Open Access
Add to BibSonomy
Abstract
Bubbles-induced turbulence poses a significant challenge to the stability of underwater wireless optical communication (UWOC) system. Existing methods for understanding channel characteristics rely on the pilot information from the feed-back channel, which are ineffective and inaccurate due to the rapidly changing nature of the underwater channel. We propose a machine-vision-based channel prediction mechanism which contains three modules of motion judgment module, image processing module and scintillation index (SI) prediction module. The mechanism captures images of bubbles and calculates the bubble density. Subsequently, a relational function is applied to acquire the predicted SI which quantifies the impacts of bubbles on the channel. Experimental results validate the effectiveness of the proposed mechanism.
© 2023 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Underwater wireless communication (UWC) is essential for the exploration and development of ocean, which transmits data in the underwater environment through wireless carriers, i.e., radio-frequency (RF) waves, acoustic waves, and optical waves [1] . Compared with the high latency of acoustic waves communication and the high attenuation of RF communication, underwater wireless optical communication (UWOC) receives the widespread attention with the ability to support higher data rates at low latency levels and the advantages of richer spectrum resources, better security, stronger anti-interference capability, and lower cost [2,3]. UWOC is increasingly used as a primary solution for underwater communication and plays an important role in many kinds of civil, military, and scientific missions, such as undersea rescue and disaster response, ocean monitoring, and exploration [4]. However, the underwater environment is complex and dynamic, which makes the UWOC channel easily interfered by various random factors, among which bubbles-induced turbulence is an typical factor [5].
Bubbles can be generated by the movement of marine organisms and unmanned underwater vehicle, the decomposition of inorganic materials, or the leakage from submarine gas fields in oceans. In the UWOC systems, the presence of bubbles in the communication link causes fluctuations in the underwater refractive index and induces bubble-beam interactions, leading to beam misalignment between the transmitter (Tx) and receiver (Rx) [6], which is manifested as beam wandering, spot dancing, and intensity fluctuations in the received optical signal [7]. The presence of bubbles greatly increases the chances of link outage and the probability of bit errors increase [8], leading to degradation in the performance of the communication system [9].
To design a reliable UWOC system, it is critical to characterize the effect of bubbles on the channel. In recent years, many researchers have conducted study on underwater bubbles channel statistical models. Zedini et al. proposed a mixture exponential-generalized gamma (EGG) distribution model to characterize turbulence-induced fading in UWOC channels in the presence of bubbles [10]. M. Singh et al. proposed a Gaussian mixture model (GMM) model to describe the intensity fluctuations of the received optical beam when an optical beam propagates through the underwater environment in the presence of bubbles [11]. Qiu et al. proposed a unified Weibull-generalized gamma distribution model to characterize the effect of ocean turbulence on optical beam propagation [12]. However, for better acquiring channel characteristics, these works need to utilize the pilot information from the feed-back channel to analyze channel statistical model and calculate parameters, which are ineffective and inaccurate due to the rapidly changing nature of the underwater channel. Scintillation index (SI) is a key parameter used to quantify the fluctuations in received optical signals and assess the impact of bubbles on UWOC. Bernotas et al. associated channel statistical model with SI [13]. Jamali et al. found that the change of bubble density caused significant fluctuations in the received optical signal intensity [14]. Huang et al. first realized a UWOC system incorporates with machine vision technology and conducted verification analysis [15]. If we establish a correlation between the bubble density and the SI, and utilize machine vision to identify the bubble density, we can obtain the predicted SI, which helps us to acquire the optimal channel statistical model and the parameters faster.
In this paper, we propose a channel prediction mechanism which integrates machine vision into UWOC system for active channel prediction. The mechanism captures on-line images of bubbles and adopts image processing technology to obtain the bubble density. Combining the bubble density with relational function which characterized the relationship of the bubble density and the SI, UWOC system can obtain the predicted SI. Based on the mechanism, the impact of bubbles can be predicted actively, allowing the transmitter to adopt corresponding coding or equalization strategies in advance, which is meaningful for building a stable, flexible, and fast-reacting UWOC system.
The rest of the paper is organized as follows: In Section 2, the principle of channel prediction mechanism is introduced. Section 3 shows the methodology of the conducted experiments and the experimental results. Finally, the conclusion is drawn in Section 4.
2. Principle of channel prediction mechanism
Figure 1 presents the block diagram of the channel prediction mechanism. The prediction mechanism integrates camera into UWOC system and consists of three modules: motion judgment module, image processing module and SI prediction module.
(1) Motion judgment module: Capture on-line image of bubbles and judge whether the communication link will be obstructed.
The camera captures on-line images of bubbles possibly generated by autonomous underwater vehicles (AUV), marine organisms, underwater plants, etc. The classical Harris and Lucas-Kanade algorithms are then used to predict the movement direction of bubbles for judging whether the bubbles would obstruct communication link. The Harris algorithm is utilized to detect the feature points of bubbles, and the Lucas-Kanade algorithm is adopted to track and predict the motion of feature points, so as to predict the movement direction of bubbles. Based on the predicted motion direction and the known position of the UWOC system, whether the communication link will be obstructed can be judged. The mechanism will proceed to image processing module once the output result indicates that the communication link will be obstructed by bubbles.
(2) Image processing module: Processes images for subsequent calculation of the bubble density.
Image processing module includes image enhancement, image preprocessing, binarization, morphological transformation, edge detection, and contour extraction. Firstly, image is enhanced and preprocessed to eliminate background noise interference and sharpen the edges of bubbles. Subsequently, based on Eq. (1), the processed image is divided into bubbles area and background area, with the bubbles area is set 0 and the background area is set 1.
Morphological transformation is applied to effectively fill the holes of binarized bubbles image. Then, canny algorithm and contour extraction are used for edge detection and better characterization of bubbles, which are conducive to calculate the bubble density in prediction module.
(3) SI prediction module: Establish a relational function which characterizes the relationship between the bubble density and the SI.
With the benefit of image processing module, the bubble density can be easily calculated. To predict the SI by the bubble density, it is crucial to establish a relational function for characterizing the relationship between the bubble density and the SI.
The SI (denoted by $\sigma _I^2$ ), which is common in the literature for measuring the strength of turbulence, defined as the normalized variance of intensity fluctuations, is given by [16]:
Assuming that the generated normalized current only has two states (i.e., ${I_1}$ and ${I_{\text {2}}}$), where ${I_1}$ represents the current when channel is obstructed, ${I_{\text {2}}}$ represents the current when the channel is unobstructed with ${I_2} = 1$. Suppose that $p$ is the probability of the channel is obstructed by bubbles, we get that $P\left ( {I = {I_1}} \right ) = p$ and $P\left ( {I = {I_2} = 1} \right ) = 1 - p$ . By using Eq. (2), the SI function can be derived as
The presence of bubbles in the communication link results in random fluctuations of the received optical signal intensity which causes the generated normalized current ${I_1}$ of PD is dynamic. Hence, the SI function can be further expressed as
where $a,b,c$ are the scale parameters. We select $n$ cases for averaging among the dynamic normalized currents generated by the PD when the bubbles obscure communication link, and then obtain the Eq. (5).Furthermore, we adopt $N$ to present the bubble density. The relationship between $N$ and $p$ can be shown as Eq. (6).
Combing Eq. (5) and Eq. (6), the relational function between the bubble density and the SI can be derived as
Based on the above three modules, the channel prediction mechanism achieves the active channel prediction, which is helpful for the transmitter adjust the communication strategy in time.
3. Experiments and results analysis
In this section, the effectiveness of three modules of the channel prediction mechanism were validated. Figure 2 shows the experimental setup for emulating the UWOC system. We utilized a water tank with dimensions of 0.5 $\times$ 0.2 $\times$ 0.2 ${m^3}$. To minimize the interference of light reflected from the sidewalls of the tank, we wrapped it with black stickers. We used tap water as a replacement for seawater due to their similar attenuation coefficients [17]. At the transmitter, a 450 $nm$ blue laser diode (LD) with a maximum power output of 1600 $mW$ was drived by a direct current (DC) power supply. The air pump placed in the middle of the water tank was used to generate underwater bubbles. The on-line images were acquired by a 4K camera and transferred to a personal computer (PC), which adopted the prediction mechanism to predict the channel state. At the receiver, the beam was captured by a PIN PD with an active area diameter of 800 $\mu m$ for amplitude measurement. We further collected samples of PD’s generated current through a 3.5 $GHz$ bandwidth oscilloscope with a sampling rate of 5 $kSa/s$.
To generate bubbles of different densities, we used a power-adjustable air pump which was always working throughout our experiments. The air pump has a flow rate range of 0-15 $L/min$ with two working modes, i.e., single-port and double-port, which was connected to an air tube with an air disc at the end. To manipulate the bubble density, we fixed the air tube diameter and divided the adjustable air pump into 15 levels according to the air flow rate. It can be seen from Fig. 3 that as the power of the air pump increases, the number of bubbles increases. Besides, in comparison to single-port work mode, the double-port work mode generates more bubbles and the bubbles are arranged more closely.
By observing the experimental results, we validated the effectiveness of motion judgement module, image processing module and prediction module. In the motion judgment module, Harris algorithm was utilized to obtain the feature point of the bubbles, and Lucas-Kanade algorithm was adopted to obtain the motion vector of the feature point. The motion vector represents the motion direction and distance of feature point, which can be applied to predict the motion of objects. Assuming that the position of the feature point in the previous frame is $\left ( {{x_0},{y_0}} \right )$, and the position of the feature point in the next frame is $\left ( {{x_1},{y_1}} \right )$. The motion vector is denoted as $\left ( {{x_1} - {x_0},{y_1} - {y_0}} \right )$, where ${x_1} - {x_0}$ represents the displacement of the feature point in the horizontal direction and ${y_1} - {y_0}$ represents the displacement of the feature point in the vertical direction.
The experimental results of motion judgement module when air flow rate at 4 $L/min$, 8 $L/min$, and 12 $L/min$ respectively were shown in Table 1 and Fig. 4. We calculated the angular difference between the predicted motion vector by Lucas-Kanade algorithm and actual motion vector which was upward in the experiments. Smaller angular difference meant better prediction, the value of angular difference less than 0 meant the prediction direction is on the left, and the value of angular difference bigger than 0 meant the prediction direction is on the right. As shown in Table 1, the angular differences were all less than ${{\rm {1}}^ \circ }$ when air flow rate at 4 $L/min$, 8 $L/min$, and 12 $L/min$. Figure 4 could be plotted based on the motion vectors in Table 1, the actual direction of motion was indicated by blue arrows and the predicted direction of motion was represented by red arrow with positive correlation between length of arrows and movement speed. Figure 4 and the data in Table 1 proved the availability of the motion judgment module. Based on the predicted motion direction and the known position of the UWOC system, the motion judgment module could judge whether the communication link would be obstructed by bubbles.
The experimental result of image processing module is shown as Fig. 5. The image processing involves a series of operations, e.g., image enhancement, image preprocessing, binarization, morphological transformation, edge detection, and contour extraction. As can be seen from Fig.5, after image processing, the edges of bubbles can be accurately detected, which was beneficial for the subsequent calculation of the bubble density and proved the availability of the image processing module. Benefiting from the image processing module, we easily calculated the bubble density.We selected the unit area whose width was the size of the air disk connected to the end of the air tube and the length was from the air disk to the communication link. We fixed the air flow rate and collected 400 frames of images continuously by a 4K camera. With the help of machine vision, we sequentially calculated the bubble density within the unit area of 400 frames of images and averaged them to obtain the corresponding bubble density at the fixed air flow rate.
Table 1. The experimental results of motion judgement module.
In order to obtain the SI, we collected 50,000 samples of PD’s generated current through oscilloscope. We conducted twenty experiments and selected several typical waveforms as shown in Fig. 6 and Fig. 7. The typical waveforms when the air flow rate at at 4 $L/min$, 8 $L/min$, and 12 $L/min$ respectively with air pump working in single-port mode was shown in Fig.6, where higher air flow rate meant a higher number of air bubbles generated by the air pump. Fig.7 showed the typical waveforms when the air flow rate at 4 $L/min$, 8 $L/min$, and 12 $L/min$ respectively with air pump working in double-port which generated more air bubbles compared to single-port mode. It was observed that there were many points in the three waveforms where the ordinate was 0 or close to 0, which meant that the receiver did not receive the signal due to the communication link was obscured by bubbles. As the air flow rate increased, the number of points with the ordinate of 0 increased, implying that the probability of bubbles obscuring the communication link increased. In order to remove the effects of the underwater channel on signal attenuation, we performed Min-Max normalization on 50,000 samples to get 50,000 normalized samples. Combined with Eq. (2), we obtained the SI.
Fig. 6. Typical waveforms selected from the twenty experiments when air pump works in single-port mode.
Fig. 7. Typical waveforms selected from the twenty experiments when air pump works in double-port mode.
In the prediction module, we adopted the ${R^2}$ measure to validate the fitting degree of the relational function to experimental data for evaluating the performance of the prediction module. The ${R^2}$ measure, also known as the goodness of fit measure, was shown as Eq. (8).
in which ${SS_{\mathrm { reg}}}$ is the sum of the square errors of the statistical distribution under consideration, i.e., $S{S_{reg}} = \sum \nolimits _{i = 1}^M {{{\left ( {{f_{m,i}} - {f_{p,i}}} \right )}^2}}$, where $M$ is the number of bins of the acquired data histogram, ${{f_{m,i}}}$ and ${{f_{p,i}}}$ are respectively the measured and predicted probability values for a given received intensity level corresponding to the $i$th bin. And $S{S_{tot}}$ is the sum of the squares of distances between the measured points and their mean, i.e., $S{S_{tot}} = {\sum \nolimits _{i = 1}^M {\left ( {{f_{m,i}} - \bar f} \right )} ^2}$, where $\bar f = \sum \nolimits _{i = 1}^M {{f_{m,i}}/M}$. Clearly, as the value of the ${R^2}$ measure for a given distribution approaches its maximum (i.e., 1), the distribution better fits the experimental data.Furthermore, the air pump was applied to generate 15 levels of bubbles according to the air flow rate in both single-port and double-port working modes, with the bubble density generated in single-port mode ranged from 100 to 300 and in double-port mode ranged from 150 to 500. We recorded the bubble density and the corresponding SI as experimental data and adopted the Eq. (7) as relational function to fit them. The fitting results and the values of ${R^2}$ and scale parameters were shown in Fig. 8.
The fitting values ${R^2}$ of both single-port mode and double-port mode were greater than 0.99 and extremely close to 1, which proved that the fitting degree of the relational function to experimental data was excellent. Furthermore, we discovered an interesting phenomenon in experiment where as the bubble density increased, the SI exhibited a trend of increasing and then decreasing. We hypothesized that when the bubble density was low, the interference of bubbles on the channel was minimal, resulting in a lower SI. As the bubble density increased, the interference of bubbles on the channel enhanced, causing an increase in the SI. However, when the bubble density reached a certain level, the repeated scattering of optical signal between bubbles became more pronounced, which implied that the scattered light was more likely to be re-received by PD, thus, resulting in the decrease of SI.
4. Conclusion
In this paper, we present a machine-vision-based channel predicting mechanism, which contains motion judgment module, image processing module, and SI prediction module. The mechanism captures bubbles and predicts their impact in advance for active channel prediction. By observing the experimental results and fitting the experimental data, we validate the effectiveness of the three modules to evaluate the performance of the proposed mechanism. Based on the prediction mechanism, the transmitter has the capability to adopt the appropriate modulation method, coding scheme or equalization design to adapt to the UWOC channel in advance, which is important for building a stable, flexible, and fast-reacting UWOC system.
Funding
National Key Research and Development Program of China (No.2013CB329205); National Natural Science Foundation of China (62021005).
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. Z. Zeng, S. Fu, H. Zhang, Y. Dong, and J. Cheng, “A survey of underwater optical wireless communications,” IEEE Commun. Surv. Tutorials 19(1), 204–238 (2017). [CrossRef]
2. G. Schirripa Spagnolo, L. Cozzella, and F. Leccese, “Underwater optical wireless communications: Overview,” Sensors 20(8), 2261 (2020). [CrossRef]
3. S. Zhu, X. Chen, X. Liu, G. Zhang, and P. Tian, “Recent progress in and perspectives of underwater wireless optical communication,” Prog. Quantum Electron. 73, 100274 (2020). [CrossRef]
4. N. U. R. Junejo, M. Sattar, S. Adnan, H. Sun, A. B. M. Adam, A. Hassan, and H. Esmaiel, “A survey on physical layer techniques and challenges in underwater communication systems,” J. Mar. Sci. Eng. 11(4), 885 (2023). [CrossRef]
5. E. Zedini, H. M. Oubei, A. Kammoun, M. Hamdi, B. S. Ooi, and M.-S. Alouini, “A new simple model for underwater wireless optical channels in the presence of air bubbles,” in IEEE Global Communications Conference, (2017), pp. 1–6.
6. I. C. Ijeh, M. A. Khalighi, M. Elamassie, S. Hranilovic, and M. Uysal, “Outage probability analysis of a vertical underwater wireless optical link subject to oceanic turbulence and pointing errors,” J. Opt. Commun. Netw. 14(6), 439–453 (2022). [CrossRef]
7. M. Singh, M. L. Singh, R. Singh, H. Kaur, P. Chohan, and S. Kaur, “Comprehensive experimental and statistical analysis of the effects of salinity gradient, temperature gradient, air bubbles on the performance of underwater wireless optical communication link,” J. Mod. Opt. 69(17), 978–994 (2022). [CrossRef]
8. M. Singh, M. L. Singh, H. Kaur, H. S. Gill, Priyanka, and S. Kaur, “Statistical channel model to characterize turbulence-induced fluctuations in the underwater wireless optical communication links,” Int. J. Commun. Syst. 35(15), e5298 (2022). [CrossRef]
9. Y. Guo, M. Kong, M. Sait, S. Marie, O. Alkhazragi, T. K. Ng, and B. S. Ooi, “Compact scintillating-fiber/450-nm-laser transceiver for full-duplex underwater wireless optical communication system under turbulence,” Opt. Express 30(1), 53–69 (2022). [CrossRef]
10. E. Zedini, H. M. Oubei, A. Kammoun, M. Hamdi, B. S. Ooi, and M.-S. Alouini, “Unified statistical channel model for turbulence-induced fading in underwater wireless optical communication systems,” IEEE Trans. Commun. 67(4), 2893–2907 (2019). [CrossRef]
11. M. Singh, M. L. Singh, G. Singh, and H. S. Gill, “Statistical channel model for underwater wireless optical communication system under a wide range of air bubble populations,” Opt. Eng. 60(03), 036111 (2021).
12. H. Qiu, Z. Huang, J. Xu, W. Zhai, Y. Gao, and Y. Ji, “Unified statistical thermocline channel model for underwater wireless optical communication,” Opt. Lett. 48(3), 636–639 (2023). [CrossRef]
13. M. P. Bernotas and C. Nelson, “Probability density function analysis for optical turbulence with applications to underwater communications systems,” Proc. SPIE 9827, 98270D (2016). [CrossRef]
14. M. V. Jamali, A. Mirani, A. Parsay, B. Abolhassani, P. Nabavi, A. Chizari, P. Khorramshahi, S. Abdollahramezani, and J. A. Salehi, “Statistical studies of fading in underwater wireless optical channels in the presence of air bubble, temperature, and salinity random variations,” IEEE Trans. Commun. 66, 4706–4723 (2018). [CrossRef]
15. Z. Huang, L. Zu, Z. Zhou, X. Tang, and Y. Ji, “Computer-vision-based intelligent adaptive transmission for optical wireless communication,” Opt. Express 27(6), 7979–7987 (2019). [CrossRef]
16. W. Liu, Z. Xu, and L. Yang, “Simo detection schemes for underwater optical wireless communication under turbulence,” Photonics Res. 3(3), 48–53 (2015). [CrossRef]
17. J. Giles and I. Bankman, “Underwater optical communications systems. part 2: basic design considerations,” in IEEE Military Communications Conference, (2005), pp. 1700–1705.
