by Dan Strassberg, Contributing Technical Editor, EDN
People often say that EEs are almost obscenely fortunate to have a tool that provides as much insight into fundamentally invisible processes as do oscilloscopes into the internal workings of electronic circuits and systems; no other profession has a tool that reveals as much. Despite the embarrassment of riches that scopes afford their users, manufacturers continue to find ways to make the instruments more valuable. Unquestionably, the old cries of “faster” (referring both to bandwidth and sampling rate), “deeper” (referring to depth of acquisition memory), and “less costly” continue to motivate scope designers. But the ways to make scopes even more useful are growing—seemingly just as fast as are bandwidth, sampling rate, and memory depth.
Over the past few years, scopes’ analytical and computational prowess has shown no signs of slowing its ascent. Adding analytical capabilities is, however, only part of the challenge of designing computationally intensive oscilloscopes. Another important part is ensuring that new capabilities of mind-boggling sophistication don’t actually boggle the minds of the target users. A scope is probably better off without features that are so difficult to operate that users give up trying to make them work. Scope designers often liken their progeny to motor vehicles and refer to the panoply of usability issues under the heading of “How an instrument ‘drives.’”
As important as scopes are in EEs’ jobs, most engineers still regard the instruments as mere tools—adjuncts to accomplishing the task at hand, not the objects of the work. Greater ease of use both responds to and encourages this attitude; when you can make a measurement without giving the technique much thought, it is comforting to believe that the procedure merits little thought. Moreover, in this era of constricted schedules and budgets, there is rarely time to think about problems that seem peripheral to completing a job. Alas, such thinking can be dangerous (see box, “Calibrating scopes’ high-frequency amplitude accuracy: more difficult than you might think”).
Modern scopes make inherently difficult measurements seem easy, but, all too often, the measurements are less straightforward than they appear. Failure to recognize this fact and to understand the instrument and the technique can lead to erroneous or meaningless results— whose lack of validity can go unrecognized until the consequences become painfully obvious and corrective action is prohibitively expensive.
A fool’s errand Becoming enough of a scope expert to select the best unit for your application and to use the instrument in the most advantageous possible way requires effort. Indeed, some say that attempts to find the best scope or to most effectively use it are fools’ errands. To begin with, no two engineers are likely to agree on definitions of “best” and “most advantageous” in the context of selecting and using scopes. Second, data sheets, the principal presale documents by which engineers select scopes, have become voluminous, sometimes exceeding 30 pages packed with footnotes and fine print. Third, many mid-market scopes and nearly all high-end units are now PC-based, which usually means based on standard versions of Windows. In such instruments, a Windows-based software application determines how you access the multitude of scope features.
The complexity of scope applications is certainly at least comparable with that of common shrink-wrapped office-software applications, such as Word and Excel from Microsoft (www.microsoft.com). Most office application users avail themselves of only a small fraction of the software features. So it is with scope users. Moreover, a common problem for many scope users is that they don’t use the instruments every day, yet, when they slide into the driver’s seat, they need to quickly get answers to their questions about the unit or device under test. In other words, the methods of accessing and using scope features should be intuitive—conforming, wherever possible, to the conventions with which the users are familiar.
Scope manufacturers point out that—at least with high-end instruments—your most valuable ally in selecting and effectively using the right instrument can be the field engineer who sold you the scope—or is trying to sell it to you. He can help you set up side-by-side comparisons with competitive units before your purchase and can supply advice and accessories to help you effectively use the scope. Representatives of distributors that sell scopes may offer similar services. Also, don’t assume that factory support is unavailable to you because you purchased your instrument from a distributor. Depending on the manufacturer and the scope model you purchased, the factory may offer support. And remember that most scope vendors’ Web sites offer a wealth of application notes containing information on effective use of the companies’ products. Table 1 and Table 2 at the Web version of this article at www.edn.com/060216cs, summarize key specifications of real-time-sampling scopes from four major manufacturers.
Begins with probe An appropriate place to begin a discussion of modern scopes is with the probe. The probe tip is where the instrument meets the device under test. Time was that engineers considered only a few megahertz to be a high frequency. Now, probing gigahertz signals is commonplace, and familiar serial buses transmit signals at rates in excess of 3Gbps. Scope manufacturers recommend that your scope and probe together have a -3dB bandwidth at least 1.8 times the bit rate. So, if you are working with a bus whose raw bit rate is 3.125Gbps, your scope and probe should have a combined bandwidth of at least 5.625GHz. (A bus with a raw bit rate of 3.125Gbps usually carries information at 2.5Gbps; 8-bit/10-bit clocking embedded within the data stream limits the information rate to 80% of the raw bit rate.) The bandwidth closest to 5.625GHz that scope manufacturers advertise is 6GHz. The 6.67% margin above 5.625GHz can help to compensate for bandwidth reduction attributable to the probe.
Several points are important. The first is that probing such highspeed serial buses is a job for differential active probes. At these speeds, nearly all buses are differential, and the signal swings are small for a variety of sound reasons: Unlike single-ended circuits, differential receivers tend to reject common-mode “noise,” enabling the use of smaller signal swings; differential circuits also radiate less noise and subject power-supply rails to less transient loading than do single-ended circuits. But the smaller signal swings militate against passive probes, which, to reduce capacitive loading, generally attenuate their input signals. In addition, using two scope inputs to view one differential signal is out of the question. That approach effectively not only halves the number of channels on your scope, but also provides input-terminal pairs that are inadequately matched at the frequencies involved. The result can be the appearance on the screen of waveform artifacts that don’t exist.
Multigigahertz-bandwidth differential active probes are amazingly clever, and their sophistication is likely to grow in the next few years. Manufacturers disagree about the best way to design and characterize these devices, but all manufacturers seem to agree on one fact: If you are trying to acquire multigigahertz signals, it is impossible to connect a probe to a unit under test without imposing some load on the signal you are trying to measure.
Manufacturers disagree, however, on whether that loading always has a meaningful effect on the waveforms you wish to view. Nevertheless, it is difficult to refute that, unless a probe is designed with the utmost care, the loading effects not only can be meaningful, but also can make unacceptable waveforms appear perfectly fine or vice versa. For example, probe induced errors can cause what is, in fact, a good waveform to appear to violate an eye-diagram mask or can make a waveform that violates the mask appear to comply.
That probes impose capacitive loads on units under test is well-known. However, a probe’s series inductance is also important in determining the probe’s response at several gigahertz. Moreover, the resonance between the probe’s shunt capacitance and series inductance can have even more dramatic effects both on the loading of the unit under test and on the probe’s frequency and transient response.
Probes get smarter Modern probing systems from all of the major scope manufacturers include facilities for bidirectional communication between the scope and the probe. Modern active probes do more than merely send the scope an amplified or buffered replica of the waveform at the probe tips, and the scope does more than just supply power to such probes. For example, LeCroy’s newest probes store dynamic probe-calibration data.
This data includes more than just the probe’s offset voltage and dc gain; it includes high-frequency gain- and phase (delay)- characterization data. According to Mike Lauterbach, PhD, LeCroy’s director of product management, all ultrawideband scopes from all manufacturers use DSP-based techniques to correct the vertical amplifier high-frequency-gain and -phase characteristics. The corrections improve the response so that it more closely resembles the desired response—often that of a fourth-order Bessel lowpass filter—than does the amplifier’s uncorrected response.
As far as Lauterbach knows, however, only LeCroy’s WaveLink probe family currently includes the probe response in the correction algorithm. Within seconds of your connecting a WaveLink probe to a compatible LeCroy scope, the correction routine uploads the calibration data from the probe and compensates the channel’s vertical response for the probe’s ac characteristics (as measured at the factory—or the last time you used a LeCroy-supplied fixture to characterize the probe). Including the probe in the calibration enables LeCroy, whose 11GHz scopes offer narrower -3dB bandwidth than that of the nearest competitive models from Agilent or Tektronix, to nevertheless claim the most accurate high-frequency ac and transient response among real-time scopes in the more-than-10GHz class. LeCroy also points out that, unlike at least one competitor, it does not currently use DSP to extend the bandwidth of its scopes.
In case you haven’t noticed, modern wideband scopes do not have frequency response related to the 10 to 90% rise time by the time honored formula TR=0.35/BW, where TR=10 to 90% rise time and BW=-3dB bandwidth. And you can’t determine the combined rise time of the scope and probe from.
For one thing, you must carefully check the data sheet’s notes to determine whether each rise-time spec applies to the time the signal takes to traverse 10 to 90% or 20 to 80% of the input-step amplitude. Manufacturers sometimes specify both rise times. Some standards for bus physical layers use only the 20 to 80% values; using 10 to 90% values in such cases would only cause confusion. In addition to the “which-rise-time?” issue, however, the old formulas don’t apply to new scopes and probes because the newer units’ high-frequency-roll-off characteristics differ from those of the analog scopes whose behavior formed the basis for the old rules. To learn more about deep memory and finding ephemeral anomalies in long-wave-form records, see box “Acquisition memory: a deep subject” at the Web version of this article at www.edn.com/060216cs
It takes persistence Persistence mode doesn’t work quite the way many people think it does (Figure 1). To dispel the confusion, here is a brief explanation that generally applies to all scope brands. Note that persistence mode can often correctly acquire waveforms that— because of a limited real-time sampling rate—contain frequencies too high for the scope to capture in real time. Many scope users erroneously believe that capturing such waveforms requires using random equivalent-time sampling, a mode you must use with caution to avoid little-understood pitfalls (Reference 1).
To use persistence mode, the trigger must be stable in time with respect to the waveform that you want to capture. You can trigger on a waveform feature or use another trigger source. Each time it triggers, the scope acquires waveform samples and places the corresponding dots on the screen with respect to the trigger time. It draws no line between the dots, though. By default, some scopes add sine x/x-interpolated dots, whereas others add none. The scope simply places the dots on the screen—or, to be more exact, it places the dots in an array in the display-processor IC, which draws the dots on the screen. The scope draws no line through the dots, however, and makes no attempt to re-create the shape of the incoming signal; such an attempt could violate the Nyquist criterion.
The scope then triggers repeatedly. Typically, it triggers several hundreds—or even thousands—of times. Each time, it acquires samples and places the dots on the screen, but it never attempts to “draw the trace.” The scope simply displays the acquired samples with respect to the trigger time. If the trigger and the incoming waveform are stable, the set of dots is closely packed onto a line shaped like the signal and strongly resembling a waveform. If the trigger time or the waveform is unstable because of vertical noise or timing jitter, the persistence display shows a cloudier set of dots. If the signal shape exhibits occasional large, intermittent aberrations, you may see a large number of dots that follow the normal signal shape and a fainter number that show the abnormal shape.
Slow refresh Scope manufacturers make much of their instruments’ fast screen-update rates and responsiveness to changes in control settings. Some companies refer to such attributes as “analogscope feel.” These claims are valid as well as important to the way in which you use a scope, but, if you think about the claims for a few moments, you can easily wonder how they can possibly not be exaggerations. Nearly all digital scopes refresh their screens only 30 or 60 times per second, yet many display many thousands of waveforms per second. They achieve this responsiveness by aggregating multiple changes to their screen bit maps between refreshes and displaying the aggregated result at the next refresh.
This aspect of scope operation is philosophically similar to the way in which scopes whose displays have, say, 1024 pixels horizontally display million-point-deep records without forcing you to scroll endlessly through the long records. However, you can zoom to that mode, as well, if you choose. The simplest way to compress a million samples into 1000 pixel columns, each representing 1000 samples, is to find the minimum and maximum signal values in each 1000-sample group and illuminate all pixels in the column from the one that corresponds to the lowest value to the one that corresponds to the highest. This approach produces a “fat” trace, whose illumination is constant over its width. To show greater signal detail, a scope can determine how many times since the last screen update the signal level corresponded to each point in the screen’s pixel map and relate each pixel’s brightness or color to the number of “hits” at the associated point.
Scope manufacturers are also discovering the value of the big screen—not the living-room dominating size of an HDTV and not even the wide aspect ratio of the screens on some laptop PCs but considerably larger in area than has been customary in scopes. Bigger screens on scopes make it easier to see waveform details. LeCroy started the trend a couple of years ago—at least in small-footprint scopes—with its WaveSurfer family, whose members sport 10.4-in.-diagonal, SVGA, 800 600-pixel screens in a 6-in.-deep package that occupies no more benchtop area than does a Tektronix TDS3000B, whose screen measures only 6.4-in. The WaveSurfer’s screen area is more than 2.5 times as great as that of the Tek unit. Now, LeCroy has added higher performance units to its stable of large-screen, small footprint scopes. The three members of the WaveRunner Xi series, whose prices start at $7500, are the same size as the WaveSurfers and also have 10.4-in. SVGA screens.
Not to be outdone, Tek, with its new DPO7000 series, has one upped LeCroy on screen size and resolution. The DPO7000 screens measure 12.1 in. diagonal. Their area is approximately 3.6 times that of a 6.4-in screen, and they provide XGA, 1024 768-pixel resolution. The approximately 12-in. package depth is roughly twice as great as that of LeCroy’s small-package units but is much shallower than most scopes. The DPO7000s, whose top of-the-line unit can accommodate memory as deep as 400M samples—all of which is assignable to one channel—also attack LeCroy’s long-held dominance in memory depth.
Although welcoming the large screens and small benchtop footprints, engineers who incorporate scopes into larger systems—for example, for production test—may be less than thrilled with the new package geometries. For these engineers, selecting system components that occupy a minimum of rack space is of key importance. The new packages are taller than those of most traditional scopes. It seems likely that the solution to the height problem will lie in LXI (LAN extensions for instrumentation), a new standard for systemcomponent instruments. You can imagine low-profile LXI scopes whose screens lie flat a top them until an operator pulls them forward on their rack slides and hinges the screen into a vertical position.
Beyond 20GHz A survey of the current state of digital-scope technology would be incomplete without some discussion of the widest bandwidth scopes—the class of instruments that engineers used to call sequential-sampling scopes. Until the advent, a year ago, of LeCroy’s WaveExpert and SDA100G series, the phrase “sequential sampling” was appropriate, and there were only two vendors, Agilent and Tektronix.
The LeCroy units essentially rewrote the book on how engineers design these ultrawide-bandwidth instruments (70 to 100GHz, depending on the manufacturer). At the product introduction, LeCroy referred to its instruments simply as sampling scopes, because “sequential” didn’t really apply. But the problem with not qualifying “sampling” is that all digital scopes are sampling scopes. During the year, LeCroy solved its terminology problem by inventing a new term, “NRO” (near-real-time oscilloscope) and adding an NRO series to its line.
All scopes in this category including the LeCroy units—depend on the signal’s occurring repetitively. It need not recur at a constant rate, but it must follow a trigger signal by an essentially fixed delay. Classic sequential-sampling scopes capture only one sample during each iteration of the input waveform, advancing the sampling point incrementally with each new trigger. Thus, despite their extremely wide bandwidth, these scopes acquire waveforms slowly. This low speed rules out instruments of this type in many common scope applications.
In some of these scopes, the analog sampler is separate from the scope mainframe. The sampler is a so-called zero-order hold circuit, which captures the input signal with fem to second aperture uncertainty and maintains the captured voltage for tens of microseconds. The sampler output is thus a relatively low-frequency replica of a multi-gigahertz signal. From the sampler output onward, the analog signals that the scope deals with are relatively low in frequency. The ADCs in such scopes are usually high-resolution (14 bits or more) successive approximation devices with conversion rates no higher than a few hundred kilohertz. Memory depths in classic sequential sampling scopes rarely exceed 100k samples.
Bandwidth to 100GHz Advances in sampling technology enable the LeCroy units, with the appropriate sampling plugins, to achieve industry-leading 100GHz bandwidth, whereas advances in ADC and memory technology make possible an architecture that differs considerably from that of sequential-sampling instruments. Instead of taking only one sample during each iteration of the input waveform, the LeCroy units take many. The company says that the sampling rate is 50 times that of the fastest competitive instrument. In addition, memory depths of hundreds of millions of samples are possible, and a built-in clock recovery facility allows the scopes to operate, in many cases, without an external trigger. The scopes also accommodate built-in analysis features that you would probably expect to find only in real-time sampling scopes. Thus, these scopes can handle many applications in which competitive instruments would acquire data too slowly, could not capture records of the necessary length, would require external equipment to trigger from the available signals, or would present a more complex interface to less-extensive analysis facilities.
As do Agilent and Tek, LeCroy offers optical-to-electrical converters to permit use of its ultrawideband scopes for fiberoptic communication-system measurements. Unlike its competitors, though, LeCroy does not currently offer differential-input plug-ins for these scopes. As a result, you need two of the LeCroy mainframes to simultaneously view four more-than-20GHz differential signals—a task the competitive units can perform with one mainframe.
Author information Contributing Technical Editor Dan Strassberg has been covering test and measurement for EDN for more than 18 years. He holds a bachelor’s degree in electrical engineering from Rensselaer Polytechnic Institute (Troy, NY) and a master’s degree in electrical engineering from Massachusetts Institute of Technology (Cambridge). You can reach him at strassbergedn@att.net.
For more information... For more information on products such as those discussed in this article, contact any ofs the following manufacturers directly, and please let them know you read about their products in EDN Asia.
Calibrating scopes’ high-frequency amplitude accuracy: more difficult than you might think
by Steve Sekel, LeCroy Corp.
Customer questions and complaints about scope amplitude accuracy are fairly common. Customers try to measure the accuracy with a swept sine wave from a signal generator. Users shouldn’t try this procedure themselves. Although the measurement sounds legitimate, the results are almost always wrong when the frequencies are higher than a couple of gigahertz.
The first problem is that you need to level the generator output at the output end of the cable. Even the best cables those that cost more than $1000—have some amplitude loss when you get to the several gigahertz range. The only way to use a signal generator to measure amplitude accuracy is to use a high-quality, calibrated power divider at the end of the cable that connects to the oscilloscope.
One output of the power divider is connected directly to the power head of an RFpower meter that is calibrated for the frequency range and power levels you are testing. If you are testing all of the volts/division ranges, this measurement often requires using more than one power head. The power-meter readings normalize the output level at each frequency step. In an automated-calibration system, you perform this procedure under computer control. It is possible but tedious to manually perform the procedure.
Reflections The second problem, which undoubtedly occurs in many cases, is dealing with the reflections from the scope input. In reality, the user is measuring the signal with the reflections superimposed. Scope inputs are not perfect 50Ω terminations. Different attenuators switch using relays or electronic switching. Inevitably, the paths are imperfect; they introduce some reflections at different frequencies.
Scope vendors work to minimize these reflections, but they all achieve about the same performance: a VSWR (voltage-standing-wave ratio) that, over the passband, can go from a perfect 1-to-1 to about 1.35-to-1. Whenever the termination reflects energy back into the line, the reflection creates standing waves at some frequency that relates to the length of the cable. Because they exhibit reflections at different frequencies, different models of oscilloscopes measure different amplitudes from the same generator-and-cable combination.
A user can reduce this effect by installing a high- quality, 6-dB attenuator at the scope’s input and attaching the power-divider output to the attenuator. The attenuator improves the return loss by 6 dB, reducing the effect of the reflection in the cable.
As you can see, the metrology required to accurately measure a scope’s amplitude accuracy over frequency is complex. All scope manufacturers put considerable effort into designing and verifying the complex systems that designers use to calibrate the instruments. Attempting to manually replicate this measurement by using only a signal generator and cable can’t produce results of the desired accuracy.
Steve Sekel is a product marketing manager at LeCroy Corp. He has worked in the test-instrumentation industry for 28 years, serving in marketing, product development management, and design engineering. He has a bachelor’s degree in electrical engineering from Tri-State University (Angola, IN).
