Performance Modeling and Evaluation for Optical ...

Journal of Communication and Computer 10 (2013) 686-692

Performance Modeling and Evaluation for Optical Access Networks Tony Tsang School of Professional Education and Executive Development (SPEED), Hong Kong Polytechnic University, Hung Hom, Hong Kong, China

Received: April 12, 2013 / Accepted: April 29, 2013 / Published: May 31, 2013. Abstract: GEPON (gigabit Ethernet passive optical network) is widely deployed to realize the economical FTTH (fiber-to-the home) system for supporting high speed Internet and video distribution service. Looking ahead, next home FTTH standardization activities, 10G-EPON, were completed at the IEEE 802.3av Task Force in 2009, and the standards had specified the co-existence of 10G-EPON on the same exist optical access distribution network as a cost-effective solution. A key device in such a system is an optical receiver IC (integrated circuit) with a quick response and high sensitivity that realizes high efficiency in data transmission. This paper also reports burst-mode optical receiver ICs fabricated using 0.25 m SiGe BiCMOS (Silicon-germanium Bipolar Complementary Metal Oxide Semiconductor) technologies for a 10G-EPON system, which is a promising access network for a next generation PON system. The results show that SiGe BiCMOS can provide high performance and cost effective receiver ICs for 10G-class PON systems. The authors use PEPA (performance evaluation process algebra) to evaluate a typical PON system’s performance. The approach is more convenient, flexible and lower cost than the former simulation method which needs to develop special hardware and software tools. Moreover, they can easily analyze how changes in performance depend on changes in a particular mode by supplying ranges for parameter values. Key words: GEPON, FTTH, PEPA, WDM (Wire Digram Manual).

1. Introduction The demand for broadband services has been promoting the rapid growth of optical access systems. A PON (passive optical network) is the most suitable solution for the construction of economical FTTH (fiber-to-the home) systems. The commercial installation progress of the PON systems is from STM-PON with 10 Mb/s to future PON over 10 Gb/s. In 2004, GE-PON with 1 Gb/s entered commercial service and the number of FTTH users in the Japanese market has reached more than 17 million as of December 2009. The next generation access system standardization activities for 10G-EPON were completed at the IEEE 802.3av, and those for Corresponding author: Tony Tsang, Ph.D., lecturer, research fields: mobile computing, networking, protocol engineering and formal methods. E-mail: [email protected].

NG-PON are under way at FSAN/ITU-T. This has encouraged carriers to deploy PONs that can easily be upgraded as new technologies mature and new standards evolve [1]. Among the several NG-PON (next-generation PON) requirements are the provisioning of higher bandwidth per subscriber, an increased splitting ration, and an extended maximum reach compared to current EPON and GPON architectures. NG-PONs may offer additional functionalities such as protection, support topologies other than conventional tree structures, and enable the consolidation of access, backhaul, and metro network infrastructures. In addition, substantial research activity is currently focused on the convergence of optical and wireless access architectures into bimodal FiWi (fiber wireless) access networks, a key feature of NG-PONs. An important

Performance Modeling and Evaluation for Optical Access Networks

goal of FiWi research is to combine the most promising technologies proposed for wireless and optical access. Many FTTH operators have adopted PON access systems with a point-to-multipoint configuration where several subscribers share a single optical fiber plant as shown in Fig. 1. All the ONUs (optical network units) are connected to the OLT (optical line terminal) by a passive optical splitter, and the system is cost effectively configured. The ONUs employ a passive optical coupler to allow them to share an expensive OLT. Upstream and downstream data are multiplexed using different optical wavelengths by using WDM (wavelength division multiplexing) techniques. The downstream traffic is continuous data. On the other hand, the upstream traffic is burst-mode data with different signal powers that employ TDM (time division multiplexing). To provide a more attractive service, the line rate in PON systems has been steadily increased. The line rate has now reached 10.3 Gbit/s in GE-PON or GPON, which are two of the most widespread PON systems. To make it possible to handle more attractive content, such as high-quality moving images, an access infrastructure with a broader bandwidth is needed and 10G-EPON was standardized by the IEEE for the next generation network. A high-speed burst-mode receiver is a key component that allows to boost the transmission speed from 1G to 10G. In such a system, the receiver in the OLT requires high-speed data operation with a quick, highly sensitive response to burst data and a wide dynamic range to compensate for the shared access loss. In particular with burst mode transmission, a quick response can improve the data transmission efficiency by shortening the settling time. In this paper, all provide an overview of burst-mode receiver ICs for broadband access networks, and describe the current state of the art. All also report 10G burst-mode receiver ICs fabricated by using 0.25 m SiGe BiCMOS technologies, focusing especially on the response to burst data. Optical transceivers using these

687

ICs exhibit a fast response, high sensitivity, and a wide dynamic range. Moreover, the authors propose a framework for combined timed behaviors and stochastic process algebra. This was done by a modeling language, called PEPA (performance evaluation process algebra), for describing the timed stochastic behaviors of wireless networks. This methodology can then be analyzed using an automatic tool. With this performance analysis methodology, it is possible to obtain the design parameters of implementation using simulation with a lower computational time and cost.

2. Burst-Mode Receiver in GE-PON Systems When increasing the physical speed to 10 Gb/s, the authors must consider coexistence with already installed systems, such as GE-PON and video streaming systems. An outline of a 10GEPON system coexisting with GE-PON and video systems is shown in Fig. 2. Fig. 2a shows the physical architecture and Fig. 2b shows the wavelength allocation. The OLT must handle both 1G and 10G signals to connect both optical network units 1G-ONU and 10G-ONU. By using the WDM technique, the 10G-EPON services can coexist with the present GE-PON system. 10G downstream and upstream signals are assigned to 1,577 nm and 1,270 nm, respectively. These wavelengths are different from those already employed by the GE-PON system and video system with the exception of 1G and 10G upstream signals. The TDM technique is used for the upstream signals [2]. In such systems, a burst-mode receiver in the OLT must provide high sensitivity and a wide dynamic range with a fast response to burst data. These requirements, which result from the different distances between the OLT and the ONUs and the use of the passive coupler, lead to significantly different optical powers between each data packet and a large path loss. To compensate for this shared access loss, the receiver should have a high sensitivity and a wide dynamic

688


range with regard to the power of the input signals. In burst-mode transmission, a quick response is a key function and it can improve the efficiency of data transmission. Each packet consists of a preamble, a header, and a payload. To prevent collisions between packets, the timing of each packet is controlled by the OLT in consideration of the laser on and off time (Ton and Toff). The preamble data and Ton and Toff are preliminary data. Shortening the response time of the receiver makes it possible to reduce the amount of preliminary data, thereby improving the efficiency of data transmission. To achieve smooth migration from an installed GE-PON system to a 10G-EPON system, these systems must be able to coexist. The OLT must

handle 1G and 10G data for connection with both 1G-ONU and 10G-ONU. Thus, the receiver is in the OLT. 10G-EPON system coexisting with GE-PON includes network architecture and wavelengths allocation. Structure of data packet in PON system must handle both 1G and 10G burst data. In addition, the cost of the receiver must be sufficiently low for subscriber use.

3. Performance Modeling and Tools PEPA, developed by Hillston in 1994 [3, 4], is a timed and stochastic extension of classical process algebras such as CSP (Communication Sequential Process) [5]. It describes a system as an interaction of the components and these components engage in activities. Generally, components model the physical or logical elements of a system and activities characterize the behavior of these components. An exponentially-distributed random variable is associated with each activity specifies the duration of it, that leads to a clear relationship between the model and a CTMC (continuous time Markov chain) process.

Fig. 1 Configuration of PON access system.

Fig. 2 10G-EPON system coexisting with GE-PON. (a) network architecture; (b) wavelengths allocation.

Via this underlying Markov process performance measures can be extracted from the model. The PEPA formalism provides a small set of operators which are able to express the individual activities of components as well as the interactions between them. The authors provide a brief summary of the operators here, more details about PEPA can be found in Refs. [3, 6]. Prefix: (a, r). P. The component will subsequently behave as P after it carries out the activity (a, r), a represents the action type and r represents a duration which satisfies exponential distribution with parameter r. Choice: P + Q. The component represents a system which may behave either as P or as Q. The choice depends on which activity is completed first. Cooperation: P ▹◃L1 Q. The component represents the interaction between P and Q. The set L is called the cooperation set and denotes a set of action types that must be carried out by P and Q together.


Probabilistic

Choice:

P

r

Q

denotes

the

probabilistic choice with the conventional generative interpretation, thus with probability r the process behaves like P and with probability 1 – r it behaves like Q. Hiding: P / L. Hiding make the activities whose action types in L invisible for external observer. Constant: P = Q. The equation gives the constant P the behavior as the component Q. 3.1 PEPA Eclipse Plug-in Tools The PEPA is a language for modeling systems in which a number of interacting components run in parallel, and whose behavior is stochastic. The core semantics of PEPA is in terms of CTMCs (continuous time Markov chains), and an alternative semantics in terms of ODEs (ordinary differential equations) has also been developed. PEPA has been applied in practice to a wide variety of systems, and its success as a modeling language has been largely down to its extensive tool support. Most recently, the PEPA Plug-in Project [7, 8] has integrated a range of analysis techniques based on both numerical solution and simulation into a single tool built on top of the eclipse platform [9]. As with all compositional Markovian formalisms, however, PEPA suffers from the state space explosion problem. A model can have an underlying state space that is exponentially larger than its description, meaning that it can be infeasible to analyze. Fluid flow approximation using PEPAs ODE semantics can solve this problem if all are only interested in the average behavior of the system over time. However, if the authors want to reason over all possible behaviors of the model for example, the probability that an error occurs within some time interval then they must consider the CTMC semantics. In this paper, all present a new extension to the PEPA plugin, in which a model can be abstracted by combining, or aggregating, states. To safely over-approximate the behavior of the original model (for any aggregation of its states), all use two

689

abstraction techniques － abstract CTMCs (a type of Markov decision process with infinite branching), and stochastic bounds. The authors provide a model checker for the three-valued CSL (continuous stochastic logic), which computes from the abstraction a safe bound of the probability of a quantitative property holding in the original model X if the actual probability is p, then the model checker will return an interval I = [p1, p2] such that p2 in I. The current version of the PEPA plug-in is available from http://www.dcs.ed.ac.uk/pepa/tools/plugin, and provides several views. 3.2 State Space View The state space view is linked to the active PEPA editor and provides a tabular representation of the state space of the underlying Markov chain. The table is populated automatically when the state space exploration is invoked from the corresponding top level menu item. A row represents a state of the Markov chain, each cell in the table showing the local state of a sequential component. The order in which sequential components are displayed corresponds to the order in which they are found in the cooperation set by depth-first visit of the co-operations binary tree. A further column displays the steady state probability distribution if one is available. A toolbar menu item provides access to the user interface for managing state space filters. When a set of filter rules is activated, the excluded states are removed from the table. The probability mass of the states that match the filters is automatically computed and shown in the view. Filter rules are assigned names and made persistent across workspace sessions. From the toolbar the user can invoke a wizard dialogue box to export the transition system and one to import the steady-state probability distribution as computed by external tools. The view also has a single-step debugger, a tool for navigating the transition system of the Markov chain. The debugger can be opened from any state of the chain and its layout is as follows.

690


In an external window are displayed the state description of the current state and two tables. The tables show the set of states for which there is a transition to or from the current state. The tables are laid out similarly to the views main table. In addition, the action types that label a transition are shown in a further column. The user can navigate backwards and forwards by selecting any of the states listed. 3.3 Performance Evaluation View and Graph View Performance evaluation view and graph view a wizard dialogue box accessible from the top-level menu bar guides the user through the process of performing steady-state analysis on the Markov chain. The user can choose between an array of iterative solvers and tune their parameters as needed. Performance metrics are calculated automatically and displayed in the performance evaluation view. It has three tabs showing the results of the aforementioned reward structures (throughput, utilization and population levels). Throughput and population levels are arranged in a tabular fashion, whereas utilization is shown in a two-level tree. Each top-level node corresponds to a sequential component and its children are its local states. The performance evaluation view can feed input to the graph view, a general-purpose view available in the plug-in for visualising charts. Throughputs and population levels are shown as bar charts and a top-level node of the utilisation tree is shown as a pie chart. As with any kind of graph displayed in the view, a number of converting options are available. The graph can be exported to PDF or SVG and the underlying data can be extracted into a comma separated value text file.

4. Modeling Receiver

for

Burst-mode

Optical

The receiver in the OLT consists of a PD (photo detector), a TIA (trans-impedance amplifier) and a limiting amplifier as shown in Fig. 3. To receive data packets with different amplitudes, the receiver

amplifier has AGC (auto-gain control) and AOC (auto offset cancelling). The combination of an AGC and AOC provides appropriate amplification without waveform distortion. Moreover, it eliminates the offset voltage in each data packet and generates a clear reshaping signal that can be received in the following CDR (clock and data recovery) circuit. To achieve a quick response and constant amplitude, fast AGC and AOC circuit techniques are employed in a burst-mode receiver. The OLT requires a wide dynamic range of over 20 dB owing to the varied optical path loss and large passive coupler loss. The OLT also requires a fast response to reduce the preliminary data and improve the data transmission efficiency. With 1G-class PON systems, the response time must be less than 400 ns. The receiver for a 10G-class PON also requires a response time of less than 800 ns. The receiver ICs in PON systems should be inexpensive for subscriber use. A Si IC is advantageous in terms of cost, and is widely used for optical access systems. At low speeds, ICs with CMOS technologies are mainly used. As the data rate becomes high, SiGe BiCMOS ICs are used for high speed and sensitivity. However, the sensitivity worsens as the data rate increases. This is mainly due to the thermal noise of the feedback resistor in a TIA circuit, which limits the receiver sensitivity. To widen the frequency bandwidth and reduce the noise, all adopt high-speed device technology. In addition, to improve the sensitivity, an APD (avalanche photodiode), which has gain in O/E conversion, is used instead of a PIN-PD. In a 10G-EPON system, FEC (forward error correction) is also adopted to improve sensitivity and provide a sufficient input dynamic range. The burst-mode TIA features fast gain switching and a quick AOC that allow them to achieve high sensitivity and a wide dynamic range with a fast response to burst inputs. The TIA consists of an amplifier core with a variable-feedback resistor RF


691

Fig. 3 Operation in burst-mode optical receiver.

controlled by means of a quick level-detection circuit, a single-to-balance converter, and coarse and fine AOC circuits. The level-detection circuit is configured using comparators with hysteresis characteristics. For high-speed operation with a quick response, the authors used a fast RF switch and devised a fast AOC operation technique. To widen the input dynamic range and improve sensitivity, all used a variable feedback resistor RF controlled with a quick level-detection circuit. The TIA operates in one of two gain modes according to the amplitude of the current input into the TIA. A feedback resistor (RF) with a large resistance is used for small inputs to reduce the thermal noise and output signals with a large amplitude. For large inputs, the TIA operates with a small Rf and RL to amplify the signal with less waveform distortion. As the input signal current to the TIA is increased, the feedback resistor is changed from a large resistor (RH) to a smaller one (RL). To change the gain mode, when the amplitude of the output voltage of the TIA reaches the Vth (threshold voltage), switch SW turns “on” and initiates the first change from high to low. A hysteresis comparator is employed to detect the input signal level quickly. Once the input exceeds the threshold level, it quickly switches the gain mode. The fast AOC technique widens the input dynamic range without degrading the response speed. When the TIA receives a large signal current, the coarse AOC compensates the offset level by coarsely drawing the

current at the TIA input. This offset compensation current IAOC is switched with a hysteresis comparator. Then, the fine AOC accurately compensates the remaining offset. The AOC uses a level-hold circuit and an offset control circuit. The level-hold circuit detects the peak level of each differential input and holds it. The control circuit makes these peak levels the same, which means the offset voltage between differential signals becomes zero. The AOC operates instantaneously due to the feed-forward configuration. Thus, the offset voltage is quickly and accurately compensated by the fast two-stage AOC circuit. An external reset signal is input between the packets to initialize the level detection and the condition of the offset compensation for each packet, which makes a quick response possible. The operation in burst-mode optical receiver is specified as follows: Photodetector 0 := .Optical-Signal0 ▹◃L0 < get,  0 >.Photodetector0; TIA0 := < listen, 0 >.AGC0 ▹◃L0 .TIA0; LIM0 := .AOC0 ▹◃L0 .LIM0; CDR0 := .Data-Recovery0 ▹◃L0 .CDR0. Burst-mode optical receiver defines how the components interact with each other. According to the working cycle and the definitions of model’s components the authors give before, the optical receiver is shown below: Receiver0 := Photodetector0 < 1 > TIA0 < 2 > LIM0

692


< 3 > CDR0 GEPON is a point-to-multipoint network consisting of an OLT, a number of ONUs, the fibers and 1:N splitter between them. Multiple ONUs can share one fiber plant, one splitter, and one OLT, thus it enables to reduce the cost for installation and operation. For downstream transmission from OLT to ONUs, 1.577 m optical wavelength range is assigned and signals are broadcasted with headers in the serial data to identify the particular ONU. For upstream transmission from ONUs to OLT, 1.27 m optical wavelength range is assigned and a TDMA (time division multiple access) scheme is employed. In addition, the co-existence of 10G-EPON and GE-PON on the same existing optical distribution is required as a cost-effective solution. Since upstream wavelength band of GE-PON spreads from 1,260 nm to 1,360 nm, meaning that the 10G and 1G bands are overlapped, so OLT have to receive burst packets, which have 10G/1G different line rate from several ONUs, so-called dual rate. Therefore, the dual-rate burst-mode optical transceivers are essential as a key optical device for 10G-EPON systems. Transceiver0 := Transmitter ▹◃L0 Receiver Transmitter 0 := .Transmitter0

range of over 23 dB. Using these receiver ICs, they also developed an optical transceiver for the OLT, and it exhibits a fast response, high sensitivity and a wide dynamic range. Moreover, it can handle the dual rate of 1G and 10G burst-mode data. This reveals that SiGe BiCMOS technology can provide cost effective receiver ICs with high performance for next generation PON systems. Furthermore, they propose a framework for combined timed behaviors and stochastic process algebra, called PEPA, for describing the timed stochastic behaviors of PON systems. This methodology can then be analyzed using an automatic tool. With this performance analysis methodology, it is possible to obtain the design parameters of implementation using simulation with a lower computational time and cost.

References [1]

[2]

[3]

The overall system becomes ONU and OLT: ONU0 := Transceiver0 ▹◃L0 Splitter0 OLTn := ONU0 < 1 > ONU1 < 2 > ONU2 < 3 >…

[4]

5. Conclusions

[5]

This paper presented an overview of optical receiver ICs for broadband access networks focusing on burst-mode operation, which is a key feature in a PON system. The authors also reported burst-mode optical receiver ICs fabricated by using SiGe BiCMOS technologies for a 10G-EPON system. The performance of a burst-mode TIA and a limiting amplifier were described. The developed TIA IC with APD exhibits a fast response time of 230 ns and a high sensitivity of -28.7 dBm with a wide dynamic

[6]

[7]

[8] [9]

K. Fouli, M. Maier, M. Medard, Network coding in next-generation passive optical networks, IEEE Communications Magazine 49 (2011) 38-46. M. Fujiwara, K.I. Suzuki, T. Imai, K. Taguchi, H. Ishii, N. Yoshimoto, et al., ALC burst-mode optical fiber amplifiers for 10 Gb/s-class long-reach PONs, IEEE/OSA Journal of Optical Communications and Networking 4 (2012) 614-621. J. Hillston, A compositional approach to performance modelling, Ph.D. Thesis, The University of Edinburgh, 1994. J. Hillston, Fluid flow approximation of PEPA models, in: Proceedings of the Second International Conference on the Quatitative Evaluation of Systems, IEEE Computer Society Press, Italy, 2005. C.A.R. Hoare, Communicating Sequential Process, Prentice-Hall, US, 1985. T. Tony, Performance modeling and evaluation of recirculation optical buffer, in: Proceedings of the International Conference on Computer & Communication Engineering (ICCCE 2010), Kuala Lumpur, Malaysia, 2010. J.A.S. Micheal, Abstraction and model checking in the PEPA plug-in for eclipse, in: Seventh International Conference on the Quantitative Evaluation of Systems, Virginia, USA, 2010. M. Tribastone, A. Duguid, S. Gilmore, The PEPA eclipse plug-in, Performance Evaluation Review 36 (2009) 28-33. The Eclipse Platform, http://www.eclipse.org.