The bit error ratio (BER) of a serial data stream is impacted by both jitter and noise. There are several tools that allow the visualization and measurement of jitter, such as BER contours and BER vs. decision timing, also known as the bathtub curve.
But while these measurements do a good job of validating overall channel performance, they do not necessarily provide much help when looking for the root cause of noise-related channel problems. As the data rates climb above 6 Gb/s and toward 10 Gb/s, it becomes apparent that the effects of noise must be included when predicting BER.
Jitter analysis methods are well established, but methods of handling noise analysis are not broadly agreed upon. Now there is a new method of handling noise analysis in a manner similar to jitter. This approach supports tracking BER problems back to their various jitter and noise components, giving clear insight into problems and facilitating design trade-offs. It allows for the identification of the true source of channel impairments, enhancing the reliability and accuracy of BER estimations.
Jitter and Noise Measurements
Jitter has long been broken down into a number of distinct components under total jitter (TJ). This model has proved helpful for modeling and analyzing behaviors of communications circuits. However, as data rates increase, the effects of noise need to be taken into account.
By sampling the signal at two locations, a high slew-rate edge and a zero slew location, it is possible to make similar analyses for both the jitter and noise components of a signal. This allows separation of the effects of noise from those of jitter, which is a great help when looking for root causes.
Several new terms and a few new measurements have come out of this noise analysis technique:
• Random noise (RN), like its jitter twin, is considered to be Gaussian based.
• Deterministic noise is bounded, it has limits, and it can be further broken down into two categories: data dependent noise, which is correlated with a bit sequence in the bit stream; and periodic noise (PN), which repeats in a cyclic fashion. By convention, PN is uncorrelated with any periodically repeating patterns in the data stream.
By analyzing noise together with jitter, it becomes possible to create accurate measurements of BER, the BER eye, and the BER contour. Since these BER measurements now are based on a detailed analysis rather than extrapolation, the results are much more accurate than could be obtained in the past.
Crosstalk Measurements Using Noise Analysis
Crosstalk issues have been difficult to trace to root causes in the past. In some cases, it is possible to analyze crosstalk by tracking its effect on jitter, but that is an indirect method at best.
Fundamentally, crosstalk is an amplitude effect and best analyzed in the amplitude, or noise, domain. As data rates increase, crosstalk becomes more of an issue. Accurate crosstalk measurements can become essential to the success of a project.
Crosstalk normally is bounded, or deterministic, noise. Since crosstalk is from a source other than the data stream under test, it normally is noncoherent and not data dependent.
The way in which crosstalk will be interpreted by analysis tools depends on the spectral signature of the aggressor source. If the data-dependent aggressor signal is static and repetitive, such as a clock or a repetitive serial test pattern, then crosstalk can be isolated and categorized as PN. When the data-dependent crosstalk is nonrepetitive, it becomes broadband, and current analysis tools have difficulty dealing with it.
In real-life testing situations, long test patterns or nonrepeating serial data can create broadband crosstalk. The result is a slight lifting of the noise floor from the spread-spectrum effect, which is hard to distinguish from the normal noise floor. In an equivalent time-sampling oscilloscope, this effect is aggravated due to the aliased acquisition of the jitter and noise spectra. This aliasing folds crosstalk content spanning a frequency range of hundreds of megahertz back into a Nyquist band of typically less than 100 kHz.
An experimental study of crosstalk from a noncoherent, repetitive aggressor line was done using the setup shown in Figure 1 to examine the robustness and accuracy of a jitter and noise analysis method as a function of the spectral signature of the aggressor. The aggressor signal was composed of a noncoherent, repetitive series of pseudo random binary sequence (PRBS) serial data patterns with increasing spectral complexity.
The aggressor-victim configuration on a simulated backplane is powered by two signal sources. The far-end victim signal is acquired on a sampling oscilloscope and processed by jitter, noise, and BER analysis software. The measurement setup emulates the crosstalk behavior at the receiver side so the evaluation targets far-end crosstalk (FEXT). The measured data-dependent jitter is about 30 ps.
Figure 2 shows a set of graphs for crosstalk from a PRBS 7 pattern. The victim data rate is 6.25 Gb/s, and the aggressor data rate is 6 Gb/s. The selected plots serve to visualize the effects of crosstalk due to growing spectral complexity on the victim’s BER: the (RN)(PN) spectrum; the BER eye; the horizontal bathtub curve; and a set of BER contours taken at 1e-3, 1e-6, 1e-9, 1e-15, and 1e-18 bit error ratio probabilities.
The most obvious differences between no crosstalk and crosstalk are in the bathtub graph, which shows a higher error rate in Figure 2, and the BER eye diagrams, which also are partially closed by the crosstalk in the same figure. On the other hand, horizontal jitter [PJ (h)] remains practically constant between the two conditions, showing that it is not a good indicator of a vertical axis problem, like crosstalk, as you might expect.
Some of the measurements are affected by crosstalk:
• PN grows from 1.56 mV to 25.52 mV.
• The vertical component of PN [PN (v)] has a similar increase.
• Vertical periodic jitter [PJ (v)] increases from 1.58 ps to 30.56 ps.
• TJ and total noise also grow greatly, affected mostly by their vertical components.
The BER plots and numeric results show how crosstalk causes eye closure in both the horizontal and vertical dimensions.
The set of results for jitter, noise, and BER analysis reflects the changes in the random and bounded jitter as well as the noise components as the aggressor signal changed. Figure 3 and Figure 4 plot the results to help identify the trends. Result values have been normalized to the eye height and unit interval.
PN increases greatly between no crosstalk and PRBS 7. PJ (v) also becomes strong in this region. The other graphed values either do not change dramatically or are totals and influenced by PN and PJ (v) in this area of the graph. This correctly identifies the source of the crosstalk as bounded PN. This correct identification continues from PRBS 7, the first pattern used, through PRBS 11.
Limits of Separate Noise and Jitter Measurements
As the pattern length increases to PRBS 11 and PRBS 15, a limit to the spectral separation method appears. At PRBS 11, the percent of the total power of PJ (v) and PN becomes lower than before. Between PRBS 11 and PRBS 15, PJ (v) and PN, which should be growing reflecting increased crosstalk, fall to nothing. Crosstalk is leaking into the random noise and jitter components of the signal in this region. This leads to over-pessimistic BER rate assessments for longer PRBS patterns.
Separating noise and jitter for analysis is a useful tool for crosstalk analysis and can yield very accurate BER eye results for shorter PRBS patterns. However, the BER analysis becomes overly pessimistic for longer crosstalk data patterns. An additional analysis tool is needed: software-based receiver equalization.
Compensating for Channel Loss and Crosstalk
The combination of inter-symbol interference due to a lossy backplane and the significant crosstalk from an asynchronous data source causes the BER of the data link to get worse quickly as the data pattern length gets longer. As effects of crosstalk increase with the spectral complexity of a PRBS 15 aggressor, the eye closure is almost complete. At BER 1e-12, the horizontal opening is reduced to 44 ps, and the vertical eye opening is only 10 mV.
Software provides a means to explore the use of an equalizer in compensating for the two sources of impairments. As shown in Figure 5, a 10-tap feed forward linear equalizer (FFE) coupled with 10-tap decision feedback equalizer (DFE) was used to optimize the eye opening, resulting in a 98-ps horizontal eye opening and a 91-mV vertical opening at 1e-12 BER. The horizontal eye opening more than doubled, and the vertical opening became nearly 10 times wider.
Conclusion
Analyzing jitter and noise together increases the accuracy of BER prediction. The orthogonal approach offered by the jitter, noise, and BER package evaluates and quantifies jitter and noise, such as crosstalk, and can create an accurate BER prediction. It also lends itself to root-cause problem identification of circuit faults.
A BER limited severely by the combination of channel loss, dispersion, and crosstalk can be improved by the application of FFE/DFE equalization. The software equalizer allows you to explore the effects of equalization due to lossy lines and crosstalk from adjacent aggressor sources.
About the Author
Pavel Zivny is a senior product engineer with the high-performance oscilloscopes group of Tektronix. He has been with Tektronix for more than 20 years, working in test engineering, design, and marketing, and has several patents awarded and pending. Mr. Zivny represents the company in the IEEE 802.3 (Ethernet) standards body, contributes to several other standards, sits on a DesignCon panel, and has authored papers presented at industry conferences.
Tektronix, P.O. Box 500, Beaverton, OR 97077, 800-835-9433, e-mail: [email protected]
FOR MORE INFORMATION
on the analysis of effects of crosstalk
in high-speed serial data links
www.rsleads.com/908ee-203
on understanding and
characterizing timing jitter
www.rsleads.com/908ee-204
August 2009