Every scientist, engineer, and technician involved in any form of electronics has used an oscilloscope. Scope displays of amplitude as a function of time provide intuitive and easily interpreted pictures of signals. The high-speed communications industry is no exception to the widespread use of scopes. Workers researching, developing, and manufacturing components and communications-network equipment extensively use ultrawideband sequential-sampling oscilloscopes—commonly called DCAs (digital communications analyzers) or CSAs (communications signal analyzers)—to inspect eye-diagram waveforms and to quickly and easily ascertain signals’ general health.
Although producing an eye-diagram display is a straightforward procedure, several pitfalls can lead to inaccurate or erroneous measurements. You can easily recognize and solve some of these problems. Others are subtle and can be difficult to detect and correct.
Figure 1
These eye-diagram displays for a 128-bit pattern look materially different when triggered by a full-rate clock (a), a divide-by-8 clock (b), and the data signal itself (c). |
Whereas the Y-axis of the DCA display represents signal amplitude, it is not completely correct to say that the X-axis represents time. To be exact, the X-axis represents time relative to the occurrence of a specific event. This timing event is commonly called a trigger. Improperly triggering the DCA causes many measurement problems.
A trigger signal is typically an electrical signal’s rising or falling transition. When you supply a stream of data to a DCA, the instrument displays an eye diagram if the trigger transition occurs synchronously with a large number of bits in the data stream. Typically, you measure only a small number of bits. This technique is generally valid as long as the sampling yields a good representation of the overall signal. Trigger-signal candidates include the system clock used in generating the measured data, a divided clock (synchronous with the data), and the data itself.
When the DCA responds to a trigger event, you measure a portion of the bit stream. As other triggers arrive, you measure different portions of the data. With persistence, the display builds up with several waveforms. The various combinations of signals overlay on each other and create the eye diagram. (To envision how this build-up occurs, consider all combinations of a 3-bit binary sequence. If you draw the eight associated waveforms 000, 001...111 on a common time axis, you see an eye diagram.) The three trigger signals (clock, divided clock, and data) can each generate a slightly different eye diagram.
FINDING THE OPTIMAL TRIGGER SIGNAL A typical test scenario occurs when the data stream’s original source is a PG (pattern generator), which in turn feeds the device or system under test. A critical measurement for a communications signal is jitter. Jitter is a measure of how the transmitted bits stray from their ideal positions in time. Thus, as the data stream propagates through the test device and is modified by any jitter-producing mechanism, bits slip from where they should be. You want the DCA to display this effect. Recall that the DCA X-axis represents time relative to the trigger event. The pristine PG clock signal precisely represents the ideal time positions for the data. Thus, in this scenario, the PG’s clock signal provides an ideal trigger. Using the PG clock as a trigger precisely displays any retarded or advanced edge in relation to its ideal position.
What about a divided clock? In almost all cases, a divided clock trigger provides results as accurately as the full-rate clock, except in two cases. If dividing the clock signal causes it to exhibit any jitter that is not common to the PG data signal, the divided clock becomes corrupted to some degree, and the DCA’s displayed jitter is in error. The second exception occurs when the length of the data pattern is a multiple of the clock-division ratio. For example, consider a pattern of 16 bits in which a divided-by-4 clock triggers the DCA. The first clock edge causes the measurement of bit 1. The next clock edge causes the measurement of bit 5 and so on with bits 9 and 13. The next clock edge causes the process to repeat. The problem is that you never measure or include bits 2, 3, 4, 6, 7, 8, 10, 11, 12, 14, 15, or 16 as part of the single displayed eye diagram. The eye does not represent the overall signal. Often, you can easily observe this effect as an incomplete eye. (Perhaps the leading 1 to 0 transition is missing.) As you increase the ratio of pattern length to division factor, you measure a larger percentage of the total pattern. Thus you reduce the problem, but you don’t eliminate it. However, if the pattern length isn’t an integer multiple of the clock-divide ratio, or if the division ratio is 1 (full clock rate), multiple passes through the repeating pattern cause the eye to build up and include waveforms from every bit in the pattern.
You can make many oscilloscope measurements without using an external trigger signal. By splitting off a portion of the signal as a trigger, you make the test signal itself into the timing reference. You can use this approach for eye-diagram analysis, but doing so poses two significant problems. Consider the jitter measurement. Rather than determining waveform stability by comparing the waveform with a stable reference, such as a PG or system clock, triggering on the data effectively compares the signal with itself. When you examine it at a time close to the trigger event, the jitter effectively becomes a common-mode effect and cancels itself out. Large amounts of jitter may incorrectly appear to be quite small. Conversely, if you measure the data at a time far away from the trigger event (that is, with significant DCA-timebase delay), the trigger jitter is no longer in phase with the data jitter and may constructively interfere, making the jitter appear to be larger than it really is. The bottom line is that when triggering on the signal itself, jitter measurements are likely to be inaccurate.
A second problem with triggering on the data itself is that you never measure a significant portion of the data pattern. If you configure the DCA to trigger on a rising edge, only the 0-1 bit sequence produces a trigger event; the 1-1, 1-0, and 0-0 sequences never do. When measuring repeating data patterns, you never measure 75% of the bits. It is easy to see that the eye diagram is incomplete when viewing data close to the trigger event. By increasing the timebase delay, you eventually display eye diagrams that appear complete. However, even though the eyes appear complete, the fact remains that you never measure 75% of the bits (
Figure 1).
DERIVING A TRIGGER SIGNAL FROM THE DATASometimes, you can’t use any of the three already mentioned trigger types. In such cases, you can derive a timing reference from the test signal itself through some form of clock-recovery circuit. The accuracy depends upon the clock-recovery system’s loop bandwidth. A clock-recovery system with narrow loop bandwidth transfers very little jitter from the data being tested to the derived trigger. The narrower the loop bandwidth, the more the recovered clock resembles a clean system clock.
Figure 2
In both of these waveforms, the data actually has 0.25 unit intervals of fast jitter. However, deriving the trigger from a narrowband loop (a) accurately displays the jitter, whereas deriving the trigger from a wideband loop (b) effectively removes the jitter, possibly leading to an unintended measurement error. |
On the other hand, a wide loop bandwidth can be useful for measuring signals that have propagated through a complex network and have large amounts of jitter. The wide loop bandwidth allows the recovered clock to track the data. When you trigger with this clock, you can clean up most of the jitter. Although this approach obviously prevents quantifying the jitter, it does allow measuring other eye parameters. The critical point to remember when using recovered clocks for triggering is that, unless the scheme has a very narrow loop bandwidth, jitter-measurement accuracy is suspect. The same jitter-measurement problems that exist with data triggering occur when you trigger from a recovered clock (
Figure 2).
The DCA’s frequency response also is critical for accurate waveform analysis. It is well known that distortion can appear in a measured signal that has significant frequency content beyond the oscilloscope channel’s -3-dB bandwidth. Specifically, as scope bandwidth decreases, edge speeds appear to decrease, and rapidly changing waveform features appear to “soften.” However, the -3-dB bandwidth alone may be an insufficient indicator of waveform fidelity. If the high-frequency roll-off peaks, or falls off too abruptly, it can create artificial overshoot and ringing in the displayed waveform. Although you can roughly compensate for edge-speed degradation by backing out the speed/bandwidth of the DCA (that is, by pre-emphasizing frequencies beyond the -3-dB frequency), it is much more difficult to correct a distorted waveshape. It is difficult to sort out whether the distortion is inherent in the waveform or the measuring instrument has caused it. (Knowing the DCA’s impulse response can help to indicate the worst-case distortion the instrument can cause.) The bottom line in this situation is that, when all you know is the DCA’s -3-dB bandwidth, you often can’t easily assess the trade-off between speed and waveform fidelity.
Whenever you make a measurement, it is important to have confidence that you are achieving the highest possible accuracy. Having a fundamental understanding of some of the problems that can exist when you make eye-diagram measurements is the first step to preventing these problems from affecting your measurements.
--------------------------------------------------------------------------------------------------------------------------
Author InformationGreg LeCheminant is a product-marketing engineer responsible for developing and improving measurement tools and techniques for Agilent Technologies’ Lightwave Division in Santa Rosa, CA. His work has included developments related to digital-communications analyzers and jitter analyzers.
