400 m Rolling Shutter based Optical Camera Communications Link

,


INTRODUCTION
Optical camera communications (OCC) is considered as a pragmatic form of visible light communications (VLC), which utilizes image sensor (IS) (i.e., cameras) as the receiver (Rx) and lightemitting diodes (LEDs), laser diodes (LDs) and liquid crystal displays as the transmitter (Tx) [1,2]. In OCC, the camera captures two-dimensional data in the form of image sequences, thus enabling multidimensional data transmission over the free space channel. OCC offers multiple functionalities of vision, data communications and localization, which can be used in a number of applications including all-optical Internet of things (OIoT) [3,4]. OIoT-based applications include device-to-device communications, mobile attocells, vehicle-to-everything (V2X), smart environments, etc. [2], which releases the needed radio frequency spectrum for use in areas most required. Inspired by IoT, the internet of vehicles (IoV) is currently of high interest within the research community [4]. Recently, the widespread use of LEDs as taillights, brake lights, headlights, street and traffic lights has opened up the potential opportunities for implementation of the intelligent transportation systems (ITS) to mitigate the traffic congestion and therefore improve the quality of life and the economy [2]. In addition, the availability of cameras in traffic networks, security surveillance, vehicles, etc., can be effectively exploited as an optical Rx, thus enabling the implementation of VLC-OCC links.
One of the advantages of camera-based Rxs over the photodiodes is the higher signal to noise ratio (SNR) due to longer exposure time (T exp ) and larger overall photosensitive area, which results in longer link distances. OCC can be deployed in ITS for vehicle to vehicle, infrastructure-to-vehicle and vehicle-toinfrastructure (V2I) communications, therefore a long link-span helps to establish a robust system [2,5]. The complementarymetal-oxide-semiconductor (CMOS)-based rolling shutter (RS) camera, which sequentially (row-by-row) integrates light illuminating the pixels thus operating similarly as scanning function, can be used to increase the data rate (R b ) higher than the frame rate (R f ) of the camera [6] . In [7], global shutter (GS) based OCC with a R b of 15 bps over a link range of 328 m was reported for use in smart city applications with ∼4% wrongly decoded received bit streams. While in [8], a RS-based OCC link for the outdoor application with a maximum link span of 120 m, a bit error rate (BER) of > 10 −2 (at 120 m) and an achieved maximum R b of 200 bps at a 4 m link span was reported. Further investigations on long-distance OCC links (beyond 120 m) using RS is yet to be investigated to the author's best knowledge.
In order to establish long distance VLC there are a number of options including large area light sources with high power (meeting the eye-safety), multiple light sources, wide area optical Rx (at cost of reduced bandwidth) and multiple Rxs (i.e., cameras). In [7], a large size Tx with a dimension of 48 × 48 cm 2 was used. However, using a large Tx is impractical, therefore in this paper, we propose a novel reception technique in order to increase the transmission link-span of RS-based OCC by reducing camera's spatial bandwidth in the out-of-focus areas. This helps to have a larger footprint of the light source on the IS without reducing the field of view (FoV). We also develop a detection algorithm to effectively extract the received information from captured video frames. Previous works reported on OCC have mainly used cameras in the focused mode. However, cameras can also be used in their defocused modes depending on applications e.g indoor non-line-of-sight communications or V2I-VLC [9]. We have used a light source (i.e., 2.5 × 2.5 cm 2 ), which is 19 times smaller than the source that was adopted in [7]. The proposed system is attractive in distance critical applications requiring relatively low R b as in ITS (e.g., exchange of safety and traffic messages and positioning-related information) for smart traffic management. In this work, we achieve a communication link of up to 400 m with 100% success rate in data transmission achieving a R b of 450 bps. The demonstrated system, to the best of our knowledge, is the longest link-span achieved for a RS-based OCC link.

SYSTEM MODEL AND DETECTION
The block diagram of the proposed system is shown in Fig. 1. At the Tx, an on-off keying non-return to zero (OOK-NRZ) data s(t) is used for intensity modulation of the LED (a 2.5 × 2.5 cm 2 size chip on board (COB)) via the LED driver. The intensitymodulated light x(t) is transmitted over a free space channel and is captured at the Rx using a CMOS RS camera with a 1000 mm telephoto lens. For the line of sight (LOS) link, the received signal is given by [10]: where h(t) is the combined impulse response of the channel and camera, η is the quantum efficiency of the IS and n(t) is the additive white Gaussian noise including the ambient lightinduced shot noise and the noise in the camera (i.e, fixed pattern, thermal, photocurrent shot and flicker noise sources) [11]. The channel DC gain for the LOS link can be obtained as [10]: where A Tx-img is the area of the projected illuminated light source on the IS, d L is the distance between the Tx and the camera's lens, T s (θ) and g(θ) are the gains of the optical filter and optical concentrator, respectively. Φ denotes the irradiance angle, ξ is the FoV semi-angle of the camera and m represents Lambertian order of emission of the Tx. The incidence

Fig. 1. Schematic block diagram of the long distance OCC link
Usually, CMOS camera sensors (adopted as the Rx in this work) uses the RS readout method, such that each pixel row is exposed in a row-by-row sequential manner with a fixed time delay (row readout time T r ). Moreover, in RS-based cameras, the exposure of each row takes place before the readout and not all at once as in the case of GS-based cameras. Consequently, each pixel row's exposure does not commence at the same time. This is an advantage in OCC systems, which can lead to increased R b higher than the camera's R f , but at the cost of reduced coverage distance, hence we propose a new technique to extend this coverage distance. Note that with RS-OCC, a flicker free transmission is achievable hence we employ this technique in this work. Note that, in OCC the captured Tx's focused image size decreases with the increasing link span as given by the relationship where d Tx and d img are the diameter of the Tx and the diameter of the projected Tx's image on the IS at the focal length f of the lens in use, respectively. This therefore, constitutes a limitation for RS-based OCC links as the received signal area on the IS of the camera, which determines the number of rows N rows (i.e the ON and OFF states of the Tx) obtainable decreases with the increasing transmission distance. Consequently, we reduce this limitation by operating the camera in its outof-focus (defocused) mode. Fig. 2 shows a lens, object and IS configuration. The IS could be moved from the position 1 to 2 where varying sizes of the projected object's image is obtainable. Importantly, we use the defocusing feature of the camera i.e., by altering the distance between the 1000 mm telephoto lens and the IS, to allow the Tx's image to converge beyond the focal point, whereby a larger footprint of the Tx is obtained. that, defocusing of the camera results in a disc-shaped pattern known as a circle of confusion (CoC), which is convoluted with the image as given by [12]: where, G o (x, y) and G i (x, y) are the defocused and focused image intensity functions, respectively. is the 2 dimensional convolution operator and G CoC (x, y) is the CoC disc function, which is the same shape as the camera's lens aperture. For a circular aperture, we have [12]: (4) and the diameter of CoC, D CoC , is the same as the width of the defocused image of a point source and is given as [12]: where U(.) is Heaviside step function, D lens = f / f stop , is the diameter of the lens aperture, f stop is the focal stop number of the lens aperture, d IS is the distance between the centre of the lens to the IS and d c represents the distance between the centre of the lens and the image, which can be obtained as [12]: Note that, the size of CoC depends on the aperture diameter D lens for collecting the light rays of d IS and d c . Consequently, to increase defocusing (i.e., D CoC ) in order to have the best signal area, |d c − d IS | must be maximised and lower values of f stop should be used to obtain larger D lens . However, the amount of achievable defocusing is limited to the camera's optics design configuration. Fig. 4 illustrates the N rows as a function of the transmission distances for focused and three different defocused image modes for d IS of f , 0.3 f , 0.5 f and 0.7 f , f = 1000 mm and D lens = f /10 and width of rows = 0.05 cm. The N rows increases with d IS . E.g., at d L of 20 m the N rows has increased from 2 for the focused image mode to 33, 52 and 71 for d IS of 0.7 f , 0.5 f and 0.3 f , respectively thus enabling longer RS-OCC link spans. The width of one row in pixel (representing one bit or symbol for a fixed camera pixel clock) [13], w b = 1/(2 f Tx T r ), where f Tx denotes the Tx's switching frequency. Note, the N rows increasing with the defocusing is at the cost of reduced light intensity level per pixel (i.e., lower peak SNR (PSNR)/pixel) since the received light is spread over a higher number of pixels. Interestingly, the latter can be compensated for by increasing the gain parameter of the camera (i.e. its sensitivity), the aperture size and T exp (within the required bandwidth) so as to capture more light but only to the extent that the rows still remains distinctive and not mixed-up. In addition, a robust image processing algorithm, see Fig. 5, is proposed to enhance the success rate of received bits.
Consequently, at the Rx side, the output of the camera (IS) is processed off-line in MATLAB. As portrayed in the detection flowchart (Fig. 5), every video frame is converted from the red, green and blue (RGB) colour format to the grayscale for both the calibration and data videos after pixelation (i.e., digitizing the image to obtain the intensity/pixel value). The data videos are the captured transmitted data whereas the captured calibration videos are the template shape of the Tx, which is used for equalisation or otherwise described as the intensity compensation of the data video frames. Next, the region of interest (ROI) is selected (i.e., the footprint of the light source on the IS, which is the CoC) and then averaged over the rows to form a column vector. In order to avoid noise amplification at the start and end of the pixel rows in each frame, at least 10-pixel rows are eliminated from the top and bottom of the CoC. The received signal is then up-sampled to increase the resolution of the received signal. The correlation between the transmitted and received signal is carried out to determine the delay between them in order to extract the required received samples. A matched filter (MF) (via postprocessing in Matlab) is applied to recover the data. The recovered data bits vector is then compared with the transmitted data to ascertain the success rate of received bits by determining the ratio of the wrongly decoded bits to the total number of transmitted bits (i.e BER).

EXPERIMENTAL SETUP AND RESULTS
The system configuration and experimental setup for the longdistance OCC are shown in Fig. 6. The Tx was located on the 6th floor (height of 25 m) and the Rx was placed outdoor at a height of 1.40 m above the ground. At the start of the measurement campaign, the weather was partly cloudy/sunny while by the end it was sunny, with a temperature range of 23 -25 o C and the wind speed and humidity of 2 -6 mph and 51 -65%, respectively.The key experimental parameters include The Tx data packet is 18 bits, the Rx's R f is set to 25 fps for all experimental configurations considered and the number of received data bits is ∼32 per frame. The Rx's video resolution, pixel size, lens aperture and f are 648 × 484 (RGB32), 2.2 × 2.2 µm 2 , f /10 and 1000 mm, respectively. Measurements were carried out for a range of link spans L s and T exp of 150 to 400 m and 100 to 800 µs, respectively. Image frames of the transmitted data were captured with up to 100% success rate of received bits for all links considered in this work. In order to quantify the link performance for each L s , we used the image quality metric of PSNR, which is given as [14]: where I max is the maximum possible pixel value, I max = 2 n − 1, n = 8 for a grayscale image, and (i) is the pixel luminance mean squared error, which is defined by: I Tx (i) and I Rx (i) are the difference between pixel and average pixel values for transmitted and received symbols (1 and 0), respectively, while i represents the row's index number. Fig. 7 shows the PSNR versus L s for a range of T exp . As illustrated, PSNR improves with T exp and decreases with the increasing L s e.g., for T exp of 800 µs, PSNR drops by 2.1 dB when increasing L s from 200 to 400 m. The maximum values of PSNR are 3.3 and 0.2 dB for T exp of 800 µs and 100 µs at L s of 150 m.

CONCLUSION
We have developed a novel technique to increase the link-span of RS-based OCC by reducing the spatial bandwidth of the camera in the out of focus regions. The experimental analysis of the proposed scheme demonstrates a 100% success rate of received bits for a L s of up to 400 m using a small surface area Tx of 2.5 × 2.5 cm 2 . The choice of the T exp played a key role in determining the value of PSNR. To the best of authors' knowledge, no works have been reported for long distance OCC links using RS.

Funding Information
The European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant (764461 (VISION)) and Scientific Research Instrument and Equipment CAS, China (YJKYYQ20170052).

Disclosures
The authors declare no conflicts of interest.