【正文】 could send the same information on all the transmit elements of the array and use diversity bining at the camera to achieve large transmission ranges due to the SNR gain. Though the multiplexing and diversity techniques are similar in concept to those in RF MIMO systems [11]the visual MIMO channel with very different characteristics attributes certain unique behavior to the MIMO gains in these systems.In visual MIMO the perspective distortions in the visual channel dominate over some of the important properties of a RF wireless channel such as distance based attenuation and multipath fading. Though perspective distortions in visual channels are primarily distance dependent visual MIMO channels induce perspective distortions in the image even if the transmitter and receiver are aligned at an angle with respect to each other. Two images which are clearly separated in the image plane may look overlapped when viewed from an angle. Such distortions can depreciate the signal quality and the detection capability leading to errors and thus reduction in the data rates. Further lens blur (typically due to focus imperfection or jerks while capturing the image) also can significantly depreciate the image quality and thus reduce the information capacity. In this paper we will detail how MIMO techniques such as multiplexing and diversity are characterized based on the effect of perspective distortions in the visual MIMO channel. Based our channel model we will derive the analytical channel capacity of the visual MIMO channel and using the same we illustrate the significance of parameters such as distance, viewing angle and blur in characterizing multiplexing and diversity in visual MIMO.This paper is structured as follows。 j) representing the channel between the kth transmit element and pixel (i。 3 : : :K) emitting a light beam of power Pin。 j) from each transmit element k to the pixel(i。i。i。i。 j), characterized by hk(i。 jrefk ). Given the spatial coordinates of the transmitting elements of an LEA with respect to the camera reference we can determine the image center coordinates of those transmit element through optical raytracing techniques in conjunction with some basic puter vision theory [9].IV. PERSPECTIVE DEPENDENT MIMO GAINSWhile the channel model in (1) resembles that of the familiar RF MIMO channel model, in fact it is significantly different from that. In RF MIMO systems, the channel matrix is typically a rich scattering matrix (usually full rank) whose entries are modeled well as independent and identically distributed random variables [10]. Further, this property allows the RF MIMO system to exploit either diversity and or multiplexing gains in data transmission which primarily depend on the multipath fading in the RF channel. The fact that the munication system here uses light as the munication medium, requires line of sight at the receiver, and the nature of the concentration function of the camera, renders some unique multiplexing and diversity characterizations different from RF MIMO.A. Resolvability of imagesThe notion of ‘parallel’channels to obtain the multiplexing data rate gains can be achieved only if the circumference of two transmit elements as seen on the image plane are separated by atleast a threshold () number of pixels in both dimensions (horizontal and vertical). As we see in Fig. 2 even if the circumference of the two transmit element images are separated by one pixel they may not be resolvable because of the blur in the image. Hence we set a threshold distance of separation between the image circumferences, = 2p 2ln2blur, equal to the fullwidthhalf maximum (FWHM) of the Gaussian lensblur function typically used as a parameter for image resolution in analyzing fine detailed astronomical and medical images [4], [5]. The distance of separation between the images of the transmit elements can be determined by perspective projection analysis (as described in [6]) considering circular transmitting elements. Given a fixedfocal length f of the camera, pixel side length s and a spatial distance between the circumference of two adjacent LEDs, the circumference of two transmit element images will be separated by im = fds pixels in each dimension. Therefore the separation between thecircumference of two transmit element images will be equal to the threshold (=s) at a distance d = f between the LEA and camera. This implies that, multiplexing in visual MIMO is possible only when d  d and when d d each transmit element has to transmit the same information whereby diversity bining at the receiver can ensure an SNR gain and hence an equivalent capacity gain.B. Dependence on viewing angle We can observe in equation (3) that the channel quality depreciates with the viewing angle  (angle between the camera image plane and LEA surface plane). Two images which are clearly separated in the image plane may look overlapped when viewed from an angle. Such distortions can significantly depreciate the signal quality and the detection capability leading to errors and thus reduction in the data rates. Moreover such an angular view also reduces the achievable multiplexing transmission range. This is because when the camera image detector plane is at an angle  to the transmitter array the effective spatial separation between two neighboringtransmit elements bees (as shown inFig. 3). From the earlier discussion on the resolvability of images, it implies that the distance upto which multiplexing can be achieved in visual MIMO then reduces toC. Distance dependent MIMO gainsIn the visual MIMO channel, for a static transmitter and receiver, the image of the LEA transmit elements captured by the camera spans one pixel or multiple pixels. Further, the image plane is spanned by images of each transmit element clearly delineated and the size of image span depending on the focus (concentration ratio) of the camera. As illustrated in Fig. 4, at short distances