short circuit current and open-circuit voltage.

Download Print Version | Download XLSX

Even though the royal blue LED provides the highest fiber-coupled power, the red LED system has a much higher short circuit current, caused by the higher mutual responsivity, mainly because of the larger number of photons per unit time and area at longer wavelengths at a given optical power. Even if it is not possible to determine the exact available electrical power by this simple measurement, it gives a good indication, since our own measurements have shown, that this power corresponds very well to the product of 90 % of the short circuit current and 90 % of the open-circuit voltage. The higher open-circuit voltage of the royal blue LED does not fully compensate for its lower generated current when discussing the maximum available power in the next section.

3.4 Maximum power point

The available electrical power provided by the LED depends on the load. Principally in a linear circuit, if the load resistance is high, the current will drop, while the voltage will not exceed the open-circuit voltage. If the load resistance is too low, the voltage will drop, while the current will not exceed the short circuit current. Deviations are possible for strongly nonlinear devices like an LED operated as a photo diode. The ideal balance between open and short circuit in terms of power is in the maximum power point (MPP). It defines the maximum available power P_MPP and is determined by measuring the current-voltage characteristics and therefore the power-voltage characteristics as well. Figure 10 shows for example the measured characteristics of the red LED when illuminated by the same LED in our POF coupled test system. By comparing the power-voltage characteristics in Fig. 11 for different LEDs, it can be seen that even if the maximum current of the royal blue LED is noticeably lower than the current of the red one, the maximum power is only slightly lower. Results for the current I_MPP and the voltage U_MPP of the maximum power point as well as the maximum power P_MPP are summarized in Table 5. With the red LEDs and 1 m of POF in our system, a maximum of 9.29 mW of electrical power can be supplied to an electrical sensor node. This is sufficient for many applications.

3.5 Electro-optical modulation bandwidth

The electro-optical modulation bandwidth of the LEDs gives an indication of the maximum achievable data rate. This is very important for our application. Even if the transmission of the sensor data requires only modest data rates, it affects the time needed for data transfer, during which the sensor node can not be optically powered and has to be operated from previously accumulated energy. For example, using 1 Mbit s⁻¹ instead of 10 kbit s⁻¹ results in roughly a hundredth of the energy that has to be stored at the sensor node. This is important because at present, the effort for remote powering the sensor unit is the main limiting factor for the fiber length and the system performance, and not the signal-to-noise ratio of the data transmission, which in principle is more critical at higher data rates.

Therefore the bandwidths of the different LEDs were measured as a transmitter at first, using LED bias currents from 10 to 200 mA. The measurement setup consists of a DC source to provide the bias current of the LED using a bias tee. The LED is connected to a fast optical to electrical converter (O/E converter) by a short piece of POF. A network analyzer with a frequency range from 5 Hz up to 3 GHz and a dynamic range of up to 130 dB is used to measure the frequency response in this setup shown in Fig. 12. Since the transmission bandwidth of the used fiber (3 dB bandwidth typically > 1 GHz for 1 m of standard POF) and of the used O/E converter (3 dB bandwidth = 500 MHz) is much higher than the expected bandwidths of the LEDs, the measured frequency responses are determined by the LEDs only. Additionally, the transmission bandwidth of a system consisting of two LEDs of the same type was measured, which is expected to represent the bandwidth of the LEDs as receiver, since the LEDs transmission bandwidth should be much higher than the receiver bandwidth. Therefore the measurement system was modified by using a second LED as a receiver instead of the O/E converter (Fig. 13). The frequency response was measured using different resistors in parallel as a load and the high-Z feature of the network analyzer with 1 MΩ input impedance. This load resistance is expected to influence strongly the bandwidth because of the capacitance of the LED p-n junction in parallel. For these system measurements, a constant DC bias current for the transmitting LEDs (I_Tx) of 20 mA was used.

Figure 10Current-voltage and power-voltage characteristics of the measured red LED.

Download

Figure 11P−V characteristics of the measured LEDs, including the MPPs (marked as squares).

Download

Table 5Maximum power, current and voltage.

Download Print Version | Download XLSX

Figure 12Measurement setup for measuring the bandwidth of the LEDs as a transmitter.

Download

Figure 13Measurement setup for measuring the bandwidth of the LEDs as a receiver.

Download

The results for three representative LED types (royal blue, green, red) are shown in Fig. 14 with 3 dB bandwidths listed in Table 6. All traces have been normalized to data at the lowest measured frequency (25 kHz for the Tx bandwidth and 500 Hz for the Rx bandwidth). These frequencies are limited by the network analyzer for the Rx bandwidth measurement and by the bias tee for the TX bandwidth measurement. However, the low-pass character of the O/E converter used for only the TX measurements somewhat limits the direct comparison of TX and system bandwidths without compromising the general conclusions. As can be seen, the bandwidths of these LEDs as transmitters alone should be sufficient for data rates of at least 1 Mbit s⁻¹, even at low bias currents of 5 mA. The system bandwidth on the other hand depends strongly on the impedance of the receiver circuit. The LED operated as a receiver is typically the limiting element, since the system bandwidth LED to LED is typically lower than the transmitter bandwidth of one LED alone. This is true especially for the green LED showing a very low bandwidth when operated as a receiver, indicating that internal effects like carrier diffusion might influence the bandwidth additionally to the junction capacitance. The fact that the frequency response of the green LED consists of two low-pass functions supports this assumption. With low-impedance receiver circuits like a transimpedance amplifier, a data rate of up to a few Mbit s⁻¹ should be achievable with the royal blue as well as with red LED systems. Furthermore, the bandwidth could be enhanced by using reverse bias voltage. This has not been tested so far, since the LEDs are not specified for reverse voltage.

Data rates of 1 Mbit s⁻¹ have been shown in preliminary experiments, although the realized demonstrator in the next section requires only much lower data rates. For example, transmission experiments with the two red LEDs as TX as well as RX and a standard 1 mm POF have been made. Therefore a 1 Mbit s⁻¹ pseudorandom binary sequence (PRBS) signal was transmitted and the eye pattern was recorded using a digital oscilloscope with 50Ω input impedance and 500 MHz bandwidth directly connected to the receiving LED (see Fig. 15). As can be seen, there is still potential for improvement in reach and bit rate. It is worth noting that these LEDs have not been designed as fast transmitters and even less as fast photo receivers, but to be cheap and efficient light sources in the first place.

Figure 143 dB bandwidth of different LEDs as transmitter (Tx bandwidth) for different bias currents and of complete systems consisting of two LEDs of the same type (system bandwidth) for different load resistances: (a) Tx bandwidth – royal blue LED; (b) Tx bandwidth – green LED; (c) Tx bandwidth – red LED; (d) system bandwidth – royal blue LED; (e) system bandwidth – green LED; (f) system bandwidth – red LED.

Download

Table 63 dB bandwidths of the different LEDs as transmitter and receiver.

Download Print Version | Download XLSX

Figure 15Eye pattern for 1 Mbit s⁻¹ PRBS data transmission with the two red LEDs as TX and RX over 1 m standard POF.

Download

4.1 Sensor network principle and topology

Figure 16System concept using TDD for energy and data transmission.

Download

Using the results for the different LEDs from Sect. 3, the sensor network demonstrator was designed (Schuster, 2016; Schuster et al., 2016). The realized sensor network uses polymer optical fibers for power and data transmission and commercially available high-power LEDs as light transmitters as well as receivers. It consists of a central board and four sensor nodes in a star topology as shown in Fig. 16 using red LEDs (Cree XQ-E HI red), since these provide the highest transmitted electrical power and bandwidth. As each sensor node is connected to the central board via a single fiber only, the transmission of energy and data is implemented by a time division duplex (TDD) procedure. Therefore, energy storage at the sensor nodes is necessary to provide power during the time a sensor node sends back the sensor data, even if the required time for the data transmission is much shorter than the available time for powering. The timing diagram for TDD is shown in Fig. 17. At first the sensor nodes receive energy to charge themselves (time slot t₁). Then the nodes perform measurements and send back the data (time slot t₂). Afterwards, the central board powers the sensor nodes again (time slot t₃). For the demonstrator a cycle time of 1 s has been chosen, although a faster readout is possible.

Figure 17Timing diagram for the power and data transmission with a period of 1 s.

Download

4.2 Central board

The central board powers the connected sensor nodes and receives the measurement data from the sensors. A block diagram is shown in Fig. 18. The key element is a microcontroller (MCU) “Kinetis MKL05” of NXP based on a ARM Cortex-M0+ architecture. To power the sensors the MCU controls the LED via a current driver using pulse-width modulation (PWM). Reception of the sensor data is realized by addressing the sensors one by one, stopping the supply of optical power for a certain time which triggers the microcontroller on the sensor boards to perform a measurement and send back data. During this time, the powering LEDs are used as a receiver for the sensor data. Serial data transmission is realized as a universal asynchronous receiver/transmitter (UART) at a rate of 9600 baud, sufficient for sensor data read-out here. By using an 8-fold switch (MUX), the currently receiving LED is toggled to a single operational amplifier, which in turn is connected to the UART interface of the microcontroller. This additional multiplexing is necessary since there is only one UART communication interface available on the selected MCU. Finally, the data are sent to a connected PC using a UART/USB converter. Figure 19 shows the realized PCB for the hardware of the central board.

Figure 18Block diagram of the sensor central board.

Download

Figure 19Central board hardware setup (Schuster, 2016).

Download

4.3 Sensor nodes

Four different sensor nodes have been built for the demonstrator. Each node uses a microcontroller (MCU) to read out the data from different electronic sensors connected to the sensor nodes via an electronic I²C interface and to send the data back optically to the central board. The block diagram of the sensor node is shown in Fig. 20.

Figure 20Block diagram of the sensor node.

Download

To perform a data transmission, the sensor board must have been charged previously by the received optical power from the central board, previously. An energy harvester chip (“bq25570” from Texas Instruments) is used to charge a 22 mF supercapacitor with the photocurrent generated by the LED by an internal boost charger and a maximum charged voltage of 5 V. Furthermore, the same energy harvester chip provides the supply voltage of 3.3 V for all consumers on the board with an integrated and adjustable step-down converter (Fig. 21).

Figure 21Block diagram of the energy harvester consisting of the energy harvester chip, the LED and the capacitor.

Download

Figure 22Sensor board hardware setup (TMP75): the sensor attachment is a prototype board containing the different sensors and is connected to the sensor node using the contacts around the MCU.

Download

The microcontroller is the same as on the central board (MKL05). The reasons for using this MCU are its available different low-power modes. As soon as the harvester supplies the microcontroller with the required 3.3 V, it boots and then switches into a stop mode (VLPS). For this mode, the current consumption of the controller is only at 2.25 µA, enabling further charging of the sensor node. When the power LED on the central board is shut off, the microcontroller recognizes an interrupt and switches to another mode to perform the measurements and send the data to the central board. In this mode the current consumption of the MCU is still below 243 µA (Freescale, 2014). Since the LED on a sensor node is alternatively a power receiver as well as a data transmitter, a field-effect transistor (FET) controlled by the microcontroller is used to switch the LED between these two modes. Any electronic sensor with a standard I²C interface is suitable for connection with these sensor nodes. The only limitation is the total power consumption required for the sensor itself, the microcontroller and the data transmission. It must be lower than the maximum power provided by the energy harvester for the time of sensing and data transmission. The layout for the sensor nodes is the same for all sensor types. Only a small adaptor board has to be changed, depending on the sensor used. As an example the circuit board for the temperature sensor is shown in Fig. 22. For this demonstrator the following four different sensors were used.

• A temperature sensor (TMP75)

  Range: −40–125 ^∘C

  Resolution: 0.125 ^∘C

• An atmospheric pressure sensor (MS5637)

  Range: 10–2000 mbar

  Resolution: < 0.11 mbar

• A relative humidity sensor (SHT21)

  Range: 0–100 %RH

  Resolution: 0.04 %RH

• A photoelectric sensor (TCST2103)

  Range: open–closed

The precision and accuracy of the used sensors are defined by the manufacturer specifications and are not discussed here. These examples demonstrate the versatility of our approach, which enables almost any electronic sensor to operate electrically isolated with an economic solution.

4.4 Demonstrator parameters

Finally, the operational parameters of the realized demonstrator are discussed. The time until the sensor node is fully charged and ready for measurements after turn-on of the system depends on the fiber length which connects the sensor nodes to the central board, and the residual charge in the capacitor. The capacitor keeps a residual charge above 2 V for several days, enabling a faster wake-up after this time compared to a completely discharged sensor node. For the demonstrator described above, the following parameters are achieved.

• Measurement cycle time per sensor: 1 s

• Maximum fiber length: 10 m of a standard 1 mm PMMA POF

• Time until sensor is ready after initial turn-on (1 m fiber): 131 s (cold-start operation), 17 s (capacitor charge: 2 V)

• Up to 120 measurements possible even without optical powering of the sensor nodes (capacitor charge: 5 V)

This illustrates just a first realization with much potential for more frequent sensor read-out or faster data rates, and fewer start-up times by using ultra-low-power electronics.