Data-Dependent Jitter (DDJ) In Serial Links

WHAT IS DATA-DEPENDENT JITTER (DDJ)

Jitter, defined as variation of a signal edge from its ideal position in time, is an important performance measure of a serial link, or clock signal. Jitter is generally divided into two types, deterministic and random. The deterministic jitter (DJ) is bounded and may be correlated to known sources. DJ has three main parts: periodic jitter (PJ), data-dependent jitter (DDJ), duty-cycle distortion jitter (DCD). Random jitter, on the other hand, is unbounded and is due to sources that can only be characterized statistically. DDJ behaves as high-frequency jitter that is strongly correlated to bit pattern within the data stream. The main sources of DDJ are: · Inter-symbol interference (ISI) ISI is basically due to bandwidth limitations of the transmission channel, which causes single bit information to spread into adjacent data bits. Two major mechanisms affect the impact of ISI on DDJ: o Slew rate.

Due to bandwidth limitation of the transmission channel, the transition rate from a 0 bit to 1 or vice versa is finite. This results in leaking of bit information into adjacent data bits. o Phase distortions. Some channels may have very fast changing phase characteristics within specific frequency ranges (often close to the pass-band to stop-band transition range). In such cases, slight variations of data bit rate, or channels parameters, can result in significant variations in bit transition edge delay. · Reflections. Reflection occurs in a channel, which is compromised of transmission lines with mismatched termination impedances. If mismatch exists in both ends of a transmission line, a delayed and attenuated version of the transmit signal will be received at receiver in addition to the main signal. The amount of delay and attenuation depends on transmission lines characteristics and the amount of termination mismatches. In practical transmission channels, often the primary source of DDJ is slew-rate related ISI. However, in situations where the channel consists of multiple transmission lines (e.g., including multiple PCB traces, relays, connectors, intermediate terminations…), reflections and phase-distortion ISI may also become significant. ATE test fixtures may fall within this category if the channel bandwidth is close to the signal bit rate.

WHY IS DDJ IMPORTANT TO KNOW?

DDJ manifests itself as data-dependent shifts of the data transition edges relative to the data sampling point in the receiver. DDJ includes very high-frequency jitter components, which clock recovery circuits cannot track because most of its frequency components fall outside the receiver's clock recovery bandwidth. These variations result in shifting of bit-error-rate (BER) bathtub curve toward the sampling edge, which deteriorates the link BER performance. Although deterministic, DDJ is fully characterized by forming the histogram of DDJ-related shifts for all the edges in the data stream. Because of limited ISI depth (i.e., the number of adjacent bits affecting a specific data bit), DDJ is bounded within a range. ISI depth is a function of the transmission channel characteristics. In many Serial link test characterization experiments, data stream is composed of repetitions of a finite length bit pattern. In such cases, there are a finite number of DDJ-related edge shifts that can occur within the data stream signal. Therefore, the DDJ histogram, which is an estimate of DDJ probability distribution function (pdf), will consist of separate distinct lines, called DDJ delta lines. Because ISI depth is limited, such repetitive patterns can produce a good estimate of complete DDJ pdf as long as the pattern repetition includes all the bit combination within the ISI depth (e.g., CJTPAT in FiberChannel [2]).

Each DDJ delta line results in a PDF which is a summation of scaled and shifted versions of the rest of jitter component PDFs. Assuming the other jitter components mostly consist of normally-distributed random jitter, it can be shown that only DDJ delta lines located at the maximum and minimum of DDJ range significantly affect the BER performance. Therefore, peak-to-peak DDJ ( ) is used in many serial link standards to quantify DDJ. Although sufficient in some case, peak-to-peak DDJ does not completely describe DDJ impact on BER in all cases; it is often important to consider DDJ delta lines that lie close to the maximum and minimum lines, and also take into account their frequency of occurrence relative to the rest of delta-lines. Nevertheless, is an important parameter for comparing the performance of different links. 3 DDJ ESTIMATION METHODS In many test experiments, a bit-pattern is repeated in the test data stream to qualify the serial link. For example, K28.5 pattern, which includes 20 bits and 10 transition edges, may be repeated continuously to generate a test data stream. For such cases, two main methods may be used to measure DDJ: "frequency domain", and edge lock". The following section describes these methods. 3.1 FREQUENCY DOMAIN DDJ MEASUREMENT Real-time sampling oscilloscopes can be used to digitize a data stream signal. The collected samples provide fairly accurate estimate of each edge location relative to a trigger time. Such information forms time interval error (TIE) sequence, which is the difference between measured and ideal transition times. Passing the TIE sequence passes through FFT operation produces frequency domain representation of jitter signal. In frequency domain, components that are harmonics of pattern repetition rate represent DDJ-related jitter. Isolating these components and using inverse FFT operation reproduces the DDJ signal in time domain, which may be used to estimate . This method is fairly accurate when the pattern length is short (e.g., K28.5 pattern).

For long patterns, the energy of DDJ components spread over many frequency bins, causing some to be hidden by noise floor. In such cases, the inverse FFT will reconstruct only portions of DDJ, which may render an inaccurate estimate of . 3.2 "EDGE LOCK" DDJ MEASUREMENT Another DDJ measurement method is to use the TIE data directly in time domain. The TIE sequence may be generated by a real time oscilloscope or Time interval Analyzer (TIA). The DDJ component of TIE for a specific pattern edge is the same in different pattern repetitions because the data bit history before that pattern edge is similar for each repetition. Therefore, to estimate DDJ for a specific pattern edge, it is sufficient to collect a number of TIE samples for that edge from different pattern repetitions (in other words lock to that pattern edge) and compute the sample average. The averaging reduces the contributions of random and periodic jitter on the TIE data and provides an estimate of DDJ component. Repeating this procedure for the rest of the pattern edges provides DDJ for all those edges. This data can be used to identify DDJ histogram delta lines, or compute . GuideTech's Femto3200 TIA uses this method for DDJ estimation For long patterns (typically longer than 10000 bits for TIAs), it might be time consuming to collect many samples of TIE for all pattern edges. To reduce the test time, DDJ may be measured only for a subset of pattern edges that are more likely to cause maximum or minimum DDJ. Analysis of the pattern transition density may be used to identify such pattern edges. 4 DDJ MEASUREMENT ACCURACY DDJ estimation accuracy depends on two major factors: test fixture and measurement instrument. The test fixture in combination with the instrument input termination form a transmission channel from the source to the instrument internal circuitry. Due to the frequency characteristics of the transmission path and the frequency content of the source signal (rise/fall time, pre-distorted or pre-emphasized signal), the DDJ received by the instrument is different from the DDJ at the source. Typically, the path results in increased relative to the source .

This, however, should not be considered a general rule, because for some transmission channels, the pattern edges experience shifts relative their ideal positions that is in opposite direction of source DDJ, causing a lower reading of . This can specially happen if transmission path-related DDJ is dominated by phase distortions and/or reflections. For accurate DDJ measurement in the presence of test fixture effects, two methods are possible: · Ensure that the transmission path has bandwidth that is larger than bit rate, and produces negligible phase distortion in the pass band. · Calibrate the impact of transmission path on DDJ, either by obtaining the frequency response of the path and simulating the source with the path model, or by using a controlled calibration signal source. The instrument can also introduce inaccuracies. Such sources for TIAs include: · The number of TIE sample for each pattern edge ( ) · The RJ power ( ) and PJ power ( ) · Frequency characteristics of the instrument front end · TIA non-linearity and noise floor Typically, the error due to RJ and PJ are reduced by and , respectively. The PJ-induced error decreases if it is not synchronous to pattern repetition rate, otherwise, it will be difficult to separate it from DDJ. The only solution in such cases is to alter the pattern repetition rate to break the synchronization. The instrument front-end frequency characteristics can also contribute to DDJ measurement. This is, however, usually predictable because of known internal channels inside the instrument, and therefore, is calibrated out during instrument calibration. TIA noise floor is often not an accuracy limiting factor because it behaves similar to the signal random noise, which is attenuated by averaging operation. Non-linearity issues are predictable and hence calibrated out within the instrument algorithms.

It is evident that measurement error reduces as increases in the presence of RJ and PJ. It also shows that accurate DDJ measurements are possible even in the presence of significant RJ and PJ. Figure 2: RMS error for edge lock DDJ measurement method, 1Gbps, , , 5 CONCLUSIONS DDJ measurement methodology described in Section 3.2 provides a powerful, accurate, and flexible technique. These measurements also includes DCD. Since the effects of DCD and DDJ on critical BER performance are similar, they are often not separated. They, however, can be separated by comparing DDJ for rising edges and falling edges; this may be needed for characterization or diagnosis reasons. When measuring DDJ, particular attention has to be paid to the test fixture. At high data rates, test fixture is often the major source of inaccuracy. For production testing application, one approach is to estimate the test fixture impact and adjust the fail test threshold accordingly.