Dos & Don'ts: getting the most from wideband scopes

Modern high-performance digital oscilloscopes offer a wealth of features for making users' lives easier—so many features, in fact, that their effect can sometimes be the opposite of what the manufacturer intended. Whereas, typically, the presence of extra capabilities doesn't complicate instrument use, figuring out whether a wanted feature is present and learning how to invoke and properly use it sometimes can induce considerable frustration. Tutorials built into top-of-the-line scopes' help facilities simplify the job of getting up to speed, but manufacturers caution users to adopt the proper mindset when learning how to use unfamiliar features. As Mike Lauterbach, PhD, director of product management at LeCroy, puts it, "Being in too much of a hurry can waste lots of time."

Moreover, when you need to verify a design's compliance with a high-speed serial-data-transmission standard, you are likely to encounter an annoying reality of high-data-rate protocols. According to Lon Hintze, DSO 80000 product manager at Agilent Technologies, "Different protocols specify different methods of measuring similar things." Just because you know how to measure, say, the effect of bandwidth on ISI (intersymbol interference) in a XAUI (10-Gbps attachment-unit-Interface) system doesn't mean that you will necessarily get the correct answer if you apply an identical approach to measuring the equivalent effect in a PCI Express system.

Manufacturers try to have their scopes' help facilities provide the information you need to correctly perform widely used tests, but, if you are working with a new protocol, you should check the instrument manufacturer's Web site before attempting to make a measurement that conforms to an industry standard. The measurement technique may have changed since your scope was produced, and, in the words of Mike Engbretson, program manager, High Performance Oscilloscope Solutions at Tektronix, "To avoid using an outdated procedure, you may have to download and install a new version of the help file or the file that sets up the scope to run the test."

Attempting to learn how to perform all of the procedures you will ever have to follow before you must perform them is a doomed approach. There are simply too many such procedures. Moreover, it is unrealistic to expect to know a year in advance—much less two or three years beforehand—the nature of the projects you will be working on. You'll be lucky if you can verify that the scope you buy today can do everything you will need it to do six months from now. Take heart, though, because the instrument hardware usually can perform additional tasks, and a simple download may be all your scope needs to equip it to perform tests you hadn't thought of at the time of purchase.

Sometimes, though, you won't be so lucky. For example, suppose you choose a sequential-equivalent-time-sampling scope because its bandwidth and resolution are greater and its price is lower than those of top-of-the-line real-time-sampling instruments. In addition, you decide that the sequential scope's requirement for repetitive inputs and its relatively low measurement speed are acceptable in your application. Moreover, after researching standardized tests that you may have to perform, you find none that demands the use of a real-time-sampling scope.

Then, six months later, you conclude that you must measure cycle-to-cycle jitter. At that point, you need a real-time-sampling scope. If you only occasionally require its capabilities, renting a scope probably makes perfect sense. But if you need those capabilities nearly every day, your company may decide that its best course is to purchase an instrument whose cost—depending on your performance requirements—can exceed $100,000 with a full set of accessories. Figure 1 shows tasks that are appropriate for real-time-sampling scopes and sequential-sampling scopes and tasks that you can perform with either type of scope.

One of the beauties of real-time-sampling scopes for debugging high-speed serial buses is the deep memory available on many units. A few scopes offer memory as deep as 100 million samples per channel—enough to capture 5-msec records at the maximum sampling rate of 20G samples/sec. Once the data is in memory, you can store it indefinitely on the scope's hard disk and subject it to almost limitless analyses. Moreover, with the sophisticated triggering functions available on these scopes, you can capture long records that contain rare events, including sequences of events that precede failures that you want to investigate. The ability to reproduce these situations at will can save enormous amounts of time that you would otherwise spend waiting for these seemingly random events to occur.

To avoid aliasing—the display of frequency components that aren't, in fact, present in a signal—real-time-sampling scopes must sample at more than twice the highest frequency present in their input signals at significant amplitude—usually defined as no more than 1 LSB of the scope's ADC resolution. In practice, a real-time-sampling scope's sampling rate is at least 2.5 times its –3-dB frequency and more commonly four or more times that frequency. To present a smooth display—as opposed to a series of dots—real-time-sampling scopes use a reconstruction filtering, a DSP technique that interpolates or fills in the waveform values between the sampled points. The most common reconstruction algorithm is sin x/x interpolation. The simplest is linear interpolation.

Some real-time-sampling instruments whose –3-dB bandwidth exceeds half their maximum sampling rate also offer a random-equivalent-time-sampling, or RIS (random-interleaved-sampling), mode. Like sequential sampling, RIS is useful only with repetitive inputs and creates an effective sampling rate that is higher than the scope's maximum real-time-sampling rate. Unlike sequential sampling, however, RIS is neither a method of avoiding an input amplifier's bandwidth limitations nor a way of increasing a scope's vertical resolution beyond that possible with the fast ADCs that such scopes use. In RIS mode, a scope samples its input waveform several times after each trigger event, but the sampling is sparse, and multiple triggers are necessary to fill in the missing samples and accurately reconstruct the waveform. In RIS mode, scopes don't usually use reconstruction filtering.

Unlike real-time-sampling scopes, sequential-sampling scopes intentionally undersample their input signals—that is, they sample them at much less than twice the frequency of the highest frequency significant-amplitude component. These scopes take only one sample after each trigger event, and each new sample occurs later within the waveform than did its predecessor. The incremental delay between the trigger and each new sample equals the effective sampling interval.

In many such scopes, the sampling circuit is in a sampling head external to the scope mainframe. The sampling head's output is a low-repetition-rate pulse train whose pulse amplitudes constitute a reduced-frequency replica of the input signal. This architecture keeps all of the high-frequency information within the close confines of the sampling head and out of the mainframe, enabling effective bandwidths as high as 80 GHz. In contrast, one of the widest bandwidth real-time-sampling/RIS scopes offers a bandwidth of 13 GHz—less than one-sixth of 80 GHz.

The maximum sampling rate of sequential-sampling scopes rarely exceeds 200k samples/sec (approximately 1/400,000th of the –3-dB bandwidth), and, because hundreds—and often thousands—of samples are required to reconstruct an input waveform, such a scope can take many seconds to display a waveform. In fact, after taking a sample, the scope can't take its next sample until it receives a new trigger. If no trigger is immediately available, the actual sampling rate can be considerably slower than 200k samples/sec.

For many years, sequential-sampling scopes were the only instruments that could directly display multigigahertz waveforms. However, a major problem with using these scopes was their need to receive a stable trigger from the system under test. For the past several years, though, sequential-sampling scopes have been able to internally derive triggers from their vertical-input signals—even from signals that aren't perfectly repetitive. Nevertheless, according to Agilent Technologies, neither it nor Tektronix—the only other manufacturer of multigigahertz-bandwidth sequential-sampling scopes—has until now fully integrated trigger generation into the scope architecture. Agilent calls its new 84396A the first generator to meet the full range of user requirements for triggering these scopes. The 84396A is a plug-in for the 86100A DCA (digital-communications analyzer). DCA is Agilent's name for its sequential-sampling scopes; Tektronix's name for this product category is CSA (communications-signal analyzer).

One of the first questions you are likely to ask when evaluating a scope for potential purchase is: "How much bandwidth do I need?" The obvious answer—"as much as you can afford"—is sometimes incorrect. High-speed serial-data-transmission protocols usually use an NRZ (non-return-to-zero) data format, which puts two bits of information into one UI (unit interval), or clock period. Therefore, a 2.5-Gbps data-rate signal can have a clock period as long as 800 psec, which corresponds to a 1.25-GHz fundamental frequency. On the one hand, you might say that to correctly evaluate this signal, you need a scope whose bandwidth accurately reproduces the fifth harmonic of the 1.25-GHz fundamental frequency. You might then conclude that you need a scope with a minimum –3-dB bandwidth of 6 GHz, because, even a scope with that bandwidth would attenuate the fifth harmonic by slightly more than 3 dB.

However, most high-speed serial protocols use embedded clocking, which adds overhead that increases the fundamental frequency. The most common embedded-clocking scheme, 8B/10B (8-bit/10-bit) coding, increases the fundamental frequency by 25%—to 1.5625 GHz in this example. (The latest protocols, such as PCI Express, use clock-embedding schemes that add less overhead.) If you still believe you must reproduce the fifth harmonic with little attenuation, you would now shift your focus to scopes with bandwidth of at least 8 GHz.

However, if you are evaluating data-receiver ICs, you may want the scope's response to approximate as closely as possible the response of the circuits under test. Receiver ICs are in the business of accurately recovering data—which does not necessarily require reproducing pretty waveforms. So, enough bandwidth to reproduce the third harmonic with only a small error could be sufficient. Hence, a scope with 5-GHz bandwidth might meet your needs. Several standards recommend scopes with a bandwidth of 1.8 times the raw data rate. In this example, that rate would mean 1.8×3.125 GHz, or 5.625 GHz, which gets you back to the 6-GHz scope.

High-speed serial buses now mostly use differential signaling, which generates less noise, rejects common-mode interference, and allows smaller signal swings, which reduce signal-transition times and thus are more compatible with the high data rates than are single-ended signals. Differential signals travel over pairs of wires. In the past, engineers would often display these signals on a scope by taking the difference between two channels—a practice that is subject to many problems. The two channels may exhibit time skew, in which one signal can be delayed relative to the other, thereby indicating nonexistent signal preshoot. Also, failure to exactly equalize the two channels' gain and rise time can introduce additional artifacts.

All three high-performance-scope manufacturers offer differential active probes having bandwidth consistent with that of their scopes. Although a set of four of these probes can cost half as much as the scope itself, don't scrimp on probes; appropriate differential probes can and do eliminate the aforementioned problems. Moreover, they allow a four-channel scope to display four differential signals. As more and more serial buses adopt multilane architectures to further increase their throughput, one of the last things you want is to sacrifice half of your expensive scope's channels to produce misleading displays.

Light travels through free space at approximately 1 ft/nsec. Because of their higher dielectric constants, cables and pc boards slow the signal velocity to approximately 0.6 times the speed of light, or approximately 0.6 ft/nsec. In a 6-GHz sine wave, one cycle occurs every 1/6 nsec, In other words, the wavelength of a 6-GHz signal traveling through a pc board is approximately 0.1 ft, or 1.2 in. It should be no surprise, then, that probing the multigigahertz signals that travel in high-speed serial buses requires great care; at a little more than 7 GHz, a quarter-inch of wire can constitute a quarter-wave stub. If you can arrange it, by far the best way to connect multigigahertz signals to a scope is to use probes with high-frequency connectors, such as SMAs, and to design your pc board with SMA connectors at the points at which you expect to attach probes (Figure 2). When you have no room for or can't afford this approach, you may want to instead consider probes that attach to your boards in other ways, but such probes can't always deliver signal fidelity as good as that of SMA probes.

When working with high-speed serial buses and scopes, you are bound to spend a great deal of time measuring jitter, a type of measurement that can be misleading. No instrument can be completely free of jitter—although the instrument's internal jitter can be small compared with the UUT (unit-under-test) jitter that you are tying to measure. A simple test that users often fail to perform is to measure the jitter in the instrumentation. For this measurement, you need a signal source with very low jitter. The best such sources are high-quality, high-frequency sine-wave generators, although some pulse-pattern generators also produce low-jitter signals. Ideally, the generator manufacturer should specify the unit's jitter. You hope to observe lower jitter on the generator output than in the UUT. If that situation occurs, you can be confident that the scope's internal jitter is lower than the UUT's jitter. If the scope can compute rms jitter, you can determine the UUT jitter from  where J is rms jitter. If J_{SCOPE+GENERATOR} is smaller than J_SCOPE+UUT by a sufficient margin, you may be able to safely assume that the scope's internal jitter is zero.

You should also examine probe calibration. Eye diagrams are overlaid rectangular waves in which a mixture of high-to-low and low-to-high transitions occurs at each side of the eye. Failing to properly adjust the probe's rise and fall characteristics can lead you to believe that an eye diagram shows adequate margins around the forbidden region in the center of the eye, when, in fact, the margins are inadequate—or that it shows inadequate margins when the margins are adequate (Figure 3). Scope manufacturers furnish hardware for calibrating their probes, but many users unfortunately fail to perform the calibration procedures, apparently because they don't realize that 10 minutes or so of extra work can prevent days of wasted work.

Lastly, you should recognize that many standards now specify the measurement of transition times from 20 to 80% of final value, rather than from 10 to 90%, which used to be the industry standard. This situation is not a case of the junk spec driving the good spec out of the market. Rather, the 20 to 80% rise time is less ambiguous because significant ringing can appear on waveform edges. Moreover, high-speed-bus waveforms carry digital information so that the receiver must distinguish only between one and zero states. As a rule  where t is transition—that is, rise or fall—time. A scope whose 20 to 80% transition-time spec is 45 psec typically has a 10 to 90% transition-time spec of slightly more than 60 psec. Also be aware that the familiar expression, t=0.35/BW, where BW is the –3-dB bandwidth, does not apply to modern wideband digital scopes, whose lowpass characteristics do not emulate single-pole lowpass filters. Therefore, you cannot calculate the combined transition time of such a scope and its probe by determining the square root of the sum of the squares of the scope's and the probe's 10 to 90% transition times.

Author Information

Contributing Technical Editor Dan Strassberg is in his 18th year of covering test and measurement for EDN. Over the years, he has written extensively about digital scopes. Dan holds a bachelor's degree in electrical engineering from Rensselaer Polytechnic Institute (Troy, NY) and an master's degree in electrical engineering from the Massachusetts Institute of Technology (Cambridge). He is a registered professional engineer in Massachusetts.