An effective interference suppression algorithm for visible light communication system based on DBSCAN Download: 880次
Visible light communication (VLC) has excellent features such as low cost and absolute safety, as well as large-area support for a light emitting diode (LED), which makes VLC systems hopeful to become an alternative to next-generation wireless communication technologies. Today, many VLC-based products have been proposed, including the wireless communication network light fidelity (Li-Fi) with high security[1], indoor positioning[2], vehicular communications[3], underwater VLC[4], and so on. Most of these products use photodiodes (PDs) as receivers (Rx) to improve the data transmission rate[5,6]. However, some groups believe that mobile phone cameras can also be used as VLC Rx[7–
However, in the process of receiving optical signals by using a mobile phone camera, there will be a variety of noises. Especially, at high transmission speeds, the signal-to-noise ratio and communication quality are greatly reduced. Therefore, how to design an effective algorithm to eliminate the interference in the images becomes important.
In the problem of denoising and decoding, many methods[8,10,12,13] have been proposed. Most of them used a filter to denoise and set thresholds for decoding. This method can effectively remove high-frequency interference in the case of low bandwidth, but when the bandwidth becomes larger, a filter will make a loss. However, many solutions based on machine learning have recently been used in VLC. Li et al.[14] used a support vector machine (SVM) and artificial neural networks (ANNs) to help improve the LED-ID recognition rate and its performance. Hao et al.[15] utilized an end-to-end learning network to improve the VLC system performance. Also, Lu et al.[16,17] utilized the density-based spatial clustering of applications with noise (DBSCAN) algorithm in VLC systems, and the algorithm solved the amplitude jitter problem occurring during transmission and successfully improved the system performance. The machine learning adaptive algorithm can solve many problems in modulation format recognition and optical transmission equalization demodulation. Through the study of machine learning, we designed an effective interference suppression algorithm based on the improved DBSCAN algorithm, which can effectively solve the interference problem of the CMOS-VLC system.
In the CMOS-VLC system, the signal generator is generally divided into two types: the white LED transmitter (Tx) and the red–green–blue LED (RGB-LED) Tx. In this Letter, we focus on the latter.
Many papers[8,10,12,13] used different filters to sharpen and denoise the image according to different situations. Then, a second or third-order polynomial fitting is applied for setting the threshold. The threshold is to distinguish logic 1 and 0.
Figure
Fig. 1. (a), (b) Green channel distribution of images; (c), (d) green channel distribution of images after low-pass filter and the third-order polynomial fitting (green curve); (e), (f) the judgement through comparing with the threshold.
At the same time, this method will make some noise in the image. In the ideal case, the edges of three color channels of an optical signal should be coincident, as shown in Fig.
Fig. 2. (a) Ideal edge, (b) real edge, (c) part of the sample image, (d) processed image after judgement, and (e) image processed by the improved algorithm.
The DBSCAN algorithm is a relatively common unsupervised machine learning algorithm. It can automatically find clusters of arbitrary shape and can effectively find noise points and outliers, which make it become an excellent and efficient data clustering algorithm[18]. At the same time, the DBSCAN algorithm requires two parameters: neighborhood radius (Eps) and minimum points in the Eps neighborhood (MinPts)[18]. The former determines the maximum radius of the data point in the search for similar data points, and the latter determines the minimum number of data points that a cluster is not considered as noise. These two parameters generally require artificial settings, but some studies have proposed that the algorithm can adaptively find Eps and MinPts[19,20]. Since the payload data in the samples obtained in the CMOS-VLC system is far more than the noise data, it is very suitable to use the DBSCAN algorithm for interference reduction processing. The DBSCAN algorithm clustered elements by distance between them, but the relationship between pixels in the sample images is more than just distance, so this Letter designed an effective algorithm based on the traditional DBSCAN algorithm for a CMOS-VLC system.
In the images obtained by the CMOS-VLC system, in addition to the area occupied by the RGB-LED (the area enclosed by the white circle in the Fig.
The images
The threshold needs to be set in advance according to the overall brightness of the sample, generally set to 20–30. When , the difference between two pixels is too large, and the two pixels belong to different stripes. When , the two pixels are considered to be the same type of pixels. To facilitate understanding of the idea, Fig.
In the experiment, we use RGB-LED as the Tx and on–off keying (OOK) as the three-color modulation scheme. The implementation method is taking consecutive three bits of the binary sequence, and the binary sequence is used to control the on–off state of the RGB-LED, which can make the RGB-LED realize the modulation of eight colors. Implementation methods and examples can be seen in Tables
Table 1. Color of Binary Sequence
|
Table 2. Modulation Example
|
Fig. 5. Experimental results: (a), (d) unprocessed image; (b), (e) processed and analyzed image; (c), (f) processed image.
When the transmission rate is , the effect of the algorithm is more obvious than that of the transmission rate at . Figure
We took the value of the green channel of two original images and analyzed it. As shown in Figs.
Fig. 6. Experimental results: (a), (c) green channel distribution of unprocessed images; (b), (d) green channel distribution of processed images.
At present, almost all CMOS image sensors carry the Bayer filter[21], which will cause the zipper effect[21] when the color of pixels changes so quickly, as shown in Figs.
Fig. 7. (a), (d), (g) Sample images; (b), (e), (h) details of sample images; (c), (f), (i) images processed by the improved DBSCAN algorithm.
The improved DBSCAN algorithm needs to define the parameters Eps and MinPts at first. However, the effect of clustering has a close relationship with the parameters. In order to know what the relationship is, we use the bit error ratio (BER) to evaluate the performance with different parameters. The received samples are clustered using the improved DBSCAN algorithm and decoded according to the width and mean color of each cluster. Then, the result is compared with the original binary sequence to get the BER.
Table
Table 3. Experimental BER ( ) versus Different Eps and MinPts
|
As a result, the minimum BER is , when the optimum MinPts is 5, and Eps is 5, 7, 8, and 10, respectively. Due to the rolling shutter of the CMOS sensor, the width of the stripe is mainly affected by the transmission rate[7]. The DBSCAN algorithm is based on the density of the dataset, that is, the effect of the algorithm will not decrease when the widths of stripes remain with few changes. The optimum parameters can be used at the same transmission rate but in different distances. We set the Eps as 7 and MinPts as 5, then reset the binary sequence, and record the optical signals again after setting the new parameters. After setting the new parameters, BER is remeasured as . We use the same method to determine the parameters at the transmission rate of , , and . Then, we remeasure the BER. When the transmission rate is below , no error is observed. But, the BER increases to at the transmission rate of . It is because when the transmission rate increases, the area of valid data is the same as that of the noise data, and the improved DBSCAN algorithm cannot distinguish noise from payload data. Also, we use BER to evaluate the performance of setting the threshold at the transmission rate of . Due to the disadvantage mentioned at the beginning of the Letter, the BER of this method is . Therefore, the algorithm is considered to perform well at a transmission rate below .
In this Letter, we design an algorithm based on the DBSCAN algorithm in order to reduce the interference in the system. As far as we know, it is the first time that the DBSCAN algorithm has been successfully applied to the CMOS-VLC system. The experimental results show that the algorithm can greatly reduce the interference and perform well at transmission rate below . There are many other factors that may affect the algorithm and the selection of the optimal parameters. In the future, we will continue to optimize the algorithm to make it more adaptive to complex environments. All in all, our proposed scheme shows great potential for the CMOS-VLC system, and the system can be more effective by using the algorithm we designed.
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
Guojun Lu, Hongzhan Liu. An effective interference suppression algorithm for visible light communication system based on DBSCAN[J]. Chinese Optics Letters, 2020, 18(1): 011001.