1. Introduction

Optical wireless communication (OWC) is an attractive alternative to radio-frequency (RF) links due to its various advantages. The technical merit of laser communications is derived from the fact that it offers a much higher collimated signal than conventional microwave [1–3].This super-collimated beam can result in a terminal design with greatly reduce size, weight and transmitter power. Furthermore, laser communication systems are not susceptible to radio frequency interference and are not subject to frequency or bandwidth regulation.

One of the most significant factors that limit the performance of OWC is the effects of turbulent atmosphere [3–6]. When a light beam propagates through a turbulent medium, Atmospheric turbulence results in phase distortions, optical scintillations, beam wander, and beam spreading [4, 6–7]. Scintillation effects can be reduced through aperture averaging by using the relatively large aperture receivers [8], and an active tracking system can mitigate the effects of beam wander [9]. However, phase distortions limit the extent of focusing capabilities of the receiving telescopes, which break up an incoming plane wave into several incoherent random modes [7, 10]. The magnitudes of the distortions depend on the strength of turbulence and the length of the optical path. Turbulence-induced random fluctuations of the signal intensity at the receiver aperture generate the corresponding increases in the receiver’s average point-spread function (PSF) in the focal plane [10–11]. Hence, in the absence of optical turbulence, a detector with the diffraction-limited spot size is sufficient to collect most of the signal energy. However, in the presence of atmospheric turbulence, optical turbulence distorts the phase of the propagating optical fields and limits the extent of focusing capabilities of the antennas. The focal spot size, which limited by the atmospheric Fried parameter, is much greater than the one of ideal diffraction-limited case. In this regime, to capture the widespread signal energy, a larger size detector array is required [7, 10–11]. As we increase the detector dimensions to adapt the system to the energy spread of the received beam, the required field-of-view (FOV) of the receiver also increases. Not surprisingly, due to the larger FOV, some terminals can be within the same FOV of one terminal receiving telescope [12–13]. Thus, another significant advantage of the detector-array scheme is the ability to receive a number of different signals simultaneously, because the detector array has a FOV large enough to include multiple terminals, in which signals arriving from different terminals are detected by different pixels. The focused beam location of the received signals is defined by a spatial shift, the shift is determined by the spatial locations of the terminals, and then we can estimate signal and noise level in the each pixel associated with each terminal.

Clearly, a large receiver FOV results in a corresponding increase in the number of background photons received [7, 14]. To mitigate the effects of the background noise [15], and obtain simultaneously multiple signals in atmospheric turbulence, our approach uses code division multiple access (CDMA) techniques [16–17] for the detector-array scheme. This paper focuses on the scenario, one of the important constraints for optical networks, that there is a simultaneous communication from a number of terminals to one master terminal. The dimensions of the detector array located on the master terminal must be made large enough to encompass the signal energy and include multiple terminals. In this scheme, each signal from one of a cluster of terminals is modulated by its own unique pseudo-random (PN) sequence, and a common optical system in the master terminal maps these incoming signals onto a high-density, high-speed 2-D detector array. The problem of detecting the scattered optical fields with the detector array has been addressed in [7, 18]. However, these papers considered only Gaussian channel approximation, and, unlike here, where the knowledge of the amplitudes of signal and noise across the array is needed. In this paper, a closed-form solution for calculating the bit-error rate (BER) of multiple signals over atmospheric turbulence channels is proposed, which is based on the singular value decomposition [19–20] using the sample covariance matrix of the received signals and Meijer G-function. This scheme proposed in this paper simultaneously estimates the signals from multiple asynchronous terminals rather than using a number of independent gimbaled telescopes and greatly reduces the need for precise pointing [13], and the intrinsic interference-tolerance of CDMA systems allows terminals to send unscheduled traffic at any time without regard for whether other terminals are also transmitting at the same time.

The remainder of the paper is organized as follows. Section 2 describes the system model that we proposed. Numeric results are shown in Section 3. Finally, conclusions are given in Section 4.

2. System model

2.1 Characterization of turbulence-induced optical distortion

Atmospheric turbulence, generated by the temperature and pressure changes in the atmosphere, gives rise to the random fluctuations in the atmospheric refractive index called the optical turbulence. Optical turbulence inflicts the deleterious effects on the propagation optical signals [10–11]. The strength of the optical turbulence is typically represented by the so-called atmospheric coherence diameter or the Fried parameter $r_0$ [21]. The stronger turbulence the atmosphere gets, the smaller the$r_0$will be. In the diffraction-limited case, the incoming plane waves is focused into a focal-spot size of $2.44f\lambda /D$, where f and D represent the focal length and the aperture diameter of the receiving telescope, respectively. However, as a beam propagates through atmospheric turbulence, the focal spot size is limited by the Fried parameter$r_0$, and is given by$2.44f\lambda /r_0$ [10], assuming that the collection aperture diameter$D>r_0$. The turbulence causes an increase in the spot size and a random distribution of the signal energy into ${(D/r_0)}^2$spatial modes in the detector plane. In this regime, to capture the same amount of the signal energy as that of the diffraction limited case, a larger detector diameter for collecting the signal is needed.

The effects of the atmospheric turbulence alter the refractive index in a statistical manner due to temperature fluctuations. The associated power spectral density can be described by von Karman spectrum for describing the spatial variations of the refractive index [10],

For optical wave propagation, what we can obtain is a statistical description of the optical variations, because the inhomogeneity of atmospheric turbulence is random. Based on the form of the extended Rytov theory [10], the received irradiance I can be expressed as the product of two statistically independent random processes$I_x$and$I_y$, that is $I=I_x{I}_y$, where$I_x$and$I_y$arise from the large-scale and small-scale turbulent eddies, respectively. Based on the assumption that both the large and small scale effects follow a gamma distribution, the so-called gamma-gamma probability density function (pdf) is therefore derived as [23]

2.2 Atmospheric BER Estimator

In this paper, we focus on the scenario that there is simultaneous communication from a number of terminals to one master terminal. For convenience, we investigate the scenario that each of terminal points to the master one, while the signals are received by the latter. Let us assume that the signals from each terminal in the space within the terminal cluster can be a spatial δ-function with temporal-spatial intensity distribution$S(\overrightarrow{R},t)$with $\overrightarrow{R}$a point in the space. This situation is formulated as

For the lth terminal, the modulating waveform $S_l(t)$forms through multiplying the spreading waveform$a_l{}^{}(t)$by the data waveform$d_l(t)$. If${\alpha }_{lj}^{}$,$j=0,\cdots ,N-1$, are the elements of the periodic pseudorandom sequence ${\mathrm{\Lambda }}_l=({\alpha }_{l0,}\cdots ,{\alpha }_{l(N-1)})$ with a period N associated with the lth terminal, i.e., for every j,${\alpha }_{lj}^{}={\alpha }_{l(j+N)}^{}$, where${\alpha }_{lj}$($j=0,\cdots ,N-1$) is either 0 or 1, then the spreading waveform $a_l{}^{}(t)$ can be expressed as

${\mathrm{\Lambda }}_l$is usually generated by a shift register with the exclusive OR [25]. Mathematically, the set (1, 0) is transferred to the set (1,-1) by${\beta }_{lj}=2{\alpha }_{lj}-1$, where${\beta }_{lj}=1$or$-1$. So that for any sequence${\mathrm{\Lambda }}_l$and${\mathrm{\Gamma }}_l$, the autocorrelation$R_{{\mathrm{\Gamma }}_l,{\mathrm{\Gamma }}_l}(q)$and cross-correlation functions$R_{{\mathrm{\Lambda }}_l,{\mathrm{\Gamma }}_l}(q)$are written, respectively, as

In the receiver site, as shown in Fig. 1 , each terminal’s optical transmitter is formed on a cluster of pixels, and each of which contains the signals of the transmitters and whose output photocurrents can be combined to detect the transmitted signals [7].

 

Fig. 1 Multi-signal receiver for optical communication system

Download Full Size | PDF

Without loss of generality, we assume a slow-varying channel, where the signal intensity is assumed to be constant for a large number of data symbols [26]. Here, the free-space channel from the lth transmitter to the mth pixel can be described as a linear time-invariant attenuation channel having impulse response$h_{l,m}$in the observation. Hereafter, at pixel m, the received signals can be written as

In order to estimate the signal-to-noise ratio (SNR), a code-matched filter containing the specified address code vector is applied to$y(n)$. Then the output vector z at the nth bit can be written as$z_l=y(n){\mathrm{\Gamma }}_l$. The cross correlation between${\mathrm{\Gamma }}_l$and${\mathrm{\Lambda }}_l$is given by (9), and then, after taking P successive frames of the received signals, the$N\times P$data matrix is formed as

In (12), the columns of$U_s$span the signal subspace uniquely associated with O, and the columns of$U_n$span the noise subspace. The diagonal matrix ${\displaystyle \sum }=diag({\zeta }_i)$ contains the corresponding eigenvalues, which in the nonincreasingly order are given by${\zeta }_1\ge {\zeta }_2\ge \cdots \ge {\zeta }_{NP}$. The eigenvalues of O have the following structure

Thus, the signal subspace can be identified. Using (13), the observation noise, the interference power and the desired signal power are estimated, respectively, as follows

Consequently, the estimate of the SNR associated with terminal l on the detector array is then obtained as

It can be shown that the BER of the On-Off CDMA is equal to the following [9]$p_e(I)=\frac{1}{2}erfc(\sqrt{\frac{{\tilde{\xi }}_l}{2}}I)$, where$erfc(\cdot )$denotes the complimentary error function. The average BER over the gamma-gamma channel can be obtained by averaging $p_e(I)$ over the fading coefficientI, i.e., $p_e={\displaystyle {\int }_0^{\infty }f(I)}{p}_e(I)dI.$

In order to solve the integral, we express $erfc(\cdot )$ in terms of the Meijer G-function [24]$erfc(x)={\pi }^{-1/2}{G}_{1,}^{2,}{}_2^0\left[x^2|\begin{array}{c}1\\ \begin{array}{cc}0& 1/2\end{array}\end{array}\right]$ . Further, by applying the formula in [24], the average BER can be written as a closed-form solution

Note that Meijer G-function is a standard function and has been embedded into most of the well-known mathematical software packages such as Mathematica and Maple. Thus, using the above closed mathematical forms as shown in (16), we can obtain a criterion to appraise the system performance efficiently.

3. Numerical results

In this section, we show some numerical results using the following parameters:$L=4$,$N=7$and$\lambda =850\text{nm}$for all figures. The array output located in the master terminal is collected in parallel over the observation time and processed for signal detection. A $256\times 256$-pixel uniform array matched to the FOV of the receiver is used where the distance between adjacent pixels is one half the wavelength. At the transmitter side, a modulator modulated the laser with pseudorandom M-sequence code [25]. For the reception procedure, the incoming signals, which are the superposition of all the transmitted ones, are collected by a telescope, detected by a detector array, and then processed for the decision.

We will first compare the derived BER based on (16) with the system simulations over the gamma-gamma channel. Figure 2 presents the BER of optical networks as a function of SNR, for three different Rytov variances${\sigma }_R^2$of 0.5, 1, and 2. It is clear from Fig. 2 that the derived error probability (plotted in dashed style) provides a good approximation and coincides with that of simulation results for all the cases depicted here, indicating that the estimator is robust to the effects of turbulence when the optical beam is subjected to gamma-gamma irradiance distribution model. Figure 3 shows the estimate error versus the received SNR over a chip interval with$T_C$for${\sigma }_R^2=0.5$. We can see that the system can realize communications when received SNR is smaller than or equal to 0.1, because the system employs CDMA instead of uncoded binary date. Hence, there is no need to adapt the threshold. From Fig. 3, we can clearly see that the algorithm presented dramatically decreases the estimate error when received SNR increases.

 

Fig. 2 BER comparison of the theoretical and simulation results

Download Full Size | PDF

 

Fig. 3 Estimate error versus the received SNR over a chip interval

Download Full Size | PDF

To study the effect of the frame-number sampled on estimator performance, in Fig. 4 , we plot the estimated SNR as a function of the number of frames where actual SNR is 8 dB and${\sigma }_R^2$is 0.5. The estimated SNR curve (shown with a “+”) is seen to converge to its actual one (shown with an “o”) for a large number of samples, validating the analysis model. Clearly, the deviation of the estimator dramatically decreases as the estimated time increases. We can see that the percentage error falls to below 20% after about 10 frames, and to below 10% after less than 18 frames.

 

Fig. 4 Estimated SNR versus the number of frames

Download Full Size | PDF

4. Conclusion

In this paper, the computational model of estimating the BER with a detector array for OWC systems over atmospheric turbulence channels is proposed, and performance of this model is examined. In this method, the observation space is separated into two orthogonal subspaces, namely, the signal subspace and the noise subspace, by applying the singular value decomposition on the covariance matrix of the received signals after the code-matched filter. Further, based on Meijer G-function, the closed-form solution of the BER is derived. The comparison between the theoretical and simulation results based on the gamma-gamma channel model indicates that such a model is adequate in estimating the BER.