1. Introduction

There is an ever-increasing need for novel wireless communications technologies due to the need to operate in infrastructure-poor and physically complex environments such as dense urban and forest environments. Optical frequencies have garnered some interest as they have been envisioned to enable communications capabilities that complement existing technologies operating at lower frequencies within the electromagnetic spectrum. Research efforts include investigation of the feasibility of wireless communications at infrared, visible light, and ultraviolet (UV) for free-space and line-of-sight (LOS) scenarios [1,2].

Recent studies have shown that a portion of the UV-C band (i.e., $200$ to 280 nm) has favorable channel characteristics for more secure LOS and non-LOS (NLOS) communications. The interest in this specific band stems from the fact that the increased atmospheric scattering at these wavelengths can be exploited to create a robust communication link between near-ground transceiver nodes [3–5]. A significant portion of the photons radiated from a UV light-emitting-diode (LED) source on the ground are scattered by the atmosphere back to the ground. Also, noise due to solar radiation becomes negligible at the ground levels [6,7]. These unique characteristics enable UV NLOS links (e.g., over large buildings) to be established for applications such as ad hoc networking in dense urban environments where conventional radio-frequency (RF) technologies may not work reliably due to multipath effects. Another important aspect is that the increased propagation pathloss at longer ranges due to atmospheric absorption means that the UV-C band has an inherent advantage in terms of improved low-probability-of-detection (LPD) characteristics compared to RF systems [8].

In the literature, the potential use of the UV solar-blind band has been studied over the last few decades. Early investigations focused on long-range ($\ge$ 5 km) LOS [9] and NLOS [10] wireless communications, using large, high-power UV flash tubes and mercury arc lamps. The work in [15] focused on experimental investigation of the potential NLOS data communication via atmospheric scattering. A review article in [16] summarizes various research efforts focused on using UV LEDs for UV wireless communication systems. More recently, the use of these UV-C wavelengths have been investigated for networking applications [5,17]. Specifically, it was reported that it is feasible to form short-range, moderate-rate NLOS communications links supporting up to $100$-m links with data rates that could support voice using a low-power LED array providing $15$-mW transmit power.

Research focused on studying the channel characteristics, including the effects of absorption and scattering caused by gaseous molecules and suspended particles in the atmosphere, have been undertaken to develop models for the atmospheric scattering of UV frequencies [11]. Other efforts focused on developing accurate UV NLOS channel models have also been pursued [12–14]. Other efforts include path loss modeling via Monte Carlo simulations as well as theoretical models to compute first- and higher-order scattering components [18,21].

In this work, we seek to produce a precise experimental UV measurement system with carefully characterized components, wherein we can ensure agreement between measured and predicted results. Such a system can be useful for channel model validation as well as for adaptive communication and networking experimentation [22]. The key contributions of the paper are:

2. Experimental system

In this section, we describe the components of the developed UV communication and channel measurement system. This includes Tx and Rx nodes, both of which consist of a gimbal structure, laser diode, stepper motors, controller circuits, and driver boards. Additionally, the Tx node consists of a source device and a thermoelectric cooler (TEC) temperature controller, whereas the Rx node includes a photon detector and photon counting circuits. Although the experimental UV testbed developed in this work can be considered as a custom-designed system, almost all the parts are selected from commercial off-the-shelf components, such as UVCLEAN265-15 (UV-C source) from QPhotonics, H10721-110 (PMT module) and C9744 (photon counting unit) from Hamamatsu. All the hardware components and connections are integrated on the gimbal structure. Fig. 1(a) illustrates the custom-built Tx gimbal system whose dimensions are 30 cm × 30 cm × 50 cm. Connected to the gimbal structure are a data acquisition (DAQ) device, Global Positioning System (GPS) disciplined reference clock, PC, and DC power supply. The DAQ device and PC establish an interface for software control and to automate data collection, and the GPS unit provides synchronization between Tx and Rx nodes. In the rest of this section, these components and the overall system are described in more detail.

 

Fig. 1. The gimbal system enables precise positioning for automated experimentation at various azimuth and elevation angles up to $0.1^\circ$ resolution. (a) Tx gimbal, (b) Rx node (Rx gimbal along with other components).

Download Full Size | PDF

2.1 Gimbal structure

The gimbals are designed to allow precise positioning and control both in azimuth $(\phi )$ and elevation $(\theta )$. For a given gimbal, either the Tx or Rx unit is integrated on the gimbal’s head. The azimuth motor shaft is connected to the middle plate that holds the upper half of the gimbal as shown in Fig. 1(a). The structure provides a reference plane that is used to compute the movement of the head unit in $\phi$ and $\theta$ through the motors. Connections from the driver boards to the power supply and DAQ control pins are accessed through the slip-ring interface, which enables the system to rotate in azimuth direction uninterruptedly. The system also has a capability of sweeping $180^\circ$ in elevation; hence, the pointing angles can span an entire sphere. The motor movement resolution (i.e., step size), can be set through the micro-stepping features of the motor driver boards. During the measurements, we used a step angle of $0.1125^\circ (\text {i.e., }$ 3200 steps/rev) for precise pointing. A Class II laser diode is attached to the gimbal head, which is used when an alignment of Tx and Rx nodes with each other is needed, such as at the beginning of LOS channel measurements.

2.2 Transmitter system

Integrated on the Tx gimbal head, as shown in Fig. 1(a), is a low-power, narrow-band, miniature, solid-state UV-C band LED module that is used as the UV photon source. The module is composed of four separately controlled LED subarrays placed in a 2-by-2 grid. The hemispherical lens sealing of the LED module shapes and collimates the radiation. When all the LED subarrays are ON, the output power is given as 10–15 mW in the datasheet. The normalized power spectral density of the LED array module is shown in Fig. 2. The peak wavelength occurs near 265 nm. We carefully characterize the radiation pattern and absolute transmit power and describe the methodology in Section 3.

 

Fig. 2. Power spectral density (PSD) of the non-monochromatic source that we use as provided by the manufacturer [23].

Download Full Size | PDF

The UV LED module dissipates heat during photon emission. Operating at higher temperatures can reduce the lifetime of these devices. Moreover, the number of photons radiated by the source is affected by such thermal effects. When the source is continuously kept ON we observe that the output power of the LED gradually drops to less than $90\%$ of its peak power in 3 min. To address this challenge, we incorporate a thermoelectric cooling circuit that monitors and keeps the temperature of the UV source at a desired setting ($ {25}\,^{\circ}\rm{C}$ in our case). The feedback for temperature control is obtained via thermistor pins of the LED. The TEC circuit adjusts the direction and amplitude of a control signal to transfer heat, i.e. positive and negative current for heating and cooling, respectively. After employing the TEC, the output power drops to $97\%$ of its peak after 60 min, when the source is kept on. We further address this issue by alternating the source between ON and OFF states every second. In this case, the output remains steady around $99\%$ of the peak output power for 4 h.

2.3 Receiver system

The key component in the Rx system is a photomultiplier tube (PMT), which is a device that converts incident photons into electric currents. Specifically, a photon enters through the device window and transfers its energy to the photo-cathode, which results in electron emission. That free electron under a high electric field gets accelerated and impinges on the first dynode (metal), which causes excitation of several electrons that can also move freely. In a similar fashion, these electrons get accelerated and impinge on the next dynode creating more free electrons inside the tube. In a typical PMT, there are usually at least $10$ dynode sections. The number of electrons are increased by nearly four times at each section yielding the gain (or multiplying factor) of around $10^6$. For a typical transit time, which is the time required for an electron emitted at the cathode to pass through the tube and reach the anode, of 10–100 ns, the PMT output current is in the range of $1$ to 10 µA, which is several orders of magnitude higher than the dark current (PMT internal noise). If the incoming photon rate is low enough, the corresponding output current appears as a discrete pulse on the time axis so that the PMT can be used in the photon counting regime. In this mode, it is possible to count the exact number of photons that hits the photo-cathode. To process the current pulses at the anode, we connect the PMT output to a photon counting unit (PCU). The PCU includes a pulse amplifier, a comparator circuit, a pulse shaper, and a pulse stretcher. The comparator eliminates the low amplitude pulses that appear at the output of the PMT due to thermal noise. The circuit outputs either a logic high or logic low signal depending on whether the amplitude at the output of the pulse amplifier exceeds a particular threshold. We set this threshold to the minimum value that yields 0 counts under complete dark conditions. The optimal value will depend on PMT and UV-C filter attributes, such as the gain, structure, type, material, out-of-band rejection, insertion loss, etc. and requires calibration of the device settings to learn. (For the devices used in this work, described earlier in this section, we used 0.8 V as this was the minimal value that yielded 0 counts under complete-dark conditions. Other devices might required a different threshold.) A voltage level that is too low will lead to dark currents generated in the PMT being counted as detected photons when photons are not present, while a level that is too high will lead to an undercount of actual photons. The cascaded pulse shaper and pulse stretcher circuits converts the signals into transistor-transistor logic pulses and increases the duration of the pulse waveform, respectively. Finally, the PCU output is connected to a counter on the DAQ through which the incoming pulses are counted. The PCU threshold is chosen such that the dark photon count rate is in the range of $0 - 20$ per second.

The spectral response of the PMT module used in this work extends from 230 nm to 700 nm. This broad range spectrum includes all the visible light band photons radiated by the Sun, and is responsible for the primary source of background noise for the system. To address this issue the PMT is cascaded with a UV-C bandpass filer as shown in Fig. 1(b). The bandpass filter has a peak wavelength at 264 nm, and provides a minimum rejection of $10^{-10}$ within 300 nm to 750 nm band. The combination of PMT and UV filter is referred as detector throughout the text. In general, field of view of the detector is determined by sensitivity of the UV filter to the angle of incident light, and the internal characteristics of the PMT, such as size, photocathode features, refractive index of the window material, etc.

2.4 Control, synchronization, and software

As displayed in Fig. 1(b), the experimental system also includes a DAQ device and a GPS module. A National Instruments PCIe-6363 X-series DAQ device is used for automated hardware control, motor movements, and data collection through LabVIEW. Specifically, the DAQ device allows high-speed data processing, such as signal modulation/demodulation, counting pulses at the output of the PCU, as well as clock recovery. The GPS module, which provides a GPS-disciplined $10$-MHz clock and a $1$-PPS (pulse-per-second) signal, is used to synchronize the Tx and Rx nodes. The $1$-PPS signal is used to initiate the measurements and synchronize PC clocks, while the $10$-MHz signal is used to phase align the DAQ’s internal clocks at both ends. High-speed timing operations (e.g., clock generations for the system, and start-stop-pause intervals for the counters) in the software are referenced to this hardware clock signal obtained from the GPS module.

3. Methodology for device characterization

One challenge for experimental systems is to ensure accuracy when compared with theoretical models. Careful calibration of devices and precise system control for power or steering is also a requirement. In particular, for UV communications and channel measurements, we also have NLOS characteristics that need to be captured. To overcome these challenges, we develop a methodology for measuring device characteristics precisely by using a software-controlled and automated UV channel measurement system. Primarily, we performed accurate measurements to obtain the radiant intensity of the source and the angular response of the detector. The proposed approach can be applied to other optical wireless communication systems that leverage the same type of device characteristics.

Here, we present a general system setup, data sampling strategy, and postprocessing of the measurements, which are common for both the source and detector characterization. The steps that are unique to each device are described in Section 4. For each characterization we perform, there are two devices involved. The first device (can be either source or detector) is the one to be characterized, and denoted as DUT, or device under test. The second device is used to complete the measurement channel, and denoted as DST, or device stationary, which refers that during the data set collection, the orientation of DST is stationary. As an example, if the DUT is considered as the source, the detector becomes the DST, or vice-versa.

3.1 Measurement setup

The main variables in the measurement setup are channel type (i.e., LOS or NLOS), range ($r$), and orientation of the DST. We choose a global reference frame so that locations of the nodes, pointing directions, and data points can be defined using a common coordinate system. For that purpose, we assume the DUT to be located at the origin, and define the $z$-axis of the coordinate system as the line that is parallel to the azimuth motor shaft with the positive $z$ direction pointing away from the gimbal base (one such instance is given in Fig. 3(a)). The $xy$-plane is chosen as the plane that includes both DUT and DST, assuming the device geometries as points. The directions of the axes are specified after the gimbal placement, and the alignment process, which are described next.

 

Fig. 3. (a) The placement of the nodes, where, in this case, the LED (at $O$) and PMT are considered as the DUT and DST, respectively. The gimbal bases are coplanar, which is achieved by tilting the gimbals until the angle measured by the digital level equals to $\arcsin (\frac {\Delta h}{r})$ on both sides, where $\Delta h$ is the difference in height from sea level. (b) Initial condition of the system before starting data sampling. At this state, the pointing directions of the DUT and DST are, by definiton, $(\theta _\text {u},\phi _\text {u})=(0^\circ,0^\circ )$ and $(\theta _\text {s},\phi _\text {s})=(0^\circ,180^\circ )$.

Download Full Size | PDF

A key consideration in the measurement setup is that the gimbal bases need to be coplanar as shown in Fig. 3(a). This is achieved by tilting the gimbals to an angle $\psi =\arcsin (\frac {\Delta h}{r})$, where $\Delta h$ is the difference in height between the devices, $r$ is the separation, and $\psi$ is the angle relative to the ground plane. A barometric sensor and a high-precision digital level are used to obtain $\Delta h$, and to adjust the bases to $\psi$, respectively. Note that the motor shafts of both gimbals become parallel, and they may not be normal to the ground.

Another consideration is the alignment of the DUT and DST such that they point toward each other. This is done by precisely adjusting the orientations using the stepper motors. Two $1$-mW laser diodes at 650 nm, one embedded in each gimbal head as shown in Fig. 1(a), are used for real-time tracking of the pointing directions at both ends. The orientation of each device is defined on a spherical surface, $\mathbb {S}^2$, using the notation $(\theta, \phi )$, where $\theta$ is the elevation angle between the $z$-axis and the $xy$-plane, and $\phi$ is the azimuth angle in the $xy$-plane and measured from the positive $x$-axis. We use $( {\theta _\text {u}}, {\phi _\text {u}})$ and $( {\theta _\text {s}}, {\phi _\text {s}})$ to refer to the orientations of the DUT and the DST, respectively. After the alignment, $(\theta _\text {u},\phi _\text {u})$ and $(\theta _\text {s},\phi _\text {s})$ will be at $(0^\circ,0^\circ )$ and $(0^\circ,180^\circ )$, respectively. It is at this step that the reference frame is explicitly defined, where the positive $x$-axis is the extension of the line segment from DUT to DST, $z$-axis is parallel to the azimuth motor shaft as mentioned before, and $y$-axis is obtained by assuming a right-handed coordinate system.

The distance $r$ is chosen based on the incoming photon rates. We assume the number of received photons has a Poisson distribution and the received photon measurements correspond to noisy observations around the true mean. The significance of this inherent noise, also known as shot noise, becomes larger as the number of detected photons decreases. The ratio of signal power to noise power increases with $\sqrt {\lambda }$, where $\lambda$ is the true mean. We set the minimum threshold of photon rate as 10,000/s, i.e., signal-to-noise ratio of 20 dB, and pick a value of $r$ so that measurements are above that threshold at all data points.

The final step in the setup is, if needed, modifying the orientation of the DST from the default alignment setting. This step can be used as a tool to tune the incoming photon rate to a desired level, which satisfies the dynamic range of the devices, and hence taking reliable measurements. The setup after the alignment, the coordinate system, and the pairs $(\theta,\phi )$ are depicted in Fig. 3(b). For the LOS channel, an opaque hardboard with an aperture is placed at both ends to reduce the number of NLOS photons that are reflected or scattered from the ground surface and nearby obstacles. The hardboard at the Rx side is also used to reduce the photon contributions originating from the sun, the background noise in the system.

3.2 Sampling strategy

We use the term observation to denote the number of photons reported by the photon counting unit (depicted in Fig. 1(b)) at a particular data point, i.e., orientation of DUT or $(\theta _\text {u},\phi _\text {u})$. An observation corresponds to the number of background noise counts when the UV source is turned off (OFF cycle), or sum of both signal and noise counts when the UV source is turned on (ON cycle). To retrieve the signal counts from noisy observations (a signal measurement), we alternate the UV source between ON and OFF cycles every second, and subtract the two consecutive observations. The receiver samples the accumulated photon counts at the end of each cycle since nodes are synchronized through the GPS module. Here, we assume the background noise counts obtained at the end of both cycles (an ON cycle followed by an OFF cycle) have the same distribution. It is also possible to obtain multiple observations at the same data point. In that case, the signal count is estimated by subtracting the mean of the observations during OFF cycles from the mean of the observations during ON cycles. This estimate is associated with the term signal measurement or measurement at the data point. We also use the term data-set as a set of measurements that are common in the measurement setup (including alignment) and sampling strategy, where the latter considers the locations of the data points on $\mathbb {S}^2$.

A key part of the characterization procedure is choosing the data points and the method of how they are sampled. We consider the data points to be uniformly distributed, and form a geodesic grid on an imaginary hemispherical surface, which is embedded in the $\mathbb {R}^3$ reference frame defined in Section 3.1. One such grid is illustrated in Fig. 4 and marked with blue circles as an example. We assume there is no back-radiation from the source, hence only the forward half of the spherical surface (with respect to the DUT) is considered. The sphere is assumed to be at the origin of the coordinate system and the radius is equal to $r$, just to indicate the desired separation between the DUT and DST. The intuitive approach is placing the DST at the data points, adjusting its orientation so that it is always pointed directly to the DUT, and then taking measurements, while the DUT is kept at $( {\theta _\text {u}}, {\phi _\text {u}})=(0^\circ,0^\circ )$. However, this approach requires precise movement of the DST in 3D, as well as realignment for each data point, which is not practical with typical gimbal systems including the ones used in our system.

 

Fig. 4. Data points (a set of $(\theta _\text {p},\phi _\text {p})$ pairs) that are used for the measurements of UV LED radiation pattern and detector’s FOV characteristics. The points are evenly spaced on a spherical surface, and equidistant from the DUT, which is assumed to be placed at the center of the sphere. The two data sets, generated using a geodesic grid function [24] and illustrated in blue and magenta, represent a coarse (low spatial resolution) and dense (high spatial resolution) sampling data points.

Download Full Size | PDF

In this work, we pursue an alternative approach, which makes the sampling process automated, faster, and easier. Specifically, the position of the DST and its pointing direction are fixed after the measurement setup. Instead, only the pointing direction of the DUT is adjusted at the beginning of a measurement based on the data point location. The idea is to rotate the orientation of the DUT (as well as the spherical surface shown in Fig. 4, and the associated coordinate system) in a way that the position of the DST corresponds to one of the data points on the sphere with respect to the rotated coordinate system. Note that this method requires calculating angles of rotation, i.e., the new value of $( {\theta _\text {u}}, {\phi _\text {u}})$ for each data point. Let the set of pairs $\{(\theta _\text {p},\phi _\text {p})\}$ be the desired data locations on the grid, where the pair $(\theta _\text {p},\phi _\text {p})$ corresponds to elevation and azimuth angles referenced to the coordinate system. We derive the relation between the pairs $(\theta _\text {p},\phi _\text {p})$ and $(\theta _\text {u},\phi _\text {u})$ (all angles are in radians) as:

Figure 4 illustrates two different data sets as an example, where the lower-resolution set (blue circles) includes 1313 data points and covers the entire hemisphere. The higher-resolution data set (magenta dots) covers an area $10$ times smaller, but includes $2383$ data points. High-resolution data sets are useful to capture rapid changes in power due to small perturbations. Also, high spatial resolution leads to a better accuracy of the location and level of the maximum power delivered by the source. Note that the accuracy of the measurements degrades at the data points near the bounds in Fig. 4 because of the gimbal structure, particularly, the presence of the middle plate and metal slab where the elevation motor is attached. The situation becomes more apparent when $\theta _\text {u}$, or $\phi _\text {u}$ becomes $-90^\circ$. The illuminated middle plate increases the number of scattered photons, while the slab partially blocks the LOS path between the source and detector.

3.3 Data processing and combining multiple data sets

The characterization process of each device usually requires multiple data sets that are aggregated in postprocessing. The main reason for collecting multiple data sets for the same device is the lack of global settings (e.g., range) and inputs (e.g., data locations) that can be applied to the system suitably and satisfy all the device-related limitations. One such limitation is due to the detector’s dynamic range. The issue is that the received power levels must lie within the detector’s dynamic range to avoid device saturation. During the measurement setup, the separation $r$, and the orientation of DST $(\theta _\text {s},\phi _\text {s})$ are two parameters that can be adjusted to ensure that the detector remains in the linear regime. However, in most cases one particular setting of $r$ or $(\theta _\text {s},\phi _\text {s})$ is inadequate to guarantee linear operation for all data points. Hence, another data set with different setup parameters of either $r$ or $(\theta _\text {s},\phi _\text {s})$ is needed.

We also consider varying spatial resolution between the data sets. We start with a coarse grid that covers the entire hemisphere, that is, $-90^\circ \leq \theta _\text {p},\phi _\text {p} \leq 90^\circ$, and then obtain another data set by decreasing the size of the grid and focusing on the center region and put limits on ($\theta _\text {p},\phi _\text {p}$). There are several reasons the center region is sampled densely. First, it is the region where the incoming photon rate reaches its maximum. Second, more change is expected near the center because of the side lobes that reach the detector through a LOS path. The impact of the side lobes can be significant considering the UV LED comprises four subarrays, each placed at one of the quadrants of a $2$-by-$2$ grid structure. Similarly, the number of incoming photons that follow any NLOS path are expected to be more around the center because of the surrounding structures (e.g., reflection from buildings or the ground) due to the contribution of all subarrays.

Finally, we repeat the measurements by changing the channel type from LOS to NLOS by keeping all the other parameters unchanged. The latter reveals the contribution of first- and higher-order scattered photons during LOS channel measurements, which can be significant, especially if there are multipaths due to the surroundings. (Such a comparison can be found in the authors’ prior work in [5]. For conciseness and not to distract from the goals of this paper we only present a result for the single-scattering NLOS scenario in Section 5.) The aperture of the UV-blocking hardboard at the Tx node is blocked to remove the LOS component. Recall that the apertures on both sides are very small. And although blocking the aperture on the Tx end eliminates the LOS path, it also removes some of the forward radiation that go through the aperture and reaches the receiver end via NLOS paths. However, we assume the error due to this issue is very small and beyond the accuracy we target.

Data postprocessing is needed to combine the multiple data sets associated to different measurement settings. For data sets that only differ in channel type, we simply subtract the contributions obtained through the NLOS channel from the measurements obtained through the LOS channel (since the LOS channel naturally includes NLOS compoonents as well). For data sets that differ in spatial resolution, range, or, ($\theta _\text {s}, \phi _\text {s}$), the link is established through common data points. Specifically, we compare measured power levels at the same locations and obtain the photon ratios. We then compute a normalization factor by taking the mean of all such ratios and apply that to all the other data points in the data set.

4. Experimental results: device characterization

We focus on the particular UV LED and PMT pair, which is mentioned in Section 2, and perform outdoor measurements to obtain the device characteristics of the UV LED source and detector. The results are presented at the end of the section.

4.1 UV LED characterization

This section describes the methodology to accurately measure UV LED radiation characteristics using a low-cost, portable, and compact system. For the source device, we focus on two key attributes: 1) the total radiated power, and 2) the distribution of that power in a $3$D free space, which we refer to as the angular radiation pattern. Next, we describe the details of our approach in obtaining these two attributes.

Finally, we describe the details of our approach in obtaining these two attributes.

4.1.1 Angular radiation pattern

We follow the methodology mentioned in Section 3, and empirically acquire the distribution of the photon counts as a function of the transmitter’s azimuth and elevation angles through the LOS link between the two gimbals. Here, we consider the DUT as the UV LED source and the DST as the PMT module. There are a total of six data sets obtained for angular radiation pattern. The details of the measurement setup for these data sets including spatial resolution, range of angles of the DUT, orientation of the DST, and the channel type are given in Table 1. The spatial or sampling resolution indicates the arc length between the neighboring data points. For the coarse data sets, we sample the entire region (i.e. $-90^\circ \leq \theta _\text {p},\phi _\text {p} \leq 90^\circ$) with a spatial resolution set to $5^\circ$ resulting in 1313 data points. The data collection procedure is described next.

Table 1. Parameters of the events used in characterization of the UV LED, $r=$30 m

View Table

We first collect a coarse resolution data set, $D_1$, in which the settings are $(\theta _\text {s},\phi _\text {s})=(0^\circ,180^\circ )$, $r=$30 m, and $5^\circ$ spatial resolution. At this range, all measurements satisfy the minimum threshold of 10,000 photons. However, we also observe that when $|\theta _\text {p}|,|\phi _\text {p}|< 10^\circ$, the number of detected photons becomes high enough, causing the PMT to saturate. Hence, we collect another data set, $D_2$, with different settings, which are $(\theta _\text {s},\phi _\text {s})=(17.5^\circ,180^\circ )$ with the same range and spatial resolution used to collect $D_1$. By doing so, we avoid PMT saturation, but now some measurements are recorded as fewer than 10,000 photons. We need to combine $D_1$ and $D_2$ to obtain the measurement data for the intended spatial resolution. Although, the settings are different, some of the data points in both data sets are valid and in common. Hence, we obtain a ratio between those valid measurements. This ratio is then used as the normalization factor that allows us to combine $D_1$ and $D_2$.

The fine-sampling data set, $D_3$, corresponds to the spatial resolution of $1^\circ$ and requires 2383 data points. The span of both $\theta _\text {p}$ and $\phi _\text {p}$ is between $-25^\circ$ and $25^\circ$. The sampling locations are shown in Fig. 4. We use the same settings for the range and $(\theta _\text {s}, \phi _\text {s})$ as used to collect $D_2$.

Finally, we collect data sets $D_4, D_5$, and $D_6$ to remove the contribution of NLOS components present in the first three data sets using the same settings in $D_1, D_2$, and $D_3$, respectively. The angular radiation pattern is plotted in Fig. 5(a). Recall that UV LED is composed of arrays of discreet LED chips placed at each quadrant. As an outcome, we see local maxima at each quadrant, where the peak powers are obtained at $(\theta _\text {p},\phi _\text {p} )=\{(8.8^\circ,-9.9^\circ ), (9.2^\circ,9.5^\circ ), (-10.6^\circ,-9.4^\circ ), (-8.9^\circ,10.8^\circ )\}$. For an elevation angle of $8.8^\circ$, the cross-sectional view of the radiation intensity is given in Fig. 5(b).

 

Fig. 5. (a) UV LED radiation pattern illustrated on $\mathbb {S}^2$. (b) The cross-sectional view of the same data for $\theta _\text {p}=8.8^\circ$. The maximum number of photons is measured at $(8.8^\circ, -9.9^\circ )$. The azimuth angles at which the peak level drops to 10% are also noted.

Download Full Size | PDF

4.1.2 Total output power

For output power calculations, a UV power meter (ILT2400) is used to measure the irradiance (defined as the radiant power per unit area) at a particular distance ($r_\text {uvm}=$1 m). PMT, UV filter, and photon counting blocks are replaced with a calibrated SED270/QT detector. As opposed to angular radiation measurements, in this case the data is collected at a single point. The orientation of the detector is set to $(\theta _\text {s},\phi _\text {s})=(0^\circ, -180^\circ )$ and $(\theta _\text {u},\phi _\text {u})$ is adjusted such that the irradiance level, $E_{e}(r_\text {uvm})$, measured by the UV meter becomes maximum. The sensitivity of the UV meter is not high enough to permit collection of data at longer ranges (i.e., $r>$ 10 m) or at any other orientation. This is also the main reason for using a photon detector instead of a UV power meter for the angular radiation pattern measurements.

The measurement obtained from the UV power meter is used to estimate the irradiance level at the distance where PMT data are collected, that is, $E_e(r_\text {pmt}) = E_e(r_\text {uvm})\times (\frac {r_\text {uvm}}{r_\text {pmt}})^2$. Consequently, the incident power at $r_\text {pmt}$ is computed as $\mathcal {P}_\text {pmt} = E_e(r_\text {pmt})\cdot A_\text {pmt}$, where $A_\text {pmt}$ is the area of PMT’s aperture. Recall that measurements are taken at $1$-s intervals while the source is alternating between ON and OFF states. Hence, the estimate of the received energy is found to be $\mathcal {E}_\text {pmt}=\mathcal {P}_\text {pmt}\times $ 1 s.

We use the spectral distribution function of the source, $S({\lambda })$, shown in Fig. 2, to compute the average energy of a photon as $\mathcal {E}_\text {avg}=\frac {h \cdot c}{\mathcal {P}_\text {tot}}\int \frac {S(\lambda )}{\lambda } d\lambda$, where $\mathcal {P}_\text {tot}\triangleq \int S({\lambda }) d\lambda$, $h$ is Planck’s constant, and $c$ is the speed of light. Then the number of incident photons is computed as $N_i = {\mathcal {E}_\text {pmt}}/{\mathcal {E}_\text {avg}}$. The ratio of the number of incident photons to the number of detected photons at the Rx gives the quantum efficiency of the bandpass filter and PMT network, $\eta = {N_d}/{N_i}$. Note that $N_d$ is the number of photons reported during PMT measurements for the data point where the orientation of the Tx gimbal is the same as with the UV meter reference measurement. Finally, the total output power is computed by taking the surface integral of the normalized angular distribution of the radiant intensity function and then scaling by $\mathcal {P}_\text {pmt}$. We compute the total power of the UV LED as 14.6 mW, which is in good agreement with the typical value provided by the vendor of 12.5 mW.

4.2 UV detector

The detector comprises a broadband PMT module and a solar-blind UV filter with a center wavelength of 265 nm. The filter response is insensitive to the incident angle. We perform similar six-set data collections as in the case of UV source characterization. The DUT is swept through the same data points as in $D_{1}-D_{6}$. In this case, the UV LED is considered as the DST, hence, its orientation is held at a constant position during a particular data set collection. The quantum efficiency ($\eta$) of the cascaded system (UV filter and PMT module) is computed as 0.01 from the power measurements. The surface plot of the detector’s angular sensitivity is depicted in Fig. 6(a). The two cross sections at $\theta =0^\circ$, and $\phi =0^\circ$ are also shown in Fig. 6(b). The plots are normalized with respect to the peak measurement. The horizontal and vertical FOV are almost identical and can be considered as $\pm 20^\circ$ around the center of the aperture. The detector’s sensitivity drops by $98\%$ beyond that range.

 

Fig. 6. (a) A 3D visualization of the receiver sensitivity (FOV) is shown. (b) Cross-sectional 2D plots when $\theta _\text {p}=0$ (against azimuth) and $\phi _\text {p}=0$ (against elevation). At $\theta _\text {p}=0$, the azimuth angles where the sensitivity drops to 2% levels of its peak are shown.

Download Full Size | PDF

5. Experimental results: system use cases

In the previous section, we characterized the Tx and Rx devices including the FOV of the UV detector as well as the radiation pattern and the total absolute power of the UV LED. In this section, we will briefly highlight experimental results obtained using the developed system to validate and demonstrate the system’s utility. Specifically, we present two different use cases where the developed system was employed for UV channel measurement and communication experiments.

The first application focuses on validation of an analytical model for a UV communications LOS and NLOS channel path loss model, which was first presented in [5]. Analytical and Monte Carlo simulation-based models have been developed where the photon count variation for the UV communications NLOS single scattering component has been shown to vary approximately as $r^{-a}$ where $r$ is the range between the Tx and Rx nodes and $a$ is $1$. This approximation is well-documented in the literature [4,19–21]. To validate the testbed, we perform an experiment to duplicate prior results using the measurement setup shown in Fig. 7(a). Here, Tx and Rx nodes are deployed to create an NLOS channel where the LOS link was blocked by placing an opaque screen on both ends of the link. The details of the setup including elevation and azimuth angles can be found in [5]. The measured photon counts as a function of range along with the least squares fit are shown in Fig. 7(b). For this particular case, the measured results show that $a$ is $1.04$. It should also be noted that, although not included in this paper, similar measurements were performed for a variety of environments, ranges, and orientations showing $a$ values that are close to the theoretical prediction. The results from these measurements indicate that the developed measurement system enables accurate UV communications channel measurements.

 

Fig. 7. (a) A diagram showing the setup to measure the single- and higher-order scattering components, (b) the measured number of photons as a function of range and the least-squares fit [5], (c) the photon rate as a function of time during the steering optimization process for an NLOS scenario where the range between Tx and Rx is 22.5 m, and (d) the photon rate as a function of time during the steering optimization process for an NLOS scenario where the range between Tx and Rx is 44 m.

Download Full Size | PDF

In the second application, we focus on determining the optimal configuration of steering angles of UV communications nodes to maximize the communication link performance. This was first presented in [22]. Specifically, we describe the experimental validation of a novel steering optimization algorithm developed based on Finite Difference Stochastic Approximation to simultaneously optimize the Tx and Rx pointing directions to maximize the channel gain, and consequently the photon rate at the Rx without any prior knowledge about the environment, locations, and relative orientations of the two nodes. The developed algorithm was implemented in LabVIEW and validated via experiments using the system presented in this paper. Two example results obtained from the experiments for NLOS scenarios are shown in Fig. 7(c) and Fig. 7(d). The results, obtained using the precise steering capability of the UV communications measurement system presented in this paper on which the steering optimization algorithm was deployed, show a photon rate increase of 167% and 161% where the range between the Tx and Rx node was 22.5 m and 44 m, respectively.

6. Conclusion

In this paper, a precise UV measurement system with carefully characterized components is presented for experimental investigation of UV wireless communication as well as channel model validations for LOS and NLOS scenarios. A key goal of this work is to develop a methodology for precise characterization of the UV source and detector modules ensuring a good agreement between measurement and predicted results. In this regard, a methodology is proposed using an adaptive sampling strategy at multiple ranges to accurately measure the LED pattern and detector’s FOV. The absolute transmit power of a specific UV LED is also measured and the results are in good agreement with the data sheet. We then experimentally demonstrate the utility of the developed system for different applications including channel model validation as well as for adaptive communication and networking experimentation. For the channel modeling use case, the pathloss measurement results acquired with the developed system were in close agreement with predicted pathloss results.