1. Introduction

Optical wireless communication (OWC) has emerged as an important means of data communication to mitigate the expected increase in radio frequency (RF) spectrum-crunching in the fifth-generation networks and beyond. In particular, solar-blind OWC offers an advantageous means of extending the bandwidth range for the current OWC effort focusing on visible-light communication and near-infrared free-space communication. Furthermore, the sunlight spectrum in the ultraviolet-C (UVC) region (200 - 280 nm) is completely absorbed by the molecules in the Earth’s stratosphere. This allows one to use UVC light for outdoor OWC without being susceptible to solar background noise. Another advantage of using UVC light is its relatively high scattering, which is desirable in both diffuse-line-of-sight (diffuse-LOS) and non-line-of-sight (NLOS) configurations for relieving pointing, acquisition, and tracking requirements and for extended aerial coverage.

UV communication studies date back to as far as 1926 [1]. The early development of UV communication systems was summarized in 1964 [1]. However, advances in UVC light-emitting diodes (LEDs) have only been achieved recently. This opens the door for broad-bandwidth and high-speed communication [2,3], as well as safe UVC links based on optimally low-power light sources to minimize health risks [4]. Fast communication links could not be achieved previously when xenon flashtubes and mercury lamps were the only available UV light sources [5,6]. For example, in an early UVC link demonstration based on a mercury lamp emitting at a wavelength of 254 nm over a 18-m NLOS link, only a low data rate of 1.2 kbit/s was achieved [7].

For high data rates, small foot-print diode-based light emitters, such as lasers and LEDs, are required. Similarly, fast UVC detectors are also crucial in achieving high-speed communication, such as by using a silicon avalanche photodetector (APD) [8]. However, a multitude of hurdles related to device configuration, device technology, and transmission methodologies could impact the achievable data rates in UVC communication links. For example, a 0.2-kbit/s outdoor NLOS link of 6 m was reported based on a 275-nm LED array [9], and a 2-kbit/s line-of-sight (LOS) link of ∼40 m was also demonstrated [10]. The performance of NLOS links was shown to improve by employing a moving average filter in a 15-m link based on an array of 265-nm LEDs [11]. In another study using frequency-shift keying, NLOS and LOS links with a data rate of 1.2 kbit/s were established over 80 and 300 m, respectively [12]. Reliable NLOS links were also shown to have enhanced performance based on spatial diversity [13–15]. A 1${\times} $2 single-input and multiple-output (SIMO) UVC link using a 265-nm LED array as a transmitter was shown to extend the possible transmission distance at a target bit error rate (BER) [16].

While the aforementioned studies achieved data rates in the kilo-bit-per-second range, a 1-Mbit/s NLOS was successfully established over 1 km using receiver diversity [17]. More recently, high-speed LOS UVC links have been demonstrated. A 1.6-Gbit/s (non-diffused) LOS link was evaluated under direct sunlight over a distance of 1.5 m [18]. The operation of a similar system was verified in an outdoor setting in [19]. Moreover, micro-LEDs with a broad modulation bandwidth of 438 MHz were recently measured and used to send 1-Gbit/s signals over 0.3 m to an APD [20].

Leveraging on the high scattering characteristic of UV light, a proof-of-concept diffuse-LOS link with a 71-Mbit/s maximum data rate was demonstrated based on an ultraviolet-B (UVB) LED and orthogonal frequency-division multiplexing (OFDM), over an 8-cm distance and ${\pm} $ 12$^\circ $ of change in the angle between the transmitter and the receiver [21].

Herein, we demonstrate a high-speed, 1${\times} $2 SIMO UVC diffuse-LOS link over a transmission distance of 5 m by using probabilistically shaped discrete multitone (DMT) modulation. To maximize the data rate without increasing the emitted power, and thus avoiding possible health risks, the DMT modulation with probabilistic constellation shaping was implemented to maximize the use of the channel capacity. In the approach, probabilistic shaping (PS) generates constellations for each subcarrier used individually according to the signal-to-noise ratio (SNR) and following a Maxwell-Boltzmann (MB) distribution [22]. Using this technique, the performance of the established communication link can approach the Shannon limit of the channel, providing even higher data rates when compared to other techniques without resorting to increasing the transmitted power. Moreover, PS can offer a continuous adjustment of the source entropy for flexible rate adaption to different communication conditions without increasing the complexity of the system and its implementation [23], which can potentially improve the performance of links in channels affected by atmospheric turbulence. This can also help for links in which the reception angle is not fixed by adapting the data rate. Moreover, the high available data rates using PS with diversity reception offers an opportunity to further increase the reception angle given a predetermined threshold data rate. The SIMO system is shown to have a reception angle of view of ${\pm} $ 9$^\circ $, which is significantly wider than that of a SISO link. In addition, the SIMO link has a peak achievable information rate (AIR) of 1.09 Gbit/s and a minimum AIR of 0.24 Gbit/s.

For practical assessment of the single-input and single-output (SISO) and SIMO UVC links, the effect of weather conditions on the received signals was studied in terms of the BER and the scintillation index by emulating foggy conditions. Our demonstration affirmed the full potential of high-speed, UVC OWC links that obviate the strict alignment requirement between the transmitter and the receiver, while providing immunity against weather conditions and solar background noise.

2. Experimental setup

As shown in Fig. 1(a), the transmitter of the UVC link is a 279-nm LED (IRTronix, UV1008M), with a measured -3-dB modulation bandwidth of 170 MHz. A 1-inch (1 inch = 2.54 cm) fused silica lens was mounted to control the beam divergence towards the receivers. At the receiver end, two variable-gain APDs (Thorlabs APD430A2) separated by ∼0.5 m are used to establish the 1${\times} $2 SIMO link. The gain values of the APDs are adjusted to maximize the SNR without saturating the detectors. The two detectors faced the transmitter in the initial position, and the transmitter was mounted on a rotating stage. The UVC light is collected using a 2-inch lens mounted in front of each APD. The output signal of each APD passes through a 25-dB amplifier (Mini-Circuits ZHL-6A-S+) and a variable attenuator for amplitude adjustment, and is further recorded using a mixed-signal oscilloscope (Agilent MSO9254A) for offline processing.

 

Fig. 1. (a) A picture of the experimental setup. The inset shows the UVC LED. (b) The block diagram of the signal generation and offline processing. (c) The L–I–V plot of the UVC LED. (d) The spectrum of the light of the UVC LED showing a peak intensity value at 279 nm with a full width at half maximum (FWHM) of 11 nm. (CCDM: constant composition distribution matching)

Download Full Size | PDF

The PS DMT signal (1.2 V_p-p) is sent through an arbitrary waveform generator (AWG, Tektronix, AWG710b) to an amplifier and an attenuator of the same models as those on the receiver side. The alternating current (AC) signal is combined with the 300-mA direct current (DC) bias using a bias tee (Mini-Circuits ZFBT-4R2GW-FT+) which is connected to the LED. The relative angle between the transmitter and the receiver, $\theta $, measured as the angle of the transmitter relative to the line connecting it to the midpoint of the two receivers, is changed by rotating the transmitter. Figures 1(a) and 1(b) show a picture of the experimental setup and a schematic diagram of the signal modulation and demodulation steps, respectively. The inset in Fig. 1(a) is a picture of the UVC LED. The light-output–current–voltage (L–I–V) plot of the LED and its spectrum are shown in Figs. 1(c) and 1(d), respectively. The spectrum exhibits a peak at 279 nm with a full width at half maximum (FWHM) of 11 nm.

To apply bit loading with the most suitable allocation of bits, the SNR value for each subcarrier was measured using DMT [24]. As shown in Fig. 1(b), a forward error correction (FEC) encoder, with a constant composition distribution matching (CCDM) [25], was applied to generate the in-phase (I) and quadrature (Q) paths based on PS-pulse-amplitude modulation-16 (PS-PAM-16) with the optimal MB distribution. The two sets are used to generate a PS-quadrature-amplitude modulation-256 (PS-QAM-256) signal. The AIR can then be calculated from the generalized mutual information (GMI) values with the binary rate-0.858 FEC code as shown in Eq. (3) in [26]. An important consideration for PS-based modulation is the complexity. Subcarriers with low SNR values in PS modulation have high shaping depth (defined as the ratio of the probability of the innermost constellation to that of the outermost constellation), which significantly increases the length of the symbol stream for entropy values below 3 bit/symbol (corresponding to a SNR of 5.1 dB). This increase in the stream length increases the complexity [25]. All subcarriers whose SNR values are below this threshold are allocated a single bit (2-QAM) to reduce the complexity of the system.

The signals from the two detectors, RX₁ and RX₂, denoted as ${s_1}(t )$ and ${s_2}(t )$, respectively, where t is the time, are combined by giving each signal a weight, ${w_i}$, $i \in \{{1,2} \}$, equal to the ratio of its amplitude to the sum of the amplitudes of the two signals. We can write:

3. Results and discussion

3.1 Angle of view

The maximum AIR value at each angle $\theta $ is measured by individually adjusting the sampling rates of both the AWG and the oscilloscope and by adjusting the MB distribution to maximize the AIR. The AIR values versus $\theta $ are presented in Fig. 2(a) with the AIR of the SISO link for comparison. As the relative angle between the transmitter and the receiver increases, the AIR drops, but it increases near the edges. This is due to the non-uniform profile of the beam, which has high intensity around the outer ring, as shown in Fig. 2(b). At the initial position, both detectors can receive the signal and the spatial diversity gain increases the quality of the signal, resulting in higher AIRs. A maximum AIR of 1.09 Gbit/s was recorded with no rotation of the transmitter, with a minimum maintained AIR of 0.24 Gbit/s across all the tested angles. To have a fair comparison, we set two conditions for the SISO diffuse-LOS link: (1) its AIR at $\theta $=0° must be around the same value of that of the SIMO link (∼1.1 Gbit/s), and (2) the AIR cannot drop below 0.1 Gbit/s at all angles. Since a single detector is used in this case, more power is needed to achieve a similar AIR, which means that the beam divergence must be smaller to increase the received intensity. From Fig. 2(a), we can see that the first condition is satisfied, and the second condition is satisfied for angle changes between the transmitter and the receiver of ${\pm} $ 5.5$^\circ $. Any change in the angle beyond that causes the AIR to drop below the set threshold (0.1 Gbit/s). The SIMO link maintains an AIR well above the threshold along ${\pm} $ 9$^\circ $ changes in the angle.

 

Fig. 2. (a) The recorded AIR values for the SISO and SIMO links versus the relative angle between the transmitter and the receiver, $\theta $. (b) A typical beam profile of the divergent beam (not showing the true color) after propagating for 1 m.

Download Full Size | PDF

Assuming the two detectors are perfectly placed on the circumference of a circle whose center is at the position of the transmitter, TX, and with a radius, R, of 5 m, and are separated by an arc of length 0.5 m, we can estimate the angle of view of our detector, assuming the beam covers an arc of the circle that is 1 m long, estimated from the beam profile in Fig. 2(b). For the light beam to exit the area of detection of the two-APD receiver, the center of the beam has to travel 0.75 m along the circle (0.25 m to face one of the detectors and 0.5 m to exit its detection area), which corresponds to an estimated maximum angle, ${\theta _{\textrm{max},\textrm{e}}}$ = 0.75 m/$R$ = 8.6$^\circ $. The experimental results are in good agreement with this prediction, with a maximum angle, ${\theta _{\textrm{max}}}$ = 9$^\circ $, which gives a total angle of view of ${\pm} \; $9$^\circ $. These results highlight the advantage of using UVC light in OWC communication, as it relieves the alignment requirements due to its characteristically high scattering coefficient in air. The reception angle can be increased further by increasing the beam divergence angle and/or the separation of the two detectors. However, this is expected to decrease the AIR at $\theta $=0°.

Figures 3(a) and 3(b) further illustrate the utilization of the channel capacity using PS by respectively showing the GMI and the normalized GMI (NGMI) for each subcarrier at an angle of 0$^\circ $. Initially, the entropy of the PS-QAM signal for each subcarrier is individually fixed at the value of the Shannon limit. We set the pre-FEC BER to 3.8${\times} $10⁻³, which is the most conventional BER threshold. For normal bit loading technology, the entropy of the uniform QAM constellation for each subcarrier can be determined according to the relation among the BER, SNR, and the uniform QAM entropy in Fig. 5 of [27]. The entropy of uniform QAM constellation for all subcarriers is shown in Fig. 3(a). The corresponding Shannon limit is also depicted in Fig. 3(a) as a reference. It is observed from Fig. 3(a) that the GMI values in the case of using PS bit loading continuously vary following the Shannon limit of the channel, owning to the possibility of continuous adjustment of the transmission entropy for each subcarrier using PS. The small GMI gap between PS bit loading and the Shannon limit comes from the necessary pre-FEC code redundancy to guarantee error-free transmission. The drop to 1 bit/symbol indicates the switch to 2-QAM to reduce the system complexity. On the contrary, the transmission entropy of normal bit loading technology can only assume discrete integer values due to the limited utilization of equiprobable discrete-alphabet QAM constellations, which means that the allocated entropy using normal bit loading technology is always sub-optimal and inflexible. Figure 3(a) simply shows a comparison between PS-bit-loading-DMT and conventional bit-loading-DMT based on entropy allocation. The detailed superiority of the PS-bit-loading-DMT scheme over conventional bit-loading-DMT scheme is compared in detail and analyzed in [28]. Moreover, the enhancement on the flexibility of transmission capacity indicates that PS bit loading technology is a more suitable and robust modulation method to take advantage of the channel capacity in the UVC solar-blind communication system with complex frequency-fading issues. We also observed that the NGMI values for all subcarriers are above 0.9 (allowing the use of a rate-0.858 FEC code), which means that using the concatenated code of 20% low-density parity-check (LDPC) and 6.25% staircase code as the FEC code can provide error-free transmission and achieve 10⁻¹⁵ post-FEC BER [29]. The FEC code redundancy can be reduced even further using a binary rate-0.9 FEC code if it is developed in the future. Figures 3(c) and 3(d) show the superimposed constellation diagrams of different subcarriers along the channel for an angle of 0$^\circ $ and 9$^\circ $, respectively. To conveniently display constellations of all subcarriers, we divided the subcarriers into several groups and superimposed all constellation diagrams from one group into one constellation diagram. For example, the last displayed constellation from the left in Fig. 3(c) is the superimposed constellations of the last 32 subcarriers. Because of the drop in the received power and the fact that one of the detectors is not receiving the signal at 9$^\circ $, the channel response and the AIR drop, but even at this extreme angle, the high speed of the link is maintained.

 

Fig. 3. (a) The generalized mutual information (GMI) values using PS bit loading and the entropy of the uniform constellation using normal bit loading (Pre-FEC BER = 3.8${\times} $10⁻³) for the different subcarriers using the SIMO link. The drop to 1 bit/symbol in PS bit loading indicates the switch to 2-QAM to reduce the complexity. (b) The normalized GMI (NGMI) values using PS bit loading showing a minimum value above 0.858 (indicated by the dashed line). The constellation diagrams at different subcarriers at (c) $\theta = 0^\circ $ and (d) $\theta = 9^\circ $.

Download Full Size | PDF

To demonstrate one of the advantages of using a SIMO link over its SISO counterpart, we show the results of the same experiment done using a SISO link in Fig. 4 at different transmission distances. The AIR values versus the relative angle were recorded at transmission distances of 1, 2, and 5 m and spline interpolation was used between the measured points. As expected, increasing the transmission distance decreases the angle of view of the detector. Moreover, the SISO link at 5 m showed an angle of view that is 40% smaller compared to the ${\pm} $ 9$^\circ $ angle of view of the SIMO system, as has been demonstrated in Fig. 2(a). The overall trend at all distances follows a similar trend of that of the SIMO link due to the nonuniform beam profile, with the AIR values increasing again near the edges.

 

Fig. 4. The AIR values of the SISO link versus the angle $\theta $ at transmission distances of 1, 2, and 5 m with spline interpolation used between the measured points.

Download Full Size | PDF

3.2 Channel performance in the presence of emulated fog

Another aspect that is of great importance is studying the effects of adverse weather conditions on UVC links. While some weather conditions can cause power losses in the received signal in LOS and diffuse-LOS, the presence of fog in the communication channel was shown to increase the received power in NLOS links [30]. Another study used Monte Carlo simulations to investigate the effects of different haze and fog densities on the path loss [31]. In addition, in [32], the effect of atmospheric turbulence was both simulated and experimentally verified.

An important advantage gained from using two detectors is the increased immunity to degradation in the signal quality induced by weather conditions, such as the presence of fog and haze particles. Unlike in NLOS links, these particles degrade the performance of the communication link in LOS and diffuse-LOS configurations. To study these effects, we used a mist generator placed right in front of the lens on the transmitter side to emulate foggy environments. To be able to benchmark the performance of the system in extreme conditions and collect data points for a longer period time, we sent a 0.5 MHz, 16-QAM DMT test signal using normal bit loading and measured the BER. We repeated the measurement with and without the emulated foggy condition for the SISO and SIMO systems.

In order to quantify the strength of the signal fading induced by the foggy channel, we characterize the variations in the envelope of the signal. We first find the upper and lower root-mean-square (RMS) envelopes using 1% of the length of the recorded signal as the size of the sliding window. We take the difference of the upper and lower RMS envelopes as the baseline of our signal. After normalizing by the mean, we can calculate the scintillation index, $\sigma_B^2$, of the signal by finding the variance of the normalized baseline. In other words:

To highlight the impact of weather conditions on UVC links, we started by testing a SISO link affected by the presence of mist where the transmitter is directly facing a single receiver. Figures 5(a) and 5(b) show the detected signals of the SISO system in the absence and presence of the emulated fog with their baselines, respectively. It is clearly observed that the foggy channel induces random variations in the received signal’s baseline. These variations are illustrated in the histograms of the baselines shown in Fig. 5(c). From the two histograms, we can see the difference in the stability of the received signal between the two scenarios in the absence and presence of the emulated fog. Moreover, the scintillation index in the foggy channel is around 6.6${\times} $10⁻³. This strong signal fading translates into poor performance of the communication link.

 

Fig. 5. The received signal from the SISO link in: (a) the presence and (b) absence of emulated fog. (c) The histograms of their baselines.

Download Full Size | PDF

We then tested the SIMO link under the emulated foggy condition. The spatial separation provides significant diversity gain since it can be assumed that the emulated fog affects each channel independently. To verify that, we calculated the correlation coefficient of the baselines of the two received signals and found that it was less than 0.05, verifying the advantage of implementing spatial diversity.

To combine the two signals, we used Eq. (1), but with the weights ${w_{1,2}}$ replaced by the functions of time, ${w_{1,2}}(t )$, which are periodically updated to account for the time-variant, random changes in the signal induced by the emulated fog. Using this algorithm, the combined signal makes use of whichever signal is less affected by the fog-induced fading at a given timeframe. The waveforms of the signals received from the two detectors, RX₁ and RX₂, in the presence of the emulated fog, as well as the combined signal, are shown respectively, in Figs. 6(a), 6(b), and 6(c) with their baselines. From the flatness of the baseline of the combined signal, we can observe the significant improvement in the stability of the signal using the SIMO system. For example, in the period shown in Fig. 6, we can see that the signal of RX₂ is affected less than that of RX₁ (the scintillation index values are 2.8${\times} $10⁻³ for RX₂ and 6.1${\times} $10⁻³ for RX₁) due to the non-uniformity of the spatial distribution of the emulated fog in the channel. In such a scenario, the SIMO system relies more on RX₂, without disregarding information from RX₁, providing a more reliable link than each independent link. However, over long periods of time, the overall effects on the two detectors are expected be symmetric.

 

Fig. 6. The signals and their baselines of: (a) RX₁, (b) RX₂, and (c) the combination of both in the SIMO link.

Download Full Size | PDF

We evaluated the performance of the SIMO link as well as the performance of each channel independently in the presence and absence of the emulated fog in terms of the BER, as shown in Fig. 7, where the BER shown for the SISO link is the average of the BER values for the two independent channels. The performance of the two channels was not identical due to the random variations in the locations of the mist molecules. These variations are slow when compared to the data rate. To highlight the effectivity of the SIMO link and its immunity to adverse weather conditions, we are considering a case where Channel 1 (corresponding to RX₁) is affected severely by the mist whereas Channel 2 (corresponding to RX₂) experiences minimal degradation, with a BER of 4.3${\times} $10⁻³ in emulated fog. Even in such a scenario, the SIMO link offered a lower BER than Channel 1, as well as Channel 2, even though it was combined with a degraded channel. Despite the fact that both channels exhibited BER values above the FEC limit, the spatial diversity reduced the BER below the FEC limit. More importantly, it also provided high immunity to the degradation caused by the foggy channel as can be seen from the small increase in the BER after introducing the mist. Therefore, in cases where the fading is more sever, it can be inferred that the BER improvement is expected to be much more significant, since the change in the SIMO link is much slower compared to the SISO link (the increase in the SISO link’s BER is around tenfold that of the SIMO link). The spatial diversity also provided improved performance in the absence of the emulated fog in the channel.

 

Fig. 7. The average BER values of the independent channels (SISO) and the SIMO link in the absence and presence of emulated fog. (EGC: equal-gain combining)

Download Full Size | PDF

Figure 7 also shows the BER using the simplest combining technique, equal-gain combining (EGC), which does not require channel state information. It is observed that by implementing this simple method, in which we set ${w_1}(t )$ = ${w_2}(t )$ = 0.5, the performance is improved significantly. This result highlights the benefit of using spatial diversity in SIMO systems to mitigate the negative effects of adverse weather conditions on UVC LOS and diffuse-LOS links even when using a simple technique.

4. Conclusions

Our demonstration leverages on UVC light and PS-DMT and diversity reception for diffuse-LOS OWC in order to establish high-speed, reliable communication links with extended coverage. The use of a 1${\times} $2 SIMO system in a diffuse-LOS link, stable over ${\pm} $ 9$^\circ $ relative angle between the transmitter and the receiver, proved to be practical in providing high-speed OWC with an AIR up to 1.09 Gbit/s while relieving the strict alignment requirements. The use of PS bit loading in the DMT modulation scheme offered high-speed communication through optimum utilization of the channel capacity. Moreover, when a single detector was used in a SISO system, the link was still maintained over a wide range of angles when compared to LOS links. This is facilitated by the inherently high scattering of UVC light in air. The stability of the SIMO link in the presence of emulated fog further emphasizes the benefits of relying on multiple receivers, taking advantage of the spatial diversity gain provided by the diversity reception. Furthermore, the SIMO link established using UVC light showed high immunity to signal fading induced by an emulated foggy condition. The results represent the current advancement in UVC communication based on compact diode devices, which allows for establishing robust links with relieved alignment requirements.