Note that jitter amplitude is expressed in units of time, typically in the picosecond range. Jitter is also represented using the dimensionless unit interval (UI), which is normalized to one bit period.

Jitter Causes
Jitter arises in telecom networks and devices for three main reasons:

1. Noise mechanisms in devices and elements.
2. The natural function of network elements and the Sonet/SDH signal structure, that occurs when two signals with slightly different timing are combined and the signal structure operates to accommodate the difference. Networks use very accurate clocks, and attempt to pass timing to all points from master sites but some variation in timing (described as "wander", a whole topic in its own right) always occur and will result in jitter produced in network elements.
3. Pattern dependency, which is the precise edge position of a pulse depending on the previous 1s and 0s history in devices such as amplifiers, modulators and detectors.

Jitter is especially important in transport systems because the long physical distances involved mean that serial data has to be sampled at the receiving end using a clock recovered from the incoming data stream itself. The clock-recovery mechanism attempts to follow timing variations and maintain sampling in the center of the eye, but it can only deal with so much timing variation, and the amount it can deal with falls off rapidly as the variation (jitter) frequency increases. If clock recovery fails, then that causes bit errors with obvious consequences for network performance. Network specifications, such as the ITU G-series specs, give limits on the amount of jitter devices should generate, tolerate and transfer.

Of these three types of measurement, jitter tolerance and jitter transfer measurements are relatively straightforward. This is mainly because they are made at relatively high levels of jitter, far away from the noise floor of the test equipment. Jitter tolerance is a particularly straightforward type of measurement since the only jitter function required in the test set is to generate an increasing amount of jitter using simple sine-wave modulation over a range of frequencies. The bit-error measuring function in the test set is used to determine the level of jitter at each frequency that can be tolerated.

Jitter transfer measurement is more exacting in that the test equipment needs to measure the amount of jitter out of a device as a function of the amount of jitter presented to it. Despite the fact that input jitter levels are high, and more importantly the measured jitter waveform is driven by the sinusoidal stimulus jitter waveform from the test set and is usually close to sinusoidal itself, the levels after attenuation by the DUT can be very low.

A tracking filter is usually used in the jitter receiver to control any out-of-band jitter transfer. Most jitter test sets perform these functions automatically and produce graphical and tabular results referenced to the limits in the various network standards

A Difficult Task
In networking designs, the really difficult measurement is jitter generation, which is also known as output jitter or intrinsic jitter. Jitter generation is the measurement of very small amounts of jitter inherent in a device or element. Standards-imposed limits on jitter generation are generally low, with 100 mUI peak-to-peak (pk-pk) being a very common limit.

Currently, there is intense interest in jitter generation measurement of optical components. This is driven particularly by the potential lower costs associated with using the latest types of XFP optical transmitters as opposed to current 300-pin MSA devices, also known as transponders.

XFP devices simplify the integrated multiplexing and retiming functions of transponders in order to save cost, complexity and power consumption, but therefore are likely to produce higher levels of jitter if not carefully matched to the devices driving them. In particular, pattern-dependent jitter in the devices driving the XFP module will be passed on to the optical output to a much higher extent than in a transponder. In general, users of jitter test equipment have become acquainted with large variations between different measurements on the same instrument or different instruments. Part these variations are caused by the logical issue that especially pk-pk measurements will respond to time-varying characteristics in the jitter being measured. In particular a single peak can occur during one measurement run, but not during another. The longer one measures, the more likely a peak will be captured. But, there is room for reducing overall measurement uncertainty by improvement in the measurement instrument itself.

Jitter Generation Measurement Problems
There are three main problems in measuring jitter generation:

1. Designing a jitter meter to be inherently "quiet" is difficult.
2. Though root mean square [rms] (average) jitter is also specified, the principal measurement is pk-pk and even narrow single peaks should be caught and measured accurately. It is peak, not average, jitter that determines the disruptive effect in terms of error performance.
3. Until now, there has not been a real way to calibrate and hence specify a jitter meter at low values with any real accuracy.

Even though Sonet/SDH test sets are often considered digital test instruments, the jitter measuring function requires precision RF design as used in instruments like network analyzers and spectrum analyzers. One of the most important parts of a jitter meter is the photo detector at the optical input, in particular its linearity. It is also very important to filter the post-detector signal prior to clock recovery to remove laser on/off effects such as overshoot and relaxation oscillation. Attention to these aspects of design minimises pattern-dependency in the jitter receiver.

The calibration of a jitter measuring receiver has to be valid over all practical jitter distributions from steady-state sine wave to isolated pulses. The ITU has decided (O.172 Appendix 7) on a set of jitter waveshapes (within ranges) to represent all types of real-world jitter. These wave shapes are based on bursts of sinusoids, with the two variable parameters of burst width and burst repetition rate (Figure 2).

In effect, O.172 Appendix 7 has defined a 100 mUI pk-pk jitter calibration reference system for which a jitter meter should read 100 mUI pk-pk for all of the defined jitter wave shapes. Such pk-pk signal measurements independent of a wave shape are a bit unusual compared to more common, power-based (rms) level measurements that depend acutely on wave shape.

It's important to note that variations of other parameters of the standard signal such as digital test pattern (PRBS, Sonet/SDH overhead) must not affect the level of this 100-mUI pk-pk jitter standard. Even though Sonet/SDH signals are scrambled to prevent excessively long runs of consecutive 1s or 0s from defeating clock recovery mechanisms, not every part of the frame overhead is scrambled. O.172 Appendix 7 endorses a method called optical retiming (described below) that achieves the required pattern independence in the calibration reference signal so that the only jitter present on the calibration signal is that which is intentionally added.

The calibration system, as shown in Figure 3, consists of a clock with modulation source, a jitter calibration section (using well-established oscilloscope and spectrum analyzer techniques for measuring the amount of jitter on clock signals), and a precision optical data generator (the shaded area in Figure 3). By making several cross-checks with the oscilloscope and spectrum analyzer, it is possible to verify that the final output optical signal is completely flat — that is, it has a constant 100 mUI pk-pk of jitter for all burst widths and repetition rates.

Understanding Optical Retiming
The optical retiming performed in the precision optical data generator requires further explanation. The laser, data modulator, and electrical pattern generator are the conventional elements used in any Sonet/SDH test set. However, some pattern-dependent and noise-based jitter may be present in the optical signal out of the data modulator. In order to remove this jitter, the output from the data modulator is itself re-modulated by the transmitter clock. This technique is referred to as pulse carving.

Because the re-modulation is performed by a clock signal, rather than a data signal, there is inherently no pattern-dependent jitter present. The re-modulating pulses are arranged to be narrow and phase-aligned to sample the data modulator output in the center of the optical eye in order to avoid the jitter present at the edges of the eye.

However, the resulting optical output is essentially now a return-to-zero (RZ) signal (because the re-modulating clock pulses used are narrow). Thus, the ITU O.172 method employs an all-optical pulse stretcher to make the final optical signal a non return-to-zero (NRZ) signal.

It is important to note that the pulse stretcher does not itself add jitter back into the system. A conventional technique would be to convert the light pulses to electrical signals, then use well-known split-and-delay techniques to achieve stretching. But that would introduce uncontrolled amounts of jitter. Instead, the all-optical approach uses a birefingent optical fiber (a fiber where propagation speed depends on polarization) to allow an optical signal to be split and delayed without conversion to an electrical signal and without pulse distortion. As described in O.172 Appendix 7, it is also possible to use over-specified optical transmitters (for example, a 10-Gbit/s modulator operating at 2.5 Gbit/s) to reduce pattern-dependent jitter in some situations.

Calibrated Source In Hand
The overall calibration system assembly discussed above provides a low-noise, calibrated jitter source free of pattern-dependent jitter. This source can then be used in the manufacture and periodic re-calibration of the receivers in Sonet/SDH jitters test sets.

Once the jitter source is implemented in a test set, a standard signal is exercised over all prescribed jitter wave shapes and the response of the jitter meter recorded for each wave shape. Responses are recorded in an accuracy map (Figure 4).

The ideal accuracy map would be flat and level at 100 mUI for all wave shapes. The actual degree of flatness is a measure of the overall quality of the jitter receiver in the test set.

Once the basic accuracy map is established then calibration becomes a matter of centering the entire map on the nominal 100-mUI point. This is where the issue of careful analog design comes in. It is legitimate to calibrate in this manner, but after such calibration the overall accuracy depends on the flatness of the accuracy map.

It is also useful to produce an overall envelope accuracy map that takes into account varying environmental conditions, jitter levels other than 100 mUI pk-pk and instrument-to-instrument variation. Any measurement with any instrument of the same type at any time would then fit within this envelope map.

References

1. ITU-T O.172, http://www.itu.int/publications/index.html.
2. Understanding Jitter and Wander Measurements and Standards, http://cp.literature.agilent.com/litweb/pdf/5988-6254EN.pdf.