Understanding Eye-Pattern Analysis

With the growing use of digital data transmissions such as SONET, SDH and fiber channel, test professionals are faced with new test requirements. One deceptively simple-looking technique widely used is eye-pattern analysis. Understanding its advantages and limitations reduces erroneous conclusions and leads to reliable test methods and improved designs.

Eye-pattern analysis is used to evaluate the subset of device performance relative to the physical layer of a transmission standard. The physical layer encompasses all of the essential electrical, mechanical and optical characteristics of the transmission medium. Among these are several time-domain characteristics that require oscilloscope-like acquisition and measurements.

Today, several specialized oscilloscopes, such as communications signal analyzers, perform the needed measurements and analysis. They are extremely well suited to quickly identify faults, gather qualitative information to determine the health of the DUT and conduct compliance testing to industry standards.

Figure 1 shows a typical eye pattern acquired for use in a SONET compliance test. The eye pattern is compared to a predefined set of boundaries in which the waveform must be confined. These boundaries are collectively called a mask.

What Is an Eye Pattern?

 

As with any tool, you need to know how it works before you use it. In this case, it’s vital to understand how the eye pattern is created.


Typically, it is created using a data generator to both stimulate the DUT and supply a trigger to the oscilloscope. The clock that generates the data is the trigger source. The data pattern should be either pseudorandom or an assortment of symbols that represent typical and stressed conditions. As each data transition is synchronous with the clock, the oscilloscope can accumulate multiple waveforms or samples of waveforms and maintain their relationship to the clock.

Figure 2 depicts the process of constructing a simple eye pattern. Note that data transitions are aligned with the falling edge of the clock. One clock period is the fastest interval in which the data can change state (high or low). This is called a unit interval or bit interval. Figure 2 shows that not all of the transitions are captured. In fact, the time between waveforms or samples of waveforms is on the order of 10 µs or more.

If the data rate is 1 Gb/s, 10,000 or more bits were missed between acquisitions. Even with so many missed events, enough waveforms can be acquired to create a composite image of the major trends found in repeating bit combinations. All that will be missed are the rare and single events. The composite image is accumulated using an infinite-persistence display found on most contemporary digital storage oscilloscopes.

The composite image reveals variations in the timing of the data transition relative to the clock edges. Ideally, the crossing of high-to-low and low-to-high transitions should occur midway between the high and low levels and form a perfect looking X.

In reality, some transitions are delayed as a result of following a long sequence of bit intervals with no change. In other cases, bandwidth limitations of the DUT may attenuate the height of pulses that are only 1 bit wide.

With the eye pattern, all of these data-dependent characteristics are recorded in a single image. At a glance, the overall health of the DUT’s ability to carry a signal can be inferred.

Figure 1 shows that the eye pattern passes the test because the signal is confined within the mask boundaries. Typically, when a mask is specified for compliance testing, there is no requirement to measure rise time, fall time, pulse width, overshoot and settling time. These are assumed to be within acceptable limits if the eye pattern passes the mask test.

Additional Considerations

The mask is not very useful unless all of the conditions are met to go along with the mask shape. Figure 3 shows two masks. Each claims to be for SONET STS-3. The shapes are quite different because each was designed for an eye pattern acquired on the same signal but under two different conditions.

Similarly, masks for SONET, SDH and fiber-channel optical signals require the optical-to-electrical converter and subsequent amplifier to have a precise frequency response. Without this, the mask test is invalid and will not constitute compliance testing.

There are still more hurdles. The waveform must still be aligned to the mask. Manually adjusting time/div, volts/div, position and offset is tedious work and very difficult for most people. That is why some signal analyzers have a built-in autoscaling algorithm unique for each mask. This ensures that with a single command or button press, the signal will be placed with the best chance of fitting within the legal boundaries of the mask.

What if there is no clock signal available to trigger the oscilloscope?

If this is the case, the data could be used to trigger the oscilloscope. This is not generally recommended because it no longer captures the data relative to the clock. Figure 4 shows the difference between using the clock and the data as the trigger source. Short bit sequences in the figure depict typical errors in the transition location.

When the clock is the trigger source, the alignment error is independent of prior errors. When the data is the trigger source, errors in the location of transitions (early or late) can be compounded if the edge at which the trigger occurs also has a location error. It is obvious that some data-dependent effects will be overly enhanced. Some may even be hidden.

When no clock is available, the trigger source can be synthesized. A clock recovery circuit can extract the data transitions and synchronize a clock to them. The clock can then be applied to the trigger.

This solution has some significant advantages if used properly. The clock recovery process will track (reject) a certain amount of lower-frequency jitter and align each clock edge to the data. Alternately, the clock recovery circuit will also not track (pass) higher-frequency jitter.

The question to ask in such a situation is whether or not the clock recovery circuit matches the expected behavior of the device that will be receiving the DUT’s transmission. If the clock recovery circuit does not match the expected behavior of the DUT’s receiver, be cautious. It might lead to failing good parts or passing bad parts.


Eye Sight

 

Now that you know how to create an eye pattern, what can you see in it?


The eye pattern shows the combined effects and interactions of the bit sequences in the data. Nearly all of the data-dependent effects and waveform characteristics can be assessed with one waveform and a mask, saving valuable test time while demonstrating performance with more realistic conditions.

Another required measurement that can be extracted from the eye pattern is an approximation of the eye opening. There is a common misconception that the eye opening of a data transmission can be directly measured from an oscilloscope eye pattern. Although eye opening means the open region between the crossings in the eye, no oscilloscopes today capture enough data without interruptions to ensure that the region is indeed void of any data transitions.

The definition for the eye opening is the region in which the bit error rate (BER) is less than some number, 10-12, for example. The best oscilloscopes still have about 10 µs of dead time between acquisitions. For a 1-Gb/s signal (109 bits/s), only one out of every 104 bit intervals is sampled or captured.

Trying to capture an error that occurs once every 1012 events is practically impossible with an oscilloscope. Instead BER testers can be used because they continuously count errors. They search the eye by iteratively measuring the BER at different times relative to the data transitions.

The search is conducted looking for the boundaries where the error rate is higher, such as 10-9. Once the boundaries are determined, additional and longer-timed measurements are taken near the boundaries where a lower error rate is anticipated. The objective is to gather enough data to extrapolate where the data’s eye will intercept the required error rate.

Extrapolation is used because the time to measure the actual error performance may take 30 minutes to a few hours. Of course, the extrapolation is based upon certain assumptions and approximations. Alternately, there is a way to estimate the eye opening for a given error rate using the eye pattern on an oscilloscope.

The data characteristics that reduce the eye’s opening can be grouped in two categories: deterministic and random jitter. Deterministic effects are dependent upon the pattern transmitted. Long runs of logic 0s followed by a single logic 1 will often yield a slightly shorter pulse with a slightly late arrival when compared with other transitions. Random effects include the remaining jitter that is uncorrelated to the data. Deterministic jitter can be found by transmitting a short repeating stress pattern and measuring the peak-to-peak width of the eye’s crossing. Random jitter can be measured by using another pattern that is uniform with 50% duty factor. A common pattern is five logic 1s followed by five logic 0s.

As the oscilloscope’s eye pattern represents only a sample of the total population, a peak-to-peak measurement of random jitter is not appropriate. For this, the standard deviation of the samples at the crossing must be measured. Figure 5 shows that the oscilloscope display represents a sample population that easily covers ±2 standard deviations (s ) but not much more. Assuming that the population is a normal distribution, the peak-to-peak random jitter can be estimated for the population of good samples.

For example, with an error rate of 10-12, the good population is found within ±7s of the mean value. To convert the rms random jitter, multiply by 14. The estimated eye opening is given as:

Opening = (Bit Interval) – (Deterministic Jitter) – (rms Random Jitter) x 14

Although it is likely that bit error will occur when the eye is barely open, there is no way to know that a very wide open eye is error free. The eye pattern can only show that the waveforms support a level of quality in the transmission.

The eye pattern will not show that a bit transition occurred at the wrong eye crossing, only that the transitions occurred within legal limits. The eye pattern shows waveforms, not the content of the data transmitted; therefore, the eye pattern cannot report BER.


Conclusion

 

Eye pattern analysis is an effective test and evaluation tool. Understanding the tool and its advantages and limitations will yield reliable and repeatable test methods that will improve the quality of your designs.

About the Author

Clark Foley is a Product Development Manager for the VXI Product Line at Tektronix. He has been affiliated with the company for 20 years, working nine years in engineering as a designer and program manager, nine years in marketing and two years in manufacturing as a sustaining engineer. Mr. Foley holds three patents and has a B.S.E.E. degree from Arizona State University. Tektronix, P.O. Box 500, M/S 39-122, Beaverton, OR 97077, (503) 627-6069.

Copyright 1996 Nelson Publishing Inc.

May 1996



Comments

To join the conversation, and become an exclusive member of Electronic Design, create an account today!