Smart scopes:Spot the secret within waveforms

The economic downturn isn’t deterring manufacturers of DSOs (digital-storage oscilloscopes) from introducing models that incorporate exciting features that enhance the instruments’ utility and make hardware engineers more productive. Even during periods of slow or negative economic growth, new-product development usually continues relatively unabated. Because hardware-design engineers rely more on DSOs than on any other measurement tool, the instruments’ capabilities must advance at least as rapidly as the workings of the products the designers create.

The rampaging speed and complexity of units under test have moved DSOs’ job far beyond the instruments’ classical mission: acquiring and displaying waveforms. For years, DSOs have made numerous automated measurements. Examples include determining signal amplitudes, periods, frequencies, transition times (that is, rise and fall times), and slew rates. Now, analysis capabilities extract information, such as histograms, which waveform inspection doesn’t reveal. But making sense of today’s increasingly complex data requires still more sophisticated analysis that demands even greater processing power. This requirement results from users’ expectation that scopes will help them to make quick decisions. Although sending waveform records to a separate PC for offline analysis might seem to be an alternative, that approach alters the character of laboratory work, making a slow, laborious task of an activity that, in today’s fast-paced business world, must be rapid and highly interactive.

As waveform complexity increases, triggering—always a problematic function—assumes more importance and requires greater sophistication. Waveform memories also become deeper, because, as bandwidth increases, sampling rates must follow suit, and higher sampling rates proportionally increase the memory needed to hold records of a given duration. In addition, capturing waveforms of the same duration as those that instruments captured five years ago is often insufficient. Many of today’s complex electronic systems produce waveforms of considerably greater duration than those that simpler systems produced (Table 1). Deep waveform memory doesn’t come cheap either. Although DSOs use lots of DRAMs, scopes’ waveform memories usually consist of much more expensive SRAMs. Because they aren’t commodity parts, SRAMs are relatively insensitive to the market forces that periodically cause DRAM prices to plummet.



TECHNOLOGY'S UNEVEN PACE
Increases in scope bandwidth don’t occur at a constant rate. The uneven pace of the introduction of technology causes bandwidth to increase rapidly and then to remain at a plateau for a year or two. Currently, thanks to the commercialization of SiGe (silicon-germanium) IC technology, the latest real-time-sampling DSOs offer bandwidths as high as 6 GHz. A year ago, before foundries started cranking out SiGe parts in production quantities, bandwidths greater than 2 GHz were uncommon. With higher bandwidth comes the imperative for instrument manufacturers to produce wider-bandwidth probes. As a general rule, you can’t use a modern wideband scope’s full bandwidth in a practical setting without wideband active probes. Even so, LeCroy has just introduced 7.5-GHz-bandwidth passive 10-to-1 and 20-to-1 attenuating probes that have a mere 0.25 pF of input capacitance. Because the resulting input impedance—127W at 5 GHz—is largely reactive, it is high enough to permit effective probing in most 50W environments.

Tektronix is the undisputed market-share leader in nearly all sectors of the worldwide benchtop-DSO market in which it competes. Depending on what and how you count, either Agilent or LeCroy holds second place. Both companies agree that ownership of second place is, for all practical purposes, too close to call. In the United States, Yokogawa is in fourth place. Gould claims a large market share in Europe, but, in wideband scopes, its US market share is smaller than Yokogawa’s.

Figure 1 With –3-dB bandwidth of 6 Ghz and 20G-sample/s acquisition, Tektronix’s TDS 6604 is the current bandwidth leader among real-time-sampling DSOs. The scope achieves its bandwidth with the aid of input amplifiers that use SiGe technology, which the manufacturer was first to incorporate into standard-product DSOs.

With its $US57,500, 6-GHz TDS 6604, Tektronix is the current king of the hill in bandwidth (Figure 1). The company also offers a very broad DSO product line, which covers a range of prices and performance. The line includes the lunchbox-sized TDS 200 series; the highly portable TDS 3000B series; the TDS 5000, 6000, and 7000 series, which are designed for the benchtop; and the ultra-wideband, sequential-sampling, special-purpose CSA and TDS 8000 series.


QUESTIONS RAISEd
One feature of many Tektronix scopes probably raises more questions and more eyebrows than all other DSO features put together. That feature is the FastAcq mode of units that Tektronix calls DPOs (digital-phosphor oscilloscopes). Although Tektronix alone offers DPOs, the company hasn’t tried to patent the approach that underlies DPO technology—immediately converting incoming data into a 3-D pixel map in which the z axis represents how often each of the display’s pixels is illuminated. The instruments display the z-axis information by varying the trace color of color displays and varying the intensity of monochrome displays. Tektronix has, however, patented its implementation of the DPO approach, and the proprietary DPX ASICs in the company’s DPOs embody those patents.

DPOs’ forte is minimizing dead time between acquisitions. Dead time is the time a scope spends manipulating the captured data to generate the display. When you attempt to find the conditions that produce elusive anomalies in a circuit’s behavior, dead time can significantly impede acquiring the information you seek. Particularly at the highest sweep speeds, you can find that a conventional DSO acquires waveforms only a fraction of 1% of the time. Suppose that you don’t know what an anomaly looks like, but you know that it occurs, on average, approximately once in every 2 million triggers the unit under test generates. If your scope acquires data only 0.5% of the time, the unit must satisfy the trigger conditions an average of 200×2 million times, or 400 million times, before the scope catches a single occurrence of the anomaly.

If the unit under test satisfies the trigger conditions 10,000 times/s, the conventional scope makes you wait, on average, more than 11 hours to catch one anomalous waveform. Moreover, if you’ve never seen the anomaly and don’t know what it looks like, you probably don’t know how to set the scope to stop triggering after capturing the anomaly. If so, you may not notice the anomaly the first time the scope captures it, and you must wait even longer to see it. (Tektronix’s TDS 3000B-series scopes can decide which waveforms are likely to be anomalous and, on detecting an anomaly, sound an alarm or stop further triggering.)


MORE AND MORE WAVEFORMS/S
The highest performance DPOs overcome these problems by capturing more than 400,000 waveforms/s—if the unit under test satisfies the trigger conditions at least that often. In the previous example, a high-performance DPO would have no problems triggering every time the unit satisfied the trigger conditions. That is, the scope would trigger 10,000 times/s and would thus capture every anomalous waveform. In this case, the first such waveform would occur approximately 200 seconds (three minutes and 20 seconds) after you started observing.

However, there are downsides to DPOs. The fast acquisition of multiple waveforms takes place only in the FastAcq mode. In that mode, the scope does not preserve the data in its original form: as a data set that contains the digitized samples in the order in which they occurred. The FastAcq mode thus fails to preserve a record of the events that precede any occurrence of an anomaly.

To obtain information of that sort, you must capture new data in the normal-acquisition mode. In that mode, the scope can’t retrigger nearly as quickly as it can in the FastAcq mode. In the normal mode, DPOs’ dead times are hundreds or even thousands of times as great as their FastAcq-mode dead times. (The exact ratio depends on the model.) In addition, the dead times are 50 or more times as great as the normal-mode dead times of several competitive scopes. Those competitive scopes, of course, have no FastAcq mode. The FastAcq view of the anomaly does, however, provide at least one enormous advantage when you switch the scope to the normal mode: the FastAcq view often proves invaluable in setting the normal-mode trigger conditions.

On the other hand, in FastAcq mode, most DPOs can’t use their full memory depth nor can DPOs with the highest acquisition rates sample at their maximum speeds. Moreover, in FastAcq mode, other useful functions are also unavailable. One such is waveform math. Competitors assert that, because of FastAcq’s limitations, most scope users rarely use it and that DPO users are sacrificing normal-mode performance to obtain a little-used feature.


MOER THAN MEETS THE EYE
DSO displays typically offer resolution of only 640×480 pixels, although 800×600-pixel displays are becoming more commonplace. When the memory depth exceeds the display’s horizontal resolution, the scope must manipulate the data to avoid losing information. Suppose that a scope displays waveforms in a 500-pixel-wide window and uses the other 140 pixels across the screen for menus and, perhaps, a scroll bar. Now suppose that the scope displays a full 1M-sample record in the waveform window. Each column of pixels must present the data from 106/500=2000 samples. The most simplistic approach, decimation, would retain only the first sample in each group of 2000 and discard the other 1999. Aside from the more expensive ADC necessary to rapidly fill the more costly 1M-sample waveform memory, this approach is equivalent to cutting the sampling rate and memory depth by a factor of 2000.

In many cases, the low effective sampling rate would cause aliasing, the generation of nonexistent, low-frequency components in what remains of the original data set. To prevent aliasing, the scope can use a slightly more sophisticated approach: peak-to-peak detection. The scope determines the highest and lowest values among the 2000 samples that the 1-pixel-wide column represents and illuminates all pixels between the two values. Although this approach does indeed prevent aliases from entering data that did not originally contain them, it leads to displays that tend to conceal, rather than reveal, the signal’s real behavior. A bright band appears on the screen. Within that band, a much narrower band represents the values the signal usually assumes. To provide a better indication of the signal’s behavior, some scopes find the numerical average of the samples that each column represents. The scope then intensifies the pixel in each column that represents this value so that the “average” waveform stands out from the other pixels in the bright band.

Several manufacturers use an even more sophisticated approach, which demonstrates how much technology instrument designers sometimes must employ to synthesize digital systems that behave like rather straightforward analog systems. Most engineers agree that, because of their inherent ability to display a gray scale, analog scopes pack more information onto the screen than do conventional DSOs. Scopes from Gould and Yokogawa, among others, make their displays more meaningful by determining the number of times the measurements represented by each pixel column correspond to the ADC readings assigned to each pixel. The scope then grades the intensity of each illuminated pixel to represent the frequency of occurrence of the voltage value the pixel represents. Although there are usually 64 or fewer intensity levels, the result faithfully mimics an analog-scope display. In DSOs, grading pixel intensity in this manner or using similar algorithms for color grading requires a lot of data processing, which, if you implement it with inappropriate hardware, can unacceptably slow the scope’s operation. Gould scopes assign this function to a dedicated DSP. Some other manufacturers use custom high-speed ASICs.


RECORD-SETTING RECORD LENGTHs
A fact of scope life in the technological fast lane is the dramatic increase in record lengths. Scope manufacturers now generally consider 1G samples/s to be a moderate sampling rate. The fastest real-time-sampling scopes now take 20G samples/s. A rate of 10G samples/s is the minimum for real-time capture of the 5-GHz signals used in wireless networks that conform to the IEEE 802.11a standard. At 10G samples/s, a 4-ms record, which is less than one-fourth the duration of one cycle at the 60-Hz US power-line frequency, contains 40M samples. At 10G samples/s, only one standard scope, LeCroy’s new WaveMaster 8500, currently captures records that long (Figure 2). In its two-channel mode at 10G samples/s, the 8500 can capture 4.8-ms records.

Figure 2 The clarity of LeCRoy’s WaveMaster 8500’s 10.4-in. (diagonal) 800×600-pixel LCD is barely the tip of the iceberg among the scope’s trend-setting capabilities. Even the 48M-sample/channel memory, 20G-sample/s acquisition rate, and 5-GHz bandwidth take a back seat to the instrument’s cache-based X-Stream architecture, which usually enables the complex calculations required for waveform analysis to begin before the scope finishes acquiring the waveforms.

Although Agilent Technologies, in its former incarnation as part of Hewlett-Packard, pioneered low- to mid-performance deep-memory DSOs, the company has until recently been noticeably absent from the market for what are now called midrange and high-performance deep-memory scopes. (Agilent’s 200-MHz-bandwidth 54600 series has offered 1M-sample/channel memories for years.) At least in the new midrange—that is, in units with 600-MHz to 1-GHz bandwidth—Agilent finally joined the fray in January with its 54830 series (Figure 3). The new series uses an updated version of the 54600 series’ MegaZoom architecture and makes notable additions to the company’s Infiniium family (see sidebar “Understanding oscilloscope display-update rates”on the Web version of this article at www.ednasia.com). The original Infiniium scopes were the industry’s first Windows-based DSOs. As such, they set a trend which Tektronix, Gould, and LeCroy have followed. Many users still consider Agilent’s rendition of the Windows user interface the most faithful and hence the most intuitive among DSOs. Moreover, touch-sensitive screens on newer Infiniium models answer the objections of most of those who refuse to use their scopes with an external pointing device, such as a mouse, a trackball, or a touchpad. The touch-sensitive screen does not, however, satisfy those who will accept neither touching the screen nor using a pointing device. For those people, Agilent’s competitors let you access all of the scope functions via front-panel controls. Agilent’s front-panel controls make available most of the scopes’ functions, including all of the commonly used ones, but a few functions require using the touch-sensitive screen or a pointing device. However, Agilent, in a move that Tektronix and Gould emulate, also offers the option of voice control, which can prove almost invaluable when you have a probe in each hand.

Figure 3 Among the most recent entries among deep-memory DSOs in the midrange- and high-performance arenas are the newest members of Agilent’s Infiniium family. The 54830 series includes two- and four-channel units with bandwidths of 600 MHz and 1 GHz, memory depths as high as 16M samples, and real-time sampling as high as 4G samples/s.

The 54830 series, whose prices begin at US$12,995, includes two- and four-channel, 600-MHz-bandwidth units and a four-channel, 1-GHz-bandwidth unit. Maximum memory depth is 8M samples when you use the channels separately and 16M samples when you interleave pairs of channels. Maximum sampling rate is 2G samples/s noninterleaved and 4G samples/s interleaved.


MORE RESPONSIVE CONTROL
A striking feature of the 54830 series is the responsiveness of the controls even when you use the maximum memory depth. Both this characteristic and the maximum waveform-capture rate of 7800 waveforms/s (approximately 60 times that of the nearest equivalent DPO) result from the use of a custom ASIC for waveform processing.

Unfortunately, however, 54830-series units lack some features that many high-performance-scope users have come to expect. Agilent says that it will correct these omissions and that purchasers will be able to upgrade their units without returning them to the factory.

The most obvious feature missing from these latest Windows-based scopes is the ability to use Windows. You’ll be disappointed—at least for now—if you thought that Agilent’s introduction about a year ago of a logic analyzer with an open Windows architecture indicated that the company’s Windows-based scopes would offer similar capabilities. As with earlier Infiniium models, the deep-memory units don’t currently let you run software of your choice, even though the ability to do so could enhance the scopes’ rather Spartan analysis capabilities.

The ability to run applications such as the Mathworks (www.mathworks.com) MatLab could go a long way toward making the Infiniium waveform-analysis functions competitive with those of LeCroy and Tektronix. Indeed, Agilent offers MatLab capabilities in a USB-analysis option for Infiniium scopes. But the optional package includes a dedicated, embedded version of MatLab; you can’t install the full package on the scopes.


OPEN WINDOWS;LET IN GLITCHES?
The principal argument in favor of not providing an open Windows environment in a Windows-based instrument is that user-installed applications, even when not in use, can degrade the performance of the primary instrument application. An example of this phenomenon would be a user-installed application that requires a DLL (dynamic-link library) whose revision level differs from that of a similarly named DLL that the primary application requires. Such conflicts shouldn’t occur because new DLLs are supposedly backward-compatible with older applications. Moreover, newer Windows versions include provisions for avoiding such problems. Still, incompatibilities can occur.

Also, using a “sneakernet” to transport data files that are too large for a standard floppy disk from your new Infiniium scope to a general-purpose PC may take more work than you’d expect. The new models’ built-in large-file-transfer capabilities are currently limited to the network connection. Although recent Infiniium units incorporated 120-Mbyte LS-120 drives, the drives that Agilent was using will soon be unavailable, and the company isn’t offering them on the 54830B series. Without an external drive, such as a Zip or CD-R/W drive, standard floppy disks are the only removable media on which the scopes can write. The scopes include CD-ROM drives but not a CD-R/W drive.

A waveform-analysis function that could nicely complement the ability to acquire long waveform records is the ability to make search masks out of anomalous waveform segments that instrument users spot while examining the records. Users often want to tag such anomalies in the hope of identifying the conditions associated with their appearance. The ability to rapidly search through previously captured data to uncover multiple similar artifacts could quickly provide crucial insights into system-failure mechanisms. Agilent may have omitted such a feature because of the existence of a competitor’s patent on it. Such a patent would effectively preclude the Infiniium scopes from incorporating the feature. Whether or not such a competitive patent exists, logic analyzers that Agilent itself introduced during 2001 incorporate a similar feature.


A WHOLE NEW ARCHITECTURE
Perhaps the most exciting DSO news in recent years was LeCroy’s introduction last week of the WaveMaster 8500 and its X-Stream architecture. Although LeCroy did not originate the concept of building a DSO around a Windows PC, the level of integration between the 8500’s scope functions and PC functions is greater than that of other PC-based scopes. The results are apparent not only from the speed with which the scope executes computation-intensive tasks, but also from its flexibility in implementing custom functions via user-generated Visual Basic and MatLab scripts. The 8500’s speed and flexibility result from hardware and software designs that LeCroy says are brand new from the ground up. The CPU is an Intel (www.intel.com) Pentium IV and, unlike most competitive Windows-based scopes, which use Windows 9x, the operating system is Windows 2000. (Gould’s Windows-based scopes use Windows NT 4.0.)

LeCroy can’t claim to be the first to use SiGe technology in a commercial DSO; Tektronix won those bragging rights last year when it introduced the TDS 7404, whose input amplifiers use SiGe. Tektronix’s TDS 6604 now also makes similar use of SiGe. In addition, the 6604’s 6-GHz bandwidth is 20% higher than the 8500’s 5 GHz. However, LeCroy uses SiGe in more ways than Tektronix currently does; both the 8500’s ADC chips and its clock generator use the technology. LeCroy says that the SiGe clock generator and the use of decimation rather than a PLL to effectively lower the sampling rate at low sweep speeds were the keys to achieving the 8500’s clock jitter of 1 ps rms over a 1-ms interval.

In addition, the 8500, with prices starting at US$54,990, comes standard with four times the memory of the 6604, with prices starting at US$57,500. Moreover, although you can’t increase the 6604’s memory depth, you can increase the 8500’s to 48 times that of the basic unit—or 192 times the 6604’s. The 6604 captures 12.5-µs records at its maximum sampling rate of 10G samples/s on four channels and 20G samples/s on two. The 8500’s maximum sampling rates are the same, and when you configure it with maximum memory, it captures 2.4-ms records. What’s more, with four channels active, the 8500, in the real-time-sampling mode, properly samples signals at its -3-dB bandwidth. The 6604 has 20% greater bandwidth, causing it to undersample and introduce aliasing. To overcome this problem, you must either use random-repetitive sampling or use only two channels.

Notwithstanding the WaveMaster’s impressive combination of bandwidth, sampling rate, memory depth, and clock stability, the new scope’s pièce de résistance is its X-Stream architecture and the benefits it brings to those who need rapid insights into the behavior of complex systems under test. The easiest way to grasp this power is probably to witness the scope simultaneously updating and displaying waveforms and histograms, not just of its four inputs, but also of four more channels derived from calculations on the inputs. Because the Pentium IV rapidly performs operations previously considered the province of DSPs, the WaveMaster not only performs the necessary math without visibly slowing down, but also presents the histograms as small, continuously updated “histicons.” Clicking on a histicon brings up a larger histogram display that includes a probability-density value for each column.

LeCroy says that the X-Stream architecture makes possible this unprecedented speed by packetizing the data in the waveform memory; transferring packets to processor cache even before waveform acquisition is complete; and, whenever possible, having the processor operate on the cached data, which is then immediately used to generate the display. In certain situations, however, calculations must await completion of the waveform acquisition. One such case is FFT calculations, the mathematics of which requires full data sets. LeCroy also says that X-Stream makes the waveforms-per-second benchmark for DSO-display responsiveness somewhat irrelevant. Nevertheless, LeCroy claims that the WaveMaster, in segmented-memory mode, betters by 50% the 400,000 waveforms/s of Tektronix’s fastest DPO displays.


Understanding oscilloscope display-update rates
—Jae-yong Chang, Product Manager, Agilent Technologies

“Display-update rate” describes the number of waveforms a scope can display per unit of time; that is, waveforms/s. A fast display-update rate is essential for capturing infrequent events or complex dynamic signals, such as modulated or video signals. In addition, a scope that has a fast display-update rate responds more rapidly when you change front-panel control settings, such as sweep time/division or sensitivity.

One of the factors that inhibited the early adoption of digital scopes was analog scopes’ superior display-update rate. Typically, a single CPU inside the digital scope had to process all of the waveform samples and respond to front-panel-control changes. This architecture created a bottleneck. The advent of very-deep-memory digital scopes aggravated the problem. With advances in digital-scope-display technologies, today’s high-performance models provide a much closer approximation of analog-scope displays.

DPOs (digital-phosphor oscilloscopes) combine a fast display-update rate and an intensity- or color-grading capability. A DPO display closely resembles an analog-scope display. DPOs usually provide normal and FastAcq acquisition modes. In normal mode, the DPO displays the waveform with several levels of gray scaling or color grading at an update rate of approximately 130 waveforms/s in midrange models. However, to achieve greater capture rates, you must operate the scope in the FastAcq mode, which creates a 3-D pixel map in which the z-axis information, which appears on the display as color or intensity grading, indicates the frequency of occurrence of each pixel on the screen. Most of the popular models in FastAcq-mode capture as many as 100,000 waveforms/s, and the top-of-the-line units capture more than that. Thus, the DPOs spend more time actively acquiring data than processing it for display. Therefore, the scope is far more likely to capture occasional transients. With the scope in FastAcq mode with color grading on, you see a nice color-graded picture of the waveform that may show hidden anomalies or glitches. However, use of the FastAcq mode requires some trade-offs. For example, use of this mode limits memory depth and sample rate and doesn’t let you use some general-purpose features, such as zoom and waveform math.

In conventional DSOs, the display-update rate typically suffers as memory depth increases. When there are more samples, the scope has to spend more time processing the data points, and the extra processing time adversely affects the display-update rate. The update rate is thus inversely proportional to the amount of data the scope acquires. However, MegaZoom deep-memory technology, which Agilent originally adopted in its 54600-series scopes, provides instant display response to front-panel control changes, even when you use the deepest memory. MegaZoom also optimizes the sample rate for maximum waveform resolution. With the MegaZoom memory architecture, a custom ASIC includes a data-processing engine that quickly reads data from the acquisition memory and processes it for display and measurement. Processing the data in the ASIC, rather than using software running on the scope’s main CPU, greatly reduces the amount of data transferred to the CPU and substantially increases the display-update rate and front-panel responsiveness. Without any special mode or limitation of general-purpose features, MegaZoom achieves fast update rates even with the deepest memory turned on.

With capture rates as high as 7800 waveforms/s, MegaZoom scopes capture many fewer waveforms per second than do DPOs. However, unlike the DPO architecture, the MegaZoom architecture targets use in combination with deep-memory acquisition, bringing the benefits of fast display-update rate and front-panel responsiveness to today’s high-performance, deep-memory scopes.



You can contact Senior Technical Editor Dan Strassberg at
(1) 1-617-558-4205, Fax (1) 1-617-558-4470
E-mail ednstrassberg@cahners.com