| 1 | \documentclass[a4paper,12pt,twocolumn]{article} |
| 2 | \usepackage{authblk} |
| 3 | \usepackage{siunitx} |
| 4 | \sisetup{group-digits=false} |
| 5 | \title{\SI{100}{GbE} Passive Optical Access Networks\\Technical Milestone Report} |
| 6 | \author{Adrian I.~Lam\vspace{-1em}\\Supervised by Dr.~Seb Savory} |
| 7 | \date{16 January 2019} |
| 8 | |
| 9 | % Generic formatting packages |
| 10 | \usepackage[T1]{fontenc} |
| 11 | \usepackage[utf8]{inputenc} |
| 12 | \usepackage{microtype} |
| 13 | \usepackage[margin=2.5cm]{geometry} |
| 14 | \usepackage{hyperref} |
| 15 | \hypersetup{unicode=true, |
| 16 | pdftitle={Technical Milestone Report}, |
| 17 | pdfauthor={ail30}, |
| 18 | pdfborder={0 0 0}, |
| 19 | breaklinks=true} |
| 20 | \usepackage{parskip} |
| 21 | \usepackage[title]{appendix} |
| 22 | |
| 23 | \usepackage[compact,small]{titlesec} |
| 24 | %\titlespacing{\section}{0pt}{0pt}{0pt} |
| 25 | %\titlespacing{\subsection}{0pt}{0pt}{0pt} |
| 26 | \usepackage{titling} |
| 27 | \setlength{\droptitle}{-8ex} |
| 28 | |
| 29 | \setlength{\columnsep}{0.5cm} |
| 30 | |
| 31 | % Math typesetting packages |
| 32 | \usepackage{amsmath} |
| 33 | \usepackage{amssymb} |
| 34 | \usepackage{physics} |
| 35 | \newcommand*{\imj}{\mathrm{j}} |
| 36 | |
| 37 | % Convenient referencing packages |
| 38 | \usepackage{cleveref} |
| 39 | \crefname{appsec}{Appendix}{Appendices} |
| 40 | |
| 41 | % Graphics-related packages |
| 42 | \usepackage[justification=centering,margin=0.5cm,font=small]{caption} |
| 43 | \usepackage{subcaption} |
| 44 | \usepackage{graphicx} |
| 45 | \usepackage{rotating} |
| 46 | |
| 47 | \usepackage{blox} |
| 48 | \usepackage{makecell} |
| 49 | |
| 50 | % code listings |
| 51 | \usepackage{upquote} |
| 52 | \usepackage{courier} |
| 53 | \usepackage{listings} |
| 54 | \makeatletter |
| 55 | \lstset{ |
| 56 | basicstyle=\ttfamily\lst@ifdisplaystyle\footnotesize\fi, |
| 57 | keywordstyle=\bfseries, |
| 58 | showstringspaces=false, |
| 59 | numbers=left |
| 60 | } |
| 61 | \makeatother |
| 62 | |
| 63 | % URL formatting, from <http://tex.stackexchange.com/a/194418> |
| 64 | \DeclareUrlCommand{\angleurl}{\def\UrlLeft{<}\def\UrlRight{>}\urlstyle{tt}} |
| 65 | \makeatletter |
| 66 | \DeclareRobustCommand*{\url}{\hyper@normalise\angleurl@} |
| 67 | \newcommand*{\angleurl@}[1]{\hyper@linkurl{\angleurl{#1}}{#1}} |
| 68 | \makeatother |
| 69 | |
| 70 | \newcommand*{\MATLAB}{MATLAB\textsuperscript{\textnormal{\textregistered{}}}} |
| 71 | |
| 72 | \renewcommand\floatpagefraction{.9} |
| 73 | \renewcommand\dblfloatpagefraction{.9} % for two column documents |
| 74 | \renewcommand\topfraction{.9} |
| 75 | \renewcommand\dbltopfraction{.9} % for two column documents |
| 76 | \renewcommand\bottomfraction{.9} |
| 77 | \renewcommand\textfraction{.1} |
| 78 | \setcounter{totalnumber}{50} |
| 79 | \setcounter{topnumber}{50} |
| 80 | \setcounter{bottomnumber}{50} |
| 81 | |
| 82 | |
| 83 | \begin{document} |
| 84 | \maketitle |
| 85 | |
| 86 | \begin{abstract} |
| 87 | This project aims at evaluating methods of achieving a \SI{100}{Gb/s} |
| 88 | Ethernet passive optical access network. An optical link based on |
| 89 | coherent receivers will be considered, and a computer model |
| 90 | will be built using \MATLAB{} to simulate various physical effects |
| 91 | in the optical fibre. Digital signal processing techniques will be |
| 92 | employed to correct for these effects and demodulate the transmitted |
| 93 | symbols. Currently, all relevant linear effects have been successfully |
| 94 | simulated and compensated for, while non-linear effects have not been |
| 95 | completed yet. After the model is complete, different designs of the |
| 96 | network, such as using different modulation schemes or multiplexing |
| 97 | methods, can be simulated and their performance compared, and the |
| 98 | feasibility to use them in an access network will be discussed. |
| 99 | The results may also be verified through off-line processing of |
| 100 | real measured data. |
| 101 | \end{abstract} |
| 102 | |
| 103 | \section{Introduction and Motivation} |
| 104 | A passive optical network (PON) is a point-to-multipoint system |
| 105 | where data for all users of the network, modulated onto an |
| 106 | optical signal, leaves the optical line terminal at the service |
| 107 | provider, and is carried along a fibre feeder, then split |
| 108 | by an unpowered beam splitter, without any routing or selection, |
| 109 | to separate distribution fibres reaching the optical network units |
| 110 | of the users~\cite[\S6.1]{PONintro}. This design allows high-speed |
| 111 | communication for a large number of consumers, with relatively low cost |
| 112 | for each user~\cite{NGPON2-1}, and is currently typically employed |
| 113 | in a fibre-to-the-home setting~\cite{ponroadmap}. |
| 114 | |
| 115 | Fibre-to-the-home applications are already well-served by the |
| 116 | currently mature implementation of \SI{1}{Gb/s} PONs, which may |
| 117 | make higher speed PON seem unnecessary. However, there are many more |
| 118 | future applications that could potentially benefit from a |
| 119 | \SI{100}{Gb/s} PON. By designing future PON specifications to be |
| 120 | compatible with existing fibre installations, less installation |
| 121 | costs will be incurred, allowing the mixing of |
| 122 | more applications onto the same network, thus increasing |
| 123 | revenues for service providers. In addition, as more and more |
| 124 | people rely on mobile networks for their daily communication |
| 125 | and entertainment needs, mobile operators are looking to increase |
| 126 | the density in cell sites, making PONs a good candidate to deliver |
| 127 | the cell data backhaul. As the 5G mobile standard develops, PONs |
| 128 | may even be useful in the fronthaul, where |
| 129 | radio signals were sampled and relayed, |
| 130 | through a PON, to a centralized |
| 131 | location for digital signal processing (DSP), seen as a way to |
| 132 | reduce costs~\cite{ponroadmap}. |
| 133 | |
| 134 | The use of coherent receivers in a high-speed PON is considered. |
| 135 | In contrast to direct detection receivers, both the real and imaginary |
| 136 | parts of each polarization of the received electrical field can be |
| 137 | detected separately |
| 138 | in a coherent receiver. This makes complex modulation schemes such |
| 139 | as phase-shift keying (PSK) or quadrature amplitude modulation |
| 140 | (QAM) possible, and combined with polarization-division multiplexing (PDM), |
| 141 | makes very high data rates possible~\cite[\S5.6]{foc},~\cite{savorydigital}. |
| 142 | |
| 143 | In this project, a model to simulate the various physical effects in |
| 144 | an optical channel is to be built. DSP will |
| 145 | then be used to attempt to correct for these effects to recover the |
| 146 | original signal. Using this model, different options for achieving a |
| 147 | \SI{100}{Gb/s} PON will be compared. The response of a real channel |
| 148 | will then be measured and, using the DSP techniques investigated, |
| 149 | processed offline to verify the model and the correction methods. |
| 150 | Feasibility of employing these techniques in a commercial PON setting |
| 151 | will also be discussed. |
| 152 | |
| 153 | The current progress in developing the simulation model is detailed |
| 154 | in \Cref{sec:simmodel}, with future plans listed in |
| 155 | \Cref{sec:future}. |
| 156 | |
| 157 | \section{The Simulation Model} \label{sec:simmodel} |
| 158 | |
| 159 | \begin{figure*}[tb] |
| 160 | \centering |
| 161 | \begin{tikzpicture} |
| 162 | \small |
| 163 | \bXInput[$x_n$]{input} |
| 164 | \bXBlocL[3]{p}{\makecell[c]{Pulse shaping\\$p(t)$}}{input} |
| 165 | |
| 166 | \bXBloc[3]{sim1}{Fibre-optic link}{p} |
| 167 | \bXLink[$x(t)$]{p}{sim1} |
| 168 | |
| 169 | \bXSumb*[6]{AWGN}{sim1} |
| 170 | \bXLink{sim1}{AWGN} |
| 171 | \path (AWGN) ++(0,-1) node (noise) {$n(t)$}; |
| 172 | \bXLink{noise}{AWGN} |
| 173 | |
| 174 | \bXOutput[3]{y}{AWGN} |
| 175 | \bXLink[$y(t)$]{AWGN}{y} |
| 176 | \end{tikzpicture} |
| 177 | |
| 178 | \begin{tikzpicture} |
| 179 | \small |
| 180 | \bXInput{yt} |
| 181 | \bXBloc[3]{q}{\makecell[c]{Matched filter\\$q(t)=p(-t)$}}{yt} |
| 182 | \bXLink[$y(t)$]{yt}{q} |
| 183 | \bXBloc[3]{sampler}{\makecell[c]{Sample\\$T_s=1/R_\text{sym}$}}{q} |
| 184 | \bXLink[$r(t)$]{q}{sampler} |
| 185 | \bXBloc[3]{sim2}{\makecell[c]{Channel\\equalization}}{sampler} |
| 186 | \bXLink[$r_n$]{sampler}{sim2} |
| 187 | \bXBlocL[3]{decision}{Decision}{sim2} |
| 188 | \bXOutput[2.5]{xhatn}{decision} |
| 189 | \bXLink{decision}{xhatn} |
| 190 | \path (xhatn) ++(0.3,0) node {$\hat{x}_n$}; |
| 191 | \end{tikzpicture} |
| 192 | \caption{Block diagram of the |
| 193 | simulation model.} |
| 194 | \label{fig:model} |
| 195 | \end{figure*} |
| 196 | |
| 197 | \Cref{fig:model} shows the current basic model, |
| 198 | involving a transmitter with a root-raised |
| 199 | cosine pulse shaping filter, processed to simulate the various |
| 200 | physical effects, then |
| 201 | transmitted through an additive white |
| 202 | Gaussian noise (AWGN) channel to a receiver with a matched filter. |
| 203 | The received signal is then sampled and DSP is used to correct for |
| 204 | the physical effects in the electrical domain. The demodulated signal |
| 205 | is then compared to the original pseudorandom data, to obtain a |
| 206 | measurement of the bit-error rate (BER) using a Monte-Carlo approach. |
| 207 | Currently, the main modulation scheme considered is quadrature |
| 208 | phase-shift keying (QPSK), with Gray coding. |
| 209 | |
| 210 | The effects considered are enumerated below. The results of the |
| 211 | methods used to correct for the effects are compared to the ideal |
| 212 | AWGN channel. |
| 213 | |
| 214 | \subsection{Chromatic Dispersion} \label{sec:CD} |
| 215 | Chromatic dispersion (CD) is the effect of the group speed of light varying |
| 216 | with the wavelength of the optical signal~\cite[\S2.7.3]{foc}. It can be |
| 217 | modelled as a linear system, with transfer function in the Fourier |
| 218 | domain |
| 219 | \[ |
| 220 | G(z, \omega) = \exp\left( -\imj \frac{D\lambda^2 z}{4\pi c} \omega^2\right) |
| 221 | \] or with impulse response in the time domain |
| 222 | \begin{equation} |
| 223 | g(z, t) = \sqrt{\frac{c}{\imj D \lambda^2 z}} |
| 224 | \exp\left( \imj \frac{\pi c}{D\lambda^2 z} t^2\right) |
| 225 | \label{eq:CDimpresp} |
| 226 | \end{equation} |
| 227 | with $z$ being the transmitted distance, $c$ the speed of light |
| 228 | in vacuum, $\lambda$ the wavelength in vacuum, and $D$ the dispersion |
| 229 | parameter of the fibre~\cite{savorydigital}. For all simulations |
| 230 | below, $D=\SI{17}{ps/(nm.km)}$. |
| 231 | |
| 232 | Using this model, constellation diagrams were obtained and |
| 233 | shown in \Cref{fig:CDconst}. It can be seen that over long |
| 234 | distances, CD would make demodulation very difficult, and as |
| 235 | such, it is necessary to compensate for this effect. Current |
| 236 | systems use dispersion compensating fibres, but DSP may be applied |
| 237 | instead to reduce cost~\cite{savorydigital}. It is noted that |
| 238 | by inverting the sign of $D$ in |
| 239 | \Cref{eq:CDimpresp}, the impulse response of the dispersion compensating |
| 240 | filter is obtained, and with truncation and discretization, |
| 241 | can be implemented as a simple tapped delay line~\cite{savorydigital}. |
| 242 | |
| 243 | \begin{figure}[htb] |
| 244 | \centering |
| 245 | \begin{subfigure}[t]{0.22\textwidth} |
| 246 | \includegraphics[width=\textwidth]{cd_qpsk_noiseless_Dz17_new.eps} |
| 247 | \caption{$z=\SI{1}{km}$.} |
| 248 | \end{subfigure} |
| 249 | \begin{subfigure}[t]{0.22\textwidth} |
| 250 | \includegraphics[width=\textwidth]{cd_qpsk_noiseless_Dz85_new.eps} |
| 251 | \caption{$z=\SI{5}{km}$.} |
| 252 | \end{subfigure} |
| 253 | \caption{QPSK constellation after chromatic dispersion, |
| 254 | without AWGN.} |
| 255 | \label{fig:CDconst} |
| 256 | \end{figure} |
| 257 | |
| 258 | \Cref{fig:CDCompz200} shows the dispersion compensating filter in |
| 259 | action. The resulting BER very closely resembles that of the ideal |
| 260 | AWGN, thus verifying the implementation. |
| 261 | |
| 262 | \begin{figure}[htb] |
| 263 | \centering |
| 264 | \includegraphics[width=.44\textwidth]{CDCompz200.eps} |
| 265 | \caption{QPSK signal with simulated chromatic dispersion and |
| 266 | CD compensation, over an AWGN channel, with |
| 267 | $z=\SI{200}{km}$.} |
| 268 | \label{fig:CDCompz200} |
| 269 | \end{figure} |
| 270 | |
| 271 | \subsection{Adaptive Equalizer} |
| 272 | Adaptive equalizers can be used to correct for time-varying effects, |
| 273 | an example of which is polarization dependent effects. |
| 274 | \cite{savorydigital} discusses the implementation of adaptive |
| 275 | equalization to PDM signals. |
| 276 | This has yet to be implemented in the simulation model. |
| 277 | |
| 278 | On the other hand, an implementation for a single polarization |
| 279 | state has been done. This would be useful for correcting for |
| 280 | fluctuations to the environment~\cite[\S11.6.1]{foc}, |
| 281 | not simulated in the model, but would be present in real life. |
| 282 | In addition, |
| 283 | it was observed that the CD compensating filter discussed |
| 284 | in \Cref{sec:CD} does not perform very well over short |
| 285 | distances, as can be seen in \Cref{fig:CDCompz2}, due to |
| 286 | truncation of the non-causal infinite-length impulse response. |
| 287 | Adaptive equalization was attempted to correct for this effect |
| 288 | as well. |
| 289 | |
| 290 | Two types of equalizing algorithms are typically considered, namely |
| 291 | the constant modulus algorithm (CMA) and the decision-directed |
| 292 | least mean square (DD-LMS) algorithm~\cite[\S11.6.1]{foc}. |
| 293 | CMA has been implemented due to its |
| 294 | simplicity. If time permits, DD-LMS can also be attempted. |
| 295 | |
| 296 | The CMA relies on the fact that for PSK signals, the transmitted |
| 297 | symbols all have unit amplitude. As a result, it attempts to minimize |
| 298 | the distance between the signal and the unit circle. |
| 299 | \Cref{fig:adaptBefAft} illustrates the adaptive nature of the algorithm. |
| 300 | \Cref{fig:CDCompz2} demonstrates the success of the CMA, bringing |
| 301 | the performance curve back to the theoretical values. |
| 302 | |
| 303 | \begin{figure}[htb] |
| 304 | \centering |
| 305 | \includegraphics[width=.44\textwidth]{CDCompz2.eps} |
| 306 | \caption{QPSK signal with CD, CD compensation, and CMA adaptive |
| 307 | equalizer, over an AWGN channel, with $z=\SI{2}{km}$.} |
| 308 | \label{fig:CDCompz2} |
| 309 | \end{figure} |
| 310 | |
| 311 | \begin{figure}[htb] |
| 312 | \centering |
| 313 | \begin{subfigure}[t]{.22\textwidth} |
| 314 | \centering |
| 315 | \includegraphics[width=\textwidth]{adaptBefore.eps} |
| 316 | \caption{Symbols 1 to 500.} |
| 317 | \end{subfigure}% |
| 318 | \begin{subfigure}[t]{.22\textwidth} |
| 319 | \centering |
| 320 | \includegraphics[width=\textwidth]{adaptAfter.eps} |
| 321 | \caption{Symbols 2001 to~2500.} |
| 322 | \end{subfigure} |
| 323 | \caption{Constellations showing the adaptive behaviour of |
| 324 | the CMA.} |
| 325 | \label{fig:adaptBefAft} |
| 326 | \end{figure} |
| 327 | |
| 328 | \subsection{Phase Noise Correction} |
| 329 | Lasers used in the transmitter and the receiver local oscillator |
| 330 | have a linewidth $\Delta\nu$ over which random frequency deviations |
| 331 | occur, resulting in a phase noise in the signal. When discretized, |
| 332 | the phase noise $\phi[k]$ can be modelled as a one-dimensional |
| 333 | Gaussian random walk, |
| 334 | \begin{gather*} |
| 335 | \phi[k] = \phi[k-1] + \Delta\phi_k \\ |
| 336 | \qq*{where} \Delta\phi_k |
| 337 | \mathrel{\overset{\makebox[0pt]{\mbox{\normalfont\tiny\sffamily i.i.d.}}}{\sim}} |
| 338 | \mathcal{N}(0, 2\pi \Delta\nu T_s) |
| 339 | \quad\text{for all }k, |
| 340 | \end{gather*} |
| 341 | with $T_s$ being the sampling period~\cite[\S11.3]{foc}. |
| 342 | |
| 343 | The effect of phase noise can be most easily understood from a plot |
| 344 | of the constellation, as shown in \Cref{fig:phaseNoiseCircle}. |
| 345 | Demodulation is |
| 346 | impossible without any correction. Fortunately there are various |
| 347 | techniques to mitigate this issue, and two of them are discussed |
| 348 | below. |
| 349 | |
| 350 | \subsubsection{Differential PSK} |
| 351 | In a normal PSK scheme, information is modulated as the |
| 352 | phase of each transmitted symbol. In contrast, in differential PSK (DPSK), |
| 353 | information is modulated as the \emph{difference} in phase between |
| 354 | two consecutive symbols~\cite[\S7.3.2]{ccsm}. |
| 355 | It can mitigate the effect of phase noise |
| 356 | if the linewidth is small (such that $\Delta\phi_k$ is sufficiently |
| 357 | smaller than, for example, $\pi/4$ for QPSK). Phase noise would then |
| 358 | have little influence to the phase difference between consecutive |
| 359 | symbols. |
| 360 | |
| 361 | It was however noted that in DPSK, the demodulator is affected |
| 362 | ``twice'' by phase noise. This increases the noise variance, |
| 363 | making bit errors more likely~\cite[\S7.3.2]{ccsm}. |
| 364 | This can be seen (among other results) in \Cref{fig:phasenoise_ult}. |
| 365 | This translates to |
| 366 | a SNR penalty compared to the normal PSK scheme. At a BER of |
| 367 | $10^{-3}$, the penalty is about \SI{2.5}{dB}. |
| 368 | |
| 369 | \subsubsection{Block phase noise estimation} |
| 370 | The phase noise can also be estimated assuming the total phase noise |
| 371 | over a small number of symbols is small. The Viterbi-Viterbi algorithm |
| 372 | used is best illustrated by an example. Consider a QPSK scheme. At the |
| 373 | receiver, the received signal $r[k]$ is given by |
| 374 | \[ |
| 375 | r[k] = \exp\left( \imj \phi[k] + \imj \frac{\pi}{4} + |
| 376 | \imj \frac{d[k]\pi}{2} \right) + n[k] |
| 377 | \] |
| 378 | where $\phi[k]$ is the unknown phase of the $k$th symbol, |
| 379 | $d[k] \in \{0, 1, 2, 3\}$ is the transmitted data, and |
| 380 | $n[k]$ is AWGN. Taking the signal to the 4th power eliminates |
| 381 | $d[k]$ from the expression, resulting in |
| 382 | \begin{equation} |
| 383 | r[k]^4 = \exp\left( \imj 4\phi[k] + \imj \pi\right) + n'[k] |
| 384 | \label{eq:rk4} |
| 385 | \end{equation} |
| 386 | where $n'[k]$ are the terms involving $n[k]$. It can be shown |
| 387 | that $n'[k]$ has zero mean, thus if $\phi[k]$ does not vary |
| 388 | much over a small range of $k$, then its value can be estimated |
| 389 | by averaging over that range (thus eliminating $n'[k]$)~\cite[\S11.5]{foc}. |
| 390 | \Cref{fig:viterbiphest} shows the algorithm |
| 391 | estimating the phase of a noisy signal. |
| 392 | |
| 393 | With a phase estimation method available, the effect of phase noise |
| 394 | can be undone simply by adding a reversed phase shift. |
| 395 | |
| 396 | \begin{figure}[htb] |
| 397 | \centering |
| 398 | \begin{subfigure}[t]{.22\textwidth} |
| 399 | \centering |
| 400 | \includegraphics[width=\textwidth]{phaseNoiseCircle.eps} |
| 401 | \caption{Phase noise randomly rotating the constellation.} |
| 402 | \label{fig:phaseNoiseCircle} |
| 403 | \end{subfigure}% |
| 404 | \begin{subfigure}[t]{.22\textwidth} |
| 405 | \centering |
| 406 | \includegraphics[width=\textwidth]{phaseEst.eps} |
| 407 | \caption{Example of the Viterbi-Viterbi algorithm |
| 408 | estimating phase noise.} |
| 409 | \label{fig:viterbiphest} |
| 410 | \end{subfigure} |
| 411 | \caption{Phase noise, and how it affects the received symbols.} |
| 412 | \end{figure} |
| 413 | |
| 414 | However, at larger linewidths, phase estimation may make mistakes. |
| 415 | This is due to the ambiguity in \Cref{eq:rk4}, where in QPSK an |
| 416 | additional phase increase of $\pi/2$ gives the same solution, |
| 417 | and phase noise makes unambiguous phase unwrapping impossible. |
| 418 | This is known as a \emph{cycle slip}~\cite{taylorphest}, and |
| 419 | is illustrated in \Cref{fig:cycleslip}. |
| 420 | |
| 421 | The result of a particular run of the simulation is shown in |
| 422 | \Cref{fig:phasenoise_ult}. |
| 423 | It can be seen that when cycle slips do not occur, the resulting |
| 424 | BER is much closer to the theoretical AWGN channel compared to |
| 425 | DPSK. However, if a cycle slip occurs, all the subsequent symbols |
| 426 | will be demodulated incorrectly~\cite{taylorphest}, |
| 427 | giving very poor performance. |
| 428 | |
| 429 | To eliminate the effect of cycle slips, principles from DPSK |
| 430 | can be incorporated into the phase estimation method, but instead |
| 431 | of differentially modulating the \emph{symbols}, the source |
| 432 | \emph{bit stream} is differentially \emph{encoded}. This is known |
| 433 | as \emph{differentially encoded} PSK (DEPSK). At the receiver, the |
| 434 | symbols are corrected after phase estimation (as above), and then |
| 435 | demodulated like conventional PSK, before differentially decoding |
| 436 | the bits. While this method transforms a single bit error into |
| 437 | a pair of bit errors~\cite{taylorphest}, it has a smaller SNR |
| 438 | penalty than DPSK~\cite[Ch.~13]{matlabcomm}, since the |
| 439 | noise variance |
| 440 | is not increased like it is in DPSK. \Cref{fig:phasenoise_ult} also |
| 441 | shows the result of DEPSK, which is immune to cycle slips, with |
| 442 | a smaller SNR penalty than DPSK. Many forward error correction |
| 443 | codes can effectively correct for short bursts of bit errors, |
| 444 | thus further reducing the penalty~\cite{taylorphest}, however |
| 445 | this will not be investigated in this project. |
| 446 | |
| 447 | \begin{figure}[htb] |
| 448 | \centering |
| 449 | %\begin{subfigure}[t]{.22\textwidth} |
| 450 | %\centering |
| 451 | \includegraphics[width=.3\textwidth]{cycleslip.eps} |
| 452 | \caption{A cycle slip.} |
| 453 | \label{fig:cycleslip} |
| 454 | %\end{subfigure}% |
| 455 | %\begin{subfigure}[t]{.22\textwidth} |
| 456 | % \centering |
| 457 | % \includegraphics[width=\textwidth]{adaptAfter.eps} |
| 458 | % \caption{Symbols 2001 to~2500.} |
| 459 | %\end{subfigure} |
| 460 | %\caption{Constellations showing the adaptive behaviour of |
| 461 | %the CMA.} |
| 462 | \end{figure} |
| 463 | |
| 464 | \begin{figure}[htb] |
| 465 | \centering |
| 466 | \includegraphics[width=.44\textwidth]{phasenoise_ult.eps} |
| 467 | \caption{Performance of various methods under a phase noise |
| 468 | of \SI{10}{MHz}, on a particular run of the simulation.} |
| 469 | \label{fig:phasenoise_ult} |
| 470 | \end{figure} |
| 471 | |
| 472 | |
| 473 | \subsection{Non-linearity: Kerr Effect} |
| 474 | Kerr effect is one of the non-linear effects investigated in this |
| 475 | project. Kerr effect describes the change in refractive index of |
| 476 | a material as the optical power of the incident beam changes. |
| 477 | The result is a phase shift proportional to the optical power |
| 478 | (i.e.~the square of the electric field, hence |
| 479 | non-linear)~\cite[\S10.2]{foc},~\cite[\S6.2.2]{nfo}. |
| 480 | To numerically simulate this effect together with other linear effects, |
| 481 | the \emph{split-step Fourier method} is used. In brief, the fibre |
| 482 | length is divided into many small bits. The signal is first transformed |
| 483 | to the Fourier domain, and chromatic dispersion is applied |
| 484 | (as in \Cref{sec:CD}). The signal is then transformed back to the time |
| 485 | domain and its power is calculated. From this, the corresponding |
| 486 | phase shift due to Kerr effect can be applied. This process repeats |
| 487 | until the total simulated length reaches the desired transmission |
| 488 | distance~\cite[\S2.4.1, App.~B]{nfo}. |
| 489 | |
| 490 | Currently, the general structure of the split-step Fourier method |
| 491 | has been coded, but there are small problems that require fixing, |
| 492 | and as such results are yet to be included in this report. However, |
| 493 | the general shape of the resulting curve matches existing |
| 494 | literature~\cite{savory100Gbps}, |
| 495 | so there should be little difficulty in having it completed soon. |
| 496 | |
| 497 | \section{Future Plan and Timeline} \label{sec:future} |
| 498 | After completing the simulation for Kerr effect, the most important |
| 499 | task would be to integrate all the effects into a single simulation |
| 500 | program, to prepare for the final model to evaluate different |
| 501 | transmission schemes. |
| 502 | Afterwards, it was planned to have a more realistic |
| 503 | model of the noise -- the AWGN channel would be replaced with |
| 504 | a combination of thermal noise (which can be modelled as |
| 505 | AWGN)~\cite[\S8.1.1]{aoe} |
| 506 | and shot noise. Finally, PDM and wavelength-division |
| 507 | multiplexing would |
| 508 | be implemented to have a ``complete'' model. To have sufficient |
| 509 | time for the remaining parts of the project, it was planned to have |
| 510 | this completed by week 3 of Lent term, i.e.\ about one week for |
| 511 | each of the three tasks. |
| 512 | |
| 513 | A few different designs of the network will be evaluated and compared, |
| 514 | and the suitability to use in a PON will be discussed. Running the |
| 515 | simulation a few times with different parameters should not take |
| 516 | too much time, but discussing real-life feasibility may involve |
| 517 | more review of current literature, so an estimate of 2 weeks is |
| 518 | reserved for this. |
| 519 | |
| 520 | The final three weeks of Lent will be spent obtaining experimental |
| 521 | data and verifying simulation results, to make further adjustments |
| 522 | to the model if necessary, and to prepare |
| 523 | for the final report and presentation. |
| 524 | |
| 525 | It is expected that most of the Easter vacation would be spent preparing |
| 526 | for the examinations. Work on the final report and presentation would |
| 527 | resume after that, which should be enough time to meet the deadline |
| 528 | in week 5 of Easter term. |
| 529 | |
| 530 | \begin{thebibliography}{10} |
| 531 | \bibitem{PONintro} |
| 532 | C.C.K.~Chan, |
| 533 | ``Protection architectures for passive optical networks,'' in |
| 534 | \textit{Passive Optical Networks: Principles and Practice}, |
| 535 | C.F.~Lam, Ed. |
| 536 | Burlington, MA: Academic Press, 2007, pp.~243-266. |
| 537 | \bibitem{NGPON2-1} |
| 538 | J.S.~Wey \textit{et al.}, |
| 539 | ``Physical layer aspects of NG-PON2 standards -- Part 1: |
| 540 | optical link design,'' |
| 541 | \textit{J.~Opt.\ Commun.\ Netw.}, vol.~8, no.~1, pp.~33-42, 2016. |
| 542 | doi:10.1364/JOCN.8.000033 |
| 543 | \bibitem{ponroadmap} |
| 544 | D.~Nesset, ``PON Roadmap,'' |
| 545 | \textit{J.~Opt.\ Commun.\ Netw.}, vol.~9, no.~1, pp.~A71-A76, 2017. |
| 546 | doi:10.1364/\allowbreak JOCN.9.000A71 |
| 547 | \bibitem{foc} |
| 548 | S.~Kumar and M.J.~Deen, |
| 549 | \textit{Fiber Optic Communications: Fundamentals and Applications}. |
| 550 | Chichester, UK: Wiley, 2014. |
| 551 | \bibitem{savorydigital} |
| 552 | S.J.~Savory, ``Digital filters for coherent optical receivers,'' |
| 553 | \textit{Opt.\ Express}, vol.~16, no.~2, pp.~804-817, 2008. |
| 554 | doi:10.1364/OE.16.000804 |
| 555 | \bibitem{ccsm} |
| 556 | J.G.~Proakis and M.~Salehi, |
| 557 | \textit{Contemporary Communication Systems Using \MATLAB}. |
| 558 | Pacific Grove, CA: Brooks/Cole, 2000. |
| 559 | \bibitem{taylorphest} |
| 560 | M.G.~Taylor, ``Phase estimation methods for optical coherent |
| 561 | detection using digital signal processing,'' |
| 562 | \textit{J.~Lightwave Technol.}, vol.~27, no.~7, pp.~901-914, 2009. |
| 563 | doi:10.1109/JLT.2008.927778 |
| 564 | \bibitem{matlabcomm} |
| 565 | The MathWorks, Inc., |
| 566 | \textit{Communications Toolbox\textnormal{\texttrademark{}} User's Guide} |
| 567 | (R2018b), |
| 568 | 2018. [Online]. Available: |
| 569 | \url{https://www.mathworks.com/help/pdf_doc/comm/comm.pdf}. |
| 570 | [Accessed: Jan.~9, 2019]. |
| 571 | \bibitem{nfo} |
| 572 | G.P.~Agrawal, |
| 573 | \textit{Nonlinear Fiber Optics}, 5th ed. |
| 574 | Oxford, UK: Academic Press, 2013. |
| 575 | \bibitem{savory100Gbps} |
| 576 | Md.S.~Faruk, D.J.~Ives, and S.J.~Savory, |
| 577 | ``Technology requirements for an Alamouti-coded \SI{100}{Gb/s} |
| 578 | digital coherent receiver using $3\times3$ couplers for |
| 579 | passive optical networks,'' |
| 580 | \textit{IEEE Photon.\ J.}, vol.~10, no.~1, 2018. |
| 581 | doi:10.1109/JPHOT.2017.2788191 |
| 582 | \bibitem{aoe} |
| 583 | P.~Horowitz and W.~Hill, |
| 584 | \textit{The Art of Electronics}, 3rd ed. |
| 585 | New York: Cambridge University Press, 2015. |
| 586 | \end{thebibliography} |
| 587 | |
| 588 | \end{document} |