Open Access
Add to BibSonomy
Abstract
We propose and demonstrate using the DIALux software with our proposed linear-regression machine-learning (LRML) algorithm for designing a practical indoor visible light positioning (VLP) system. Experimental results reveal that the average position errors and error distributions of the model trained via the DIALux simulation and trained via the experimental data match with each other. This implies that the training data can be generated in DIALux if the room dimensions and LED luminary parameters are available. The proposed scheme could relieve the burden of training data collection in VLP systems.
© 2021 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
Including additional values to the present light emitting diode (LED) lighting infrastructure, visible light communication (VLC) has been proposed [1–5]. VLC can also be employed in indoor positioning systems (IPS) providing visible light positioning (VLP). It can locate objects or persons in buildings or underground areas, where the popular Global Positioning System (GPS) fail to provide accurate positioning. Different IPS techniques have been proposed, such as using Wireless Fidelity (WiFi) or Bluetooth [6]; however, the positioning accuracy are limited by the electromagnetic interference (EMI). VLP is based on the optical signal for positioning; hence, it can be applied in radio-frequency (RF) restricted areas, such as hospitals and power stations. Different VLP systems have been proposed, such as using proximity [7], angle-of-arrival (AOA) [8,9], time-of-arrival (TOA)/time-difference-of-arrival (TDOA) [10,11], and received-signal-strength (RSS) [12,13]. Among these technologies; RSS-VLP technique is simple and cost-effective. It relies on the Lambertian model of the optical signal, and analyzes different received optical powers to estimate the distances between LED transmitter (Tx) and optical receiver (Rx). To enhance the RSS-VLP positioning accuracy, machine learning (ML) algorithms, such as regressions [14] or neural networks (NNs) [15,16] were proposed. There are only few studies reported in the literatures on the LED Tx arrangements to achieve accurate VLP systems.
In this work, we propose and demonstrate using the DIALux software with our proposed linear-regression machine learning (LRML) algorithm for designing a practical and accurate indoor VLP system. The DIALux software is one of the leading software in lighting design industry. Moreover, it is available free of charge with a variety of LED luminary emission profiles for selection. The proposed LRML is simple and non-iterative; hence, the VLP model can be built time-efficiently. Experimental analyses reveal that average position errors and the error distributions of the model trained via DIALux simulation and trained via experimental data are match with each other. The proposed scheme could be a promising tool for designing IPS, as well as relieving the burden of training data collection in VLP systems.
2. Algorithms and experiment
In VLP systems, collecting fingerprints or RSS optical powers from all the LEDs with exact indoor coordinates are very time-consuming. When the LED power changes due to aging or different brands of LED are replaced; all the training data in the database should be updated accordingly. The burden of training data collection will prohibit the implementation of large scale VLP system. Here, we propose employing DIALux together with LRML algorithm to design an indoor VLP system. It can also relieve machine learning training data collection. In this section, we use a practical indoor VLP environment to illustrate the operation mechanism of our proposed scheme. Figure 1(a) shows the top-view of the practical room with dimensions about 2 m × 1.55 m. The training point, testing point and LED lamp locations are also illustrated. It is worth to note that in our actual room, the 4 LEDs are not arranged in a perfect rectangle in order to emulate the condition with non-uniform LED arrangement. The vertical distance between the ceiling and the Rx is about 2.7 m. The LED lamp (TOA LDL030C) is commercially available with output power of 13 W consisting of 26 LED chips. Each LED is modulated by a specific Manchester-coded identifier (ID) at date rate of 3.125 kbit/s upconverted to carrier frequencies at 47 kHz, 59 kHz, 83 kHz or 101 kHz, respectively via a simple driver circuit as shown in Fig. 1(a). There is no need of synchronization among the LEDs. The odd frequencies are used to avoid harmonic spectral overlap. The client side is an autonomous mobile robot (AMR) with a photo-detector (PD) connected to a real-time oscilloscope (RTO) as the Rx, as illustrated in the photo of Figs. 1(b) and (c). The RTO acts as the analog-to-digital converter (ADC), which digitizes the signal in real-time. In the experiment, it allows continuous detection and sampling of the PD received optical signals sent from different LEDs. The sampled signals will be processed to retrieve the IDs and carrier frequencies. The RTO (PicoTechnology 5243D) has a 3-dB analog bandwidth of 100 MHz, with sampling rate of 500 MSample/s at vertical resolution of 8 bits. The PD (Thorlabs PDA100A2) is silicon based with detection wavelength range from 320 nm to 1100 nm. The bandwidth is 11 MHz with typical responsivity of 0.72 A/W. The AMR (ADATA prototype) allows continuous training and testing data collections. It has the dimensions of 95 cm × 65 cm × 36 cm and maximum speed of 1 m/s. For large scale VLP deployment, the whole area will be divided into different unit cells with specific IDs and repeatedly deployed RF carrier frequencies as shown in Fig. 1(d). Since different LEDs have different IDs, the location of the AMR at which unit cell can be identified at the beginning. After this, the position of the AMR inside that particular unit cell can be obtained by analyzing the RSS of the 4 carrier frequencies. Figure 2 shows the decoding mechanism of the RSS VLP system. The received RF signals at carrier frequencies of 47 kHz, 59 kHz, 83 kHz and 101 kHz are filtered, and amplitude measured. These amplitudes are used for the RSS positioning inside an unit cell. Then, each RF signal is down-converted to retrieve the Manchester-coded ID. These IDs are used to identify where the AMR at specific positioning unit cell.
Fig. 1. (a) Top-view of an unit cell indicating the training, testing and LED locations. Photos of (b) AMR with PD connected to a RTO as Rx, (c) actual room for VLP, (d) Real VLP implementation using different unit cells to cover the whole indoor areas. AMR: autonomous mobile robot, RTO: real-time oscilloscope, PD: photodiode.
Fig. 3. (a)-(e) Virtual rooms and (f)-(j) emitted light intensity profiles in DIALux when different LEDs are switched on separately and simultaneously.
A virtual room with the dimensions of the actual room are built in the DIALux software. The LED luminary emission profile in the Illuminating Engineering Society (IES) file format are imported to the DIALux. The IES file are provided by the LED lamp company. The processes for evaluating our proposed VLP system is as follow. (1) The AMR acquires training and testing data in the actual room based on the room map as shown in Fig. 1(a). (2) Then, we collect the training data at the corresponding locations from the virtual room in DIALux. (3) After this, we apply our LRML VLP model, and we train this model by using the experimental training data collected in the actual room in step 1; and the training data collected in the virtual room in step 2. (4) Finally, we evaluate both the trained models by the experimental testing data.
In order to verify the feasibility and the accuracy of our proposed scheme, complicated machine learning or NNs [15,16] are not used. Here, we implement a simple LRML model, which is expressed in Eq. (1), where P is the predictive coordinate matrix, WML is the weight vector, D is the dimension (D = 4 as 4 LEDs are used), and Ф is the design RSS matrix.
3. Results and discussions
Figures 3(a)-(e) and 3(f)-(j) illustrate virtual rooms and the emitted light intensity profiles in the DIALux program when different LEDs are switched on separately and simultaneously. Gaussian random noises are added to the received optical signal in the simulation. When the 4 LEDs are switched on simultaneously, the light intensity projected on the floor is circular and uniform as illustrated in Figs. 3(e) and (j). This condition also matches with the actual room used in the experiment as shown in Fig. 1(c). First of all, the light intensity profile of a single LED is studied experimentally. The LED is at the ceiling and the PD on the AMR moves horizontally on the floor away from the LED. Figure 4(a) shows the experimentally measured light intensity profile against different PD horizontal offsets. It can be observed that the profile is Lambertian shaped with nearly no detectable optical signal at the horizontal offset distance of 150 cm. As mentioned in section 2, the AMR acquires training and testing data in the actual room based on the room map as shown in Fig. 1(a). It is worth to note that the training and testing data are at different locations. Then, the collected training data is utilized to build the LRML VLP model.
Fig. 4. (a) Experimentally measured light intensity profile against different PD horizontal offsets. Error distributions of the actual room (b) trained by experimental training data set and evaluated by experimental testing data set; (c) trained by DIALux simulation training data set and evaluated by experimental testing data set.
Figure 4(b) shows the experimental error distributions of the actual room trained by the experimental training data set and evaluated by the experimental testing data set. The red dots are the testing locations and the black circles indicate the maximum position error. The average position error is 10.5 cm. Then, we train the LRML VLP model by using the simulation training data obtained from the DIALux program. Figure 4(c) shows the error distributions of the actual room trained by the simulation training data set and evaluated by the experimental testing data set. The average position error is 11.1 cm. Results show that average position errors and the error distributions of the model trained via DIALux simulation and trained via experimental data are match with each other. It is also worth to note in Figs. 4(b) and (c) that both models have the highest position errors at the center of the room. Since the room has the dimensions of 200 cm × 155 cm and the emitted light intensity profile will drop to 0 at the horizontal offset distance of 150 cm as shown in Fig. 4(a); the received optical signal strength from each LED is weak at the center of the room when the 4 LEDs are at the 4 corners of the room ceiling. As a result, the position error is high at the center of the room. The position error could be reduced by using the LED lamp with higher optical power or wider emission angle. Reducing the size of the unit cell could also reduce the position error at the center. Figures 5(a) and (b) show the experimental cumulative distribution function (CDF) of the position errors based on the actual room trained and evaluated by experimental data; and trained by DIALux simulation and evaluated by experimental data. Results show that 80% of the testing locations with position error < 17 cm in both models.
Fig. 5. CDF of the actual room (b) trained by experimental training data set and evaluated by experimental testing data set; (c) trained by DIALux simulation training data set and evaluated by experimental testing data set.
4. Conclusion
Here, we proposed and demonstrated using the DIALux software with our proposed LRML algorithm for designing a practical and accurate indoor VLP system. The DIALux software is popular, user friendly. It is free of charge with a variety of LED luminary emission profiles for selection; hence, the cost of VLP system design can be minimized. Experimental results showed that the average position errors of the model trained by experimental data and DIALux simulation data were 10.5 cm and 11.1 cm, respectively. Besides, 80% of the CDF testing locations had position error < 17 cm in both models. The results revealed that average position errors and the error distributions of the model trained via DIALux simulation and trained via experimental data are match with each other. This implies that training data can be generated in DIALux if the room dimensions and LED luminary parameters are available. The proposed scheme could be a promising tool for designing IPS, as well as relieving the burden of training data collection in VLP systems.
Funding
Ministry of Science and Technology, Taiwan (MOST-109-2221-E-009-155-MY3).
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. H. L. Minh, D. O’Brien, G. Faulkner, L. Zeng, K. Lee, D. Jung, Y. J. Oh, and E. T. Won, “100-Mb/s NRZ visible light communications using a post-equalized white LED,” IEEE Photonics Technol. Lett. 21(15), 1063–1065 (2009). [CrossRef]
2. Z. Wang, C. Yu, W. D. Zhong, J. Chen, and W. Chen, “Performance of a novel LED lamp arrangement to reduce SNR fluctuation for multi-user visible light communication systems,” Opt. Express 20(4), 4564–4573 (2012). [CrossRef]
3. H. H. Lu, Y. P. Lin, P. Y. Wu, C. Y. Chen, M. C. Chen, and T. W. Jhang, “A multiple-input-multiple-output visible light communication system based on VCSELs and spatial light modulators,” Opt. Express 22(3), 3468–3474 (2014). [CrossRef]
4. B. Janjua, H. M. Oubei, J. R. Durán Retamal, T. K. Ng, C. T. Tsai, H. Y. Wang, Y. C. Chi, H. C. Kuo, G. R. Lin, J. H. He, and B. S. Ooi, “Going beyond 4 Gbps data rate by employing RGB laser diodes for visible light communication,” Opt. Express 23(14), 18746–18753 (2015). [CrossRef]
5. C. W. Chow, C. H. Yeh, Y. Liu, Y. Lai, L. Y. Wei, C. W. Hsu, G. H. Chen, X. L. Liao, and K. H. Lin, “Enabling techniques for optical wireless communication systems,” Proc. OFC 2020, paper M2F.1.
6. J. Luo, L. Fan, and H. Li, “Indoor positioning systems based on visible light communication: state of the art,” IEEE Commun. Surv. Tutorials 19(4), 2871–2893 (2017). [CrossRef]
7. C. Xie, W. Guan, Y. Wu, L. Fang, and Y. Cai, “The LED-ID detection and recognition method based on visible light positioning using proximity method,” IEEE Photonics J. 10(2), 1–16 (2018). [CrossRef]
8. S. Cincotta, A. Neild, C. He, and J. Armstrong, “Visible light positioning using an aperture and a quadrant photodiode,” IEEE Globecom Workshops, 2017.
9. C. Y. Hong, Y. C. Wu, Y. Liu, C. W. Chow, C. H. Yeh, K. L. Hsu, D. C. Lin, X. L. Liao, K. H. Lin, and Y. Y. Chen, “Angle-of-arrival (AOA) visible light positioning (VLP) system using solar cells with third-order regression and ridge regression algorithms,” IEEE Photonics J. 12(3), 1–5 (2020). [CrossRef]
10. T. Q. Wang, Y. A. Sekercioglu, A. Neild, and J. Armstrong, “Position accuracy of time-of-arrival based ranging using visible light with application in indoor localization systems,” J. Lightwave Technol. 31(20), 3302–3308 (2013). [CrossRef]
11. P. F. Du, S. Zhang, C. Chen, A. Alphones, and W. D. Zhong, “Demonstration of a low-complexity indoor visible light positioning system using an enhanced TDOA scheme,” IEEE Photonics J. 10(4), 1–10 (2018). [CrossRef]
12. H. S. Kim, D. R. Kim, S. H. Yang, Y. H. Son, and S. K. Han, “An indoor visible light communication positioning system using a RF carrier allocation technique,” J. Lightwave Technol. 31(1), 134–144 (2013). [CrossRef]
13. C. W. Hsu, J. T. Wu, H. Y. Wang, C. W. Chow, C. H. Lee, M. T. Chu, and C. H. Yeh, “Visible light positioning and lighting based on identity positioning and RF carrier allocation technique using a solar cell receiver,” IEEE Photonics J. 8(4), 1–7 (2016). [CrossRef]
14. Y. C. Chuang, Z. Q. Li, C. W. Hsu, Y. Liu, and C. W. Chow, “Visible light communication and positioning using positioning cells and machine learning algorithms,” Opt. Express 27(11), 16377–16383 (2019). [CrossRef]
15. J. He, C. Hsu, Q. Zhou, M. Tang, S. Fu, D. Liu, L. Deng, and G. Chang, “Demonstration of high precision 3D indoor positioning system based on two-layer ANN machine learning technique,” Proc. OFC 2019, Paper Th3I.2.
16. Y. C. Wu, C. W. Chow, Y. Liu, Y. S. Lin, C. Y. Hong, D. C. Lin, S. H. Song, and C. H. Yeh, “Received-signal-strength (RSS) based 3D visible-light-positioning (VLP) system using kernel ridge regression machine learning algorithm with sigmoid function data preprocessing method,” IEEE Access 8, 214269–214281 (2020). [CrossRef]
