4.4.6 Timing

In communications systems, timing implies the following functions:

• clock extraction from the incoming signal
• jitter or wander removal by passing the serial signal through an elastic store
• retiming the signal at the output with a clock accuracy according to standards, typically ±20 ppm (parts per million)

Opaque systems extract the clock at and after the photodetector, and based on this and an elastic store buffer they time the (bits of a) signal after converting the optical signal to electrical. At the output, they retime the outgoing signal by virtue of modulation (modulating the laser directly or the modulator after the laser). The clock for retiming can be either a standard reference clock that is provided externally (one active and one standby or alternate), known as Building Information Timing Supply (BITS). In the absence of BITS, clock can be derived (extracted) from an incoming signal (such as an OC-N in SONET or a DS1 in STM or in SONET) (Fig. 4.26).

Timing accuracy is not arbitrary but is defined by standards. In the United States, the primary timing reference source (PRS) is an atomic clock of the highest accuracy (it can miss one tick in 10¹¹ or it can slip 2.523 ticks in a year). This clock is referred to as stratum 1, and it is distributed to many geographic areas from which networks and their nodes are synchronized. Subsequent to stratum 1, there are clocks derived from it and thus of lesser accuracy. Depending on the network layer, networks and their nodes must comply with the accuracy of one of the strata (Table 4.3). Typically, the external timing reference to a node is from a BITS clock of stratum 3 or better.

Figure 4.26 Node receiving BITS timing, and node extracting timing from incoming signal.

Timing extraction from the all-optical signal is not currently simple, unless a small part of the optical power of the incoming signal is split off from which clock is extracted as in opaque systems. Similarly, retiming of an all-optical signal is not trivial. Thus, current all-optical commercial systems (nodes, regenerators) do not provide the retiming function but they merely perform amplification and reshaping (dispersion compression), or 2R. However, there are coherent methods under development that promise to soon add the third R to the 2R functionality. These methods are based on fiber loops of fixed delay and circumference, as well as on SOAs with nonlinearity, a pulsed pump and interference methods (see Chapter 2).

4.4.6.1    Optical Phase-Locked Loops

An optical phase-locked loop (OPLL) is a method that extracts timing of the optical signal in the optical regime; it is based on a tunable laser source, a filter, and a photodiode bridge and its principle of operation is similar to that of electronic PLLs. For example, consider two optical signals of the same frequency and rms amplitude, one is the frequency of data signal and the other is generated by a local tunable laser source. Each optical frequency impinges on a photodiode of a balanced bridge. If the two optical frequencies are in perfect quadrature (i.e., one is described by cos ωt and the other by sin ωt) and the two frequencies are the same, then the bridge is balanced at a quiescent state. If the two frequencies are not the same, then an unbalanced current is created, amplified, and fed back through a low-pass filter to adjust the frequency of the local laser. Clearly, this arrangement assumes that both light sources (incoming and locally generated) have the same frequency and optical power level. These are two good assumptions as the incoming signal may have drifted in frequency, been dispersed (chromatic and PMD), and been attenuated. These require compensation that complicates the OPLL. However, integration soon will resolve these issues and will make the OPLL a commercially available component.

4.4.6.2   Ultrafast Optical Pattern Recognition

One of the key functions in ultrafast ultrahigh bandwidth systems is real-time pattern recognition. By pattern recognition it is meant recognize and locate the start of frame pattern (SONET, ATM, IP) which is in the overhead or header field. To achieve this, pattern synchronization and knowledge of the location of the sought pattern in the header is required. At bit rates of 10 or 40 Gb/s, ultrafast digital electronic circuitry with picosecond switching capability is challenging, and pattern recognition in the optical regime has not been cost-effectively implemented yet. Here, we describe two methods, an optoelectronic able to locate and transform byte patterns (AR-CAM), and an all-optical pattern detector (OPD) based on weighted optical splitters able to locate the "start of frame" pattern.

Figure 4.27 Architecture of an associative RAM-based CAM (AR-CAM).

AR-CAM. This is an ultrafast electronic "recognizer" circuitry limited to a clock circuitry and a simple shift register to capture a byte or a word in real time and convert it from serial to parallel. Doing so, it reduces the recognition speed by 8 or 16 and thus the "recognizer" can operate a little slower (in the nsec than the psec regime).

This device, known as associative RAM-based content addressable memory (AR-CAM), utilizes fast random access memory (RAM) devices as content addressable taking advantage of the fast access cycle of RAMs (less than 4 nsec) and their low cost. This simple approach has demonstrated pattern recognition and translation at Giga-pattern/sec cost-effectively in SONET, ATM, IP, and other applications (Fig. 4.27).

OPD. This method requires several splitters positioned in series and equidistant on a fiber that carries a single optical channel (Fig. 4.28). The separation of each splitter is such that the travel time of the optical signal from one to the next matches the period of the bit rate (at 10 Gb/s the separation is less than 2 cm). Each splitter removes a small fraction of the optical signal in such a way that all fractions are concurrently multiplexed and detected by a PIN photosensor. The output of the photosensor will provide the highest current level when all bits in the pattern are "one," thus establishing the start of frame (SoF) in the time domain; the assumption is that the SoF is always all "ones." Since the frame is synchronous by nature, other patterns that emulate the SoF will be detected, but they also will be rejected because they are by nature nonperiodic (this is the same argument in all pattern recognizers in communications systems, optical or electronic).