Waveform Ghost Busters Capturing and Analyzing Random and Infrequent Signal Anomalies

Transcription

1 Waveform Ghost Busters Capturing and Analyzing Random and Infrequent Signal Anomalies Engineers often refer to a flickering or dim waveform on their oscilloscope s display as a waveform ghost. A waveform ghost is basically an intermittent signal. Sometimes it s there sometimes it s not. Intermittent signals can be spooky. When your scope is setup to capture a repetitive input signal, perhaps it occasionally captures a wave shape that is different from the normal repetitive stream of waveforms. It could be an infrequent narrow glitch, or an infrequent shift in timing, or an infrequent runt pulse, or anything that is different than what is expected. Reliably capturing random and infrequent waveform anomalies such as these are the most difficult kinds of signal problems to capture, identify, and fix. 1

2 Oscilloscope Tools That Help Reveal and Bust Ghostly Waveforms Agenda: This presentation will begin by showing several examples of ghostly waveforms. We will then discuss the various characteristics of oscilloscopes that can enhance a scope s ability to capture and reveal waveform ghosts including bandwidth, sample rate, memory depth, and waveform update rate. Next, we will discuss the various features of a scope that can used to isolate and analyze ghostly waveforms including advanced parametric triggering, InfiniiScan zone triggering, segmented memory acquisition, and waveform intensity modulation. Let s now begin by taking a look at examples of ghostly waveforms. 2

3 Waveform Ghost Example #1: Infrequent Non-monotonic Edge Waveform ghosts will sometimes show up on your scopes display as either dim or flickering waveforms. This is an example of an infrequent non-monotonic edge, which is both dim and flickering. The waveform usually exhibits a nice clean rising edge. But occasionally the rising edge of the waveform begins to either reverse direction, or momentarily stalls. 3

4 Waveform Ghost Example #2: Infrequent Metastable State This is a an example of an infrequent metastable state. It is often caused by a marginal setup and/or hold time violation. The state of the signal begins to switch logic levels in this case from high to low but then returns to the original state due to an insufficient data-to-clock setup time. 4

5 Waveform Ghost Example #4: Very Infrequent Glitch This is an example of a very infrequent glitch (or narrow pulse). This particular glitch occurs on average just once every one million occurrences of normal rising edges of this clock signal. Capturing a signal such as this on an oscilloscope is often like looking for a needle in a haystack. 5

6 Waveform Ghost Example #5: Infrequent Runt Pulse A runt pulse is a digital pulse with insufficient high or low amplitude. In this example we know that these runt pulse are infrequent due to the relative dimness of the trace. 6

7 Waveform Ghost Example #6: Infrequent Setup Time Violation This is an example of an infrequent data-to-clock setup time violation. The yellow waveform is a synchronous clocking signal. The green waveform is a logic data signal. For logic high or logic low data (1 s and 0 s) to be reliably clocked into a device such as memory, the data signal must be stable (high or low) for a minimum amount of time before the occurrence of the rising edge of the clock. However, in this example we can see rising and falling data edges (dimmer traces) sometimes shifted to the right, which may be a violation of the device s setup time specification. 7

8 Oscilloscope characteristics/specs that enhance its ability to capture anomalies All of the examples we just showed are infrequent waveform anomalies. In other words, they are all problem type signals. So what characteristics of a scope will enhance its ability to capture and show you these various types of anomalies? The biggest misconception among oscilloscope users is that higher sample rates will improve a scope s probability of capturing these types of problem signals. If a scope s front-end hardware has sufficient bandwidth, then higher sample rates will produce higher real-time bandwidth. Real-time bandwidth will determine how fast of an edge and how narrow of a pulse can be captured. If the scope s bandwidth is too low, then fast and narrow pulses will be filtered out before they even get to the scope s A-to-D converter. But sampling faster will NOT improve a scope s probability of capturing something that is random or infrequent. You can think of a scope s real-time bandwidth - and its associated sample rate - as a prerequisite for capturing signals with high frequency content. We will go into more depth on required sample rate and bandwidth next. The second misconception (or myth), is that deeper memory will improve a scope s probability of capturing waveform ghosts (infrequent anomalies). Although it is true that deeper memory may improve the scope s ability to capturing something that happens infrequently, since with deeper memory the scope capture a longer time-span at a high sample rate, deep memory will also hide those troublesome waveform ghosts.

9 Let s now talk about minimum required bandwidth.. 8

10 Oscilloscope Bandwidth Definition The most important performance characteristic to evaluate when selecting a new scope is typically bandwidth. As mentioned earlier, a scope s bandwidth will determine how fast of an edge, as well as how narrow of a pulse can be captured. The bandwidth of an oscilloscope is defined as the frequency at which an input sine wave is attenuated by 3 db. Scopes have a variety of frequency responses, but the most typical responses are a Gaussian type response (single pole roll-off), represented by the red trace, and a maximally-flat response - sometimes called a brick-wall response - represented by the green trace in this graph. Most of today s lower bandwidth scopes ( < 1 GHz ) exhibit a Gaussian type response. Many higher bandwidth scopes exhibit a maximally-flat response. Also shown in this slide is an actual swept frequency response measurement of a 1-GHz bandwidth scope, which has an approximate Gaussian response. There are advantages and disadvantages of each type of response. But just remember, if the highest significant frequency content of your input signals is significantly lower the bandwidth specification of the scope, it really doesn t matter. Where is does matter is in determining the minimum required sample rate to produce a particular real-time bandwidth. This will be discussed later.

11 Selecting the Right Bandwidth This slide shows two different bandwidth oscilloscopes capturing the same 100 MHz digital clock signal. The screen-shot on the left shows what a 100 MHz digital clock looks like when captured by a 100-MHz bandwidth oscilloscope. The higher harmonics of this signal have been attenuated to such a degree that all that virtually remains is the fundamental frequency component of this clock signal (100 MHz sine wave). The screen-shot on the right shows what the same 100 MHz clock signal looks like when captured by a 500-MHz bandwidth scope. The 500-MHz bandwidth scope is able to not only capture the 100 MHz fundamental frequency component, but also higher harmonics with reasonable accuracy. Since input sine waves are attenuated by approximately 30% (-3 db) at the bandwidth frequency of the scope, you should never use a scope of a specific bandwidth to test signals of the same frequency. For pure analog/rf measurement applications (sine waves), it is recommended that the scope s bandwidth be at least three times higher than the highest input sine wave frequency that you may want to measure. At 1/3 the scope s bandwidth, signals are typically minimally attenuated. For digital applications, which are actually more common today, your scope s bandwidth should be at least fives times higher than the highest clock rate in your system. If your scope s bandwidth is at least five times higher than the highest clock rate, then it will be able to capture up to the fifth harmonic with minimal attenuation. Note that the 5X factor for digital applications is actually just a rule-of-thumb recommendation. There is actually a more accurate method to determine the

12 appropriate bandwidth based on actual frequency content in high speed edges regardless of the clock rate. 10

13 Required Bandwidth Based on Edge Speed Using a more accurate method to determine required bandwidth, let s now determine the appropriate scope bandwidth to capture a fast rising edge for accurate parametric measurements. The first step is to determine what the fastest edge speeds are in your design. You can typically get this information from data sheet specifications of the fastest devices in your system. The second step is to calculate the maximum practical frequency component within the signal under test based on the fastest rise/fall times of your signals. We refer to this frequency component as f knee. Dr. Howard W. Johnson has written a book that includes this topic titled, High-speed Digital Design A Handbook of Black Magic. All fast rising edges have an infinite spectrum of frequency components. However, there is an inflection (or knee ) in the frequency spectrum of fast edges where frequency components higher than f knee are insignificant in determining the shape of the signal. For high-speed signals with rise time characteristics based on 10% to 90% thresholds, f knee is equal to 0.5 divided by the rise time of the signal. The next step is to calculate the required bandwidth of the oscilloscope to measure this signal based on what degree of accuracy you require. The table in this slide shows multiplying factors for various degrees of accuracy for scopes with a Gaussian or a maximally-flat frequency response. Let s now walk through a simple example.

14 Required Bandwidth Example If your fastest digital signals have an approximate rise/fall time of 500 ps (based on a 10% to 90% criteria), then the maximum practical frequency component (f knee ) in the signal would be approximately 1 GHz. And this frequency has no relationship to a clock rate. The clock or repetitive rate could be just 1 MHz, for example. But if the signal has a rise time of 500 ps, then the signal is composed of significant frequency components up to 1 GHz. For a rise time measurement accuracy of 3%, the minimum scope bandwidth would be approximately 2 GHz (assuming a Gaussian response scope). However, if you are able tolerate up to 20% timing errors when making parametric rise time and fall time measurements on your signals, then you could possibly use a 1 GHz bandwidth oscilloscope for your digital measurement applications. For example, when measuring a signal with an actual rise time of 500 ps, a 1 GHz scope may show it as a 600 ps rise time. So if your budget is limited and you can tolerate this degree of error, then a 1 GHz scope may be the right choice for you. But if you wanted to measure the timing of this edge more accurately, then a 2 GHz scope would be the right choice if it fits your budget. Agilent s Recommendation Select a scope that has sufficient bandwidth to accurately capture the highest frequency content of your signals.

15 Sample Rate So how much sample rate is required to achieve a particular real-time bandwidth specification? Remember, real-time bandwidth simply means the bandwidth that can be achieved for each and every single-shot acquisition of the scope. Closely related to real-time bandwidth requirements are real-time sample rates. Some engineers have total trust in Nyquist and claim that just 2X sampling over the scope s bandwidth is sufficient. Other engineers don t trust digital filtering techniques based on Nyquist criteria and want their scope to sample at rates that are 10X to 20X over the scope s bandwidth specification. The truth actually lies somewhere in between. To understand why, you must have a understanding of the Nyquist theorem and how it relates to a scope s frequency response. So, let s find out exactly what Mr. Nyquist says

16 Nyquist s Sampling Theorem Nyquist s sampling theorem states that for a limited bandwidth (band-limited) signal with maximum frequency f max, the equally-spaced sampling frequency f s must be greater than twice of the maximum frequency f max, i.e., f s > 2 f max in order to have the signal be uniquely reconstructed without aliasing. The frequency 2 f max is called the Nyquist sampling frequency (f S ), which we usually call sample rate. Half of this value, f max, is sometimes called the Nyquist frequency (f N ).

17 Ideal Brick-wall Response with Nyquist (f N ) If an oscilloscope s bandwidth is specified exactly at the Nyquist frequency (f N ), this implies that the oscilloscope has an ideal brick-wall filter that falls off exactly at this same frequency. This type of frequency response filter is impossible to implement in either hardware or software.

18 Gaussian Response w/ f f S /4 (f N /2) Since scopes don t have a brickwall frequency response, significant frequency components above the Nyquist frequency (f N ) can be sampled if the scope s bandwidth is the same as the Nyquist frequency. In other words, if the scope s sample rate is just 2X the scope s bandwidth, displayed waveforms can have aliased results, as we will show next.

19 1-GHz scope 2 GSa/s (BW = f S /2 = f N ) This is an example of a 1-GHz bandwidth scope sampling at 2 GSa/s. The input test signal is a 1 ns wide pulse with very fast edge speeds. Both the rising and falling edges of this narrow pulse contain significant frequency components beyond the Nyquist frequency (f N ). The digitized waveform result is aliased. The aliasing appears as wobbling edges with varying degrees of pre-shoot and overshoot. So, where should a scope s bandwidth (f BW ) be specified relative to the scope s sample rate (f S ) and the Nyquist frequency (f N )?

20 Gaussian Response w/ f f S /4 (f N /2) To minimize sampling significant frequency components above the Nyquist frequency (f N ), most scope vendors specify the bandwidth of their scopes that have a typical Gaussian frequency response at 1/4 th to 1/5 th, or lower, than the scope s real-time sample rate. Although sampling at even higher rates relative to the scope s bandwidth, or specifying the scope s bandwidth at a lower frequency relative to the scope s sample rate would further minimize the possibility of sampling frequency component beyond the Nyquist frequency (f N ), specifying a scope s bandwidth at 1/4 th to 1/5 th is sufficient as we will show in just a minute.

21 Maximally-flat Response w/ f f S /3 (f N /1.5) As mentioned earlier, many higher bandwidth oscilloscope exhibit a sharper frequency roll-off characteristic, which we call a maximally-flat frequency response. Since a scope with a maximally-flat response approaches the ideal characteristics of a brick-wall filter, where frequency components beyond the Nyquist frequency are attenuated to a higher degree, not as many samples are required in order to produce a good representation of the input signal using digital filtering. Scope s with this type of response can theoretically specify their bandwidth (assuming the front-end analog hardware is capable) at f S/ 2.5. Let s now compare signal fidelity when capturing the same narrow pulse we showed earlier using various sample rates.

22 1-GHz Bandwidth Oscilloscope Here we show a 1-GHz scope that has a maximally-flat frequency response capturing the same 1-ns wide pulse with 500 ps edge speeds. This is about the narrowest and fastest edge speed that you can expect to capture using a 1-GHz bandwidth scope. Remember when we previously walked through an example of determining the minimum required bandwidth to capture 500 ps edges, we determined that 1 GHz bandwidth would be a minimum cost-effective solution, but that 2 GHz bandwidth would provide more accuracy. The screen image on the left shows the scope sampling at 2.5 GSa/s for a sample rate-to-bandwidth ratio of 2.5-to-1. This is the minimum recommendation that we just discuss. The screen image on the right shows the same scope sampling at 5 GSa/s for a sample rate-to-bandwidth ratio of 5-to-1. Note that although there is some improvement in this measurement at the higher sample rate, the improvement is minimal. Many engineers would expect that a scope that samples at twice the rate would produce significantly better results. If this narrow pulse were to be occurring randomly and infrequent, both scopes could capture it with the same probability. Higher sample rates do NOT improve the probability of capturing random and infrequent events. Higher sample rates only improves the accuracy of each acquisition but just minimally. 20

23 2-GHz Bandwidth Oscilloscope Here we show the same signal, but now captured on a 2-GHz bandwidth scope sampling at 10 GSa/s and 20 GSa/s. The higher bandwidth definitely provides us with improved measurement accuracy over the 1-GHz bandwidth scope. Our rise time measurement now have less than 3% error. Notice that the difference between sampling at 10 GSa/s (SR = BW x 5) versus 20 GSa/s (SR = BW x 10) is virtually imperceptible. But there can actually be some tradeoffs with higher bandwidth and higher sample rates. Higher sample rates on some scopes can sometimes produce degraded measurement accuracy and resolution due to interleaved sampling noise, which is not within the scope of this presentation. But I ve listed an application note at the end of this presentation which goes into detail on this topic. Secondly, higher bandwidth and higher sample rate means a higher sticker price. Lastly, the higher bandwidth and higher sample rate provided by this particular scope does NOT improve the probability of capturing this narrow pulse if it were to be occurring randomly and infrequently. This scope actually has lower probability of capturing this pulse (if it were random) as compared to the 1-GHz bandwidth scope we showed in the previous slide. 21

24 Raw Sample Data vs Sin(x)/x Reconstruction Many engineers would intuitively believe that capturing a 1-ns wide pulse with 500 ps edge speeds while sampling at just 2.5 GSa/s would provide very poor display and measurement resolution since raw digitized samples would be spaced 400 ps apart. But nearly all of today s modern digital scopes automatically run a Sin(x)/x waveform reconstruction filtering after each repetitive acquisition to provide for improved and higher data resolution. The screen image on the left shows a single-shot capture of our 1-ns wide pulse without Sin(x)/x waveform reconstruction. The screen image on the right shows after Sin(x)/x waveform reconstruction has been performed with data resolution improved from 400 ps down to 7.5 ps. In addition to higher data resolution, the Sin(x)/x waveform reconstruction filtering also provides us with a continuous waveform trace like you expect to see on a scope s display If Nyquist s rules are observed, Sin(x)/x waveform reconstruction always provides improved measurement accuracy and resolution, as well as significantly improved display quality (a continuous waveform). Unfortunately on many of today s scopes, Sin(x)/x waveform reconstruction degrades waveform update rate significantly, which is extremely important for capturing random and infrequent events as we will discuss in a few minutes. But on Agilent s InfiniiVision X-Series oscilloscope, Sin(x)/x reconstruction is performed within the MegaZoom IV ASIC (not software), which means that fast waveform update rates can be sustained (no hit in update rate). 22

25 Oscilloscope Memory Depth Now that we ve covered minimum required bandwidth and sample rate, let s now delve into the 2 nd myth, which is that deeper memory will improve a scope s probability of capturing random and infrequent events. A scope s memory depth will determine how long of a time-span of signals a scope can capture while sampling at its highest sample rate. Acquisition time is equal to the scope s memory depth divided by the scope s sample rate. And intuitively, if the scope can capture a longer time-span, then there is a higher probability that a random and/or infrequent event may be captured withing the scope s memory buffer. And this is true. In the example shown in this slide the scope actually captured 2 metastable states. But where are they? They are hidden by the deep memory acquisition and display. 23

26 Searching through Deep Memory for Ghosts After capturing a deep memory record, you can search through the scope s memory. Many scopes on the market today have automatic search & navigation capabilities, including Agilent s InfiniiVision 3000 and 4000 X-Series oscilloscopes. But you have to know what you are searching for in order to set up the automatic search criteria. Are you looking for glitches? Or are you looking for non-monotonic edges? And how do know what to search for if the strange anomalies haven t yet been revealed. Manually scrolling through a deep memory record looking for anomalies can be very time-consuming especially for very deep memory scopes. In this example, we set up the scope to search for glitches using a pulse-width search criteria based on a particular polarity pulse (negative pulse in this case) that was less than 30 ns wide. In this case, we found 2 metastable states. But we are in a Catch 22 situation here. The deep memory catches 2 anomalies, but doesn t readily reveal their characteristics. But to search for them, I ve got to know something about them. 24

27 Oscilloscope dead-time hides waveform ghosts Now we are finally getting around the characteristic of a scope that helps to reveal infrequent and random anomalies. All scopes have an inherent and negative characteristic called dead-time or blind-time. It s the time between each repetitive acquisition that scopes require to process and display the data that they just captured. In this example, we graphically show two acquisitions of the scope with oscilloscope dead-time in between each acquisition. But this graphic is exaggerated. Dead-time is often orders of magnitude longer than acquisition time. In this example, two signal anomalies occurred during the dead-time and weren t captured and displayed. If signal anomalies are random and infrequent, they are more likely occur during the scope s dead-time as opposed to occurring during the scope s acquisition time. With additional repetitive acquisitions, the scope will eventually capture and display the anomalies. 25

28 Faster Waveform Update Rates Reveal Ghostly Scopes with faster waveform update rates have lower dead-times. This improves the probability that glitches and signal anomalies will occur during the scope s acquisition time as opposed to occurring during the scope s dead-time. Agilent s InfiniiVision 3000 and 4000 X-Series oscilloscope have the fastest waveform update rates and lowest dead-times in the oscilloscope industry. These scopes can update waveforms as fast as 1,000,000 waveforms per second. These rates have been achieved through Agilent s exclusive MegaZoom IV ASIC technology. 26

29 Capturing an Infrequent Non-monotonic Edge The screen image on the left shows the scope updating at 1,000 waveform per second. This means that the scope s dead-time is approximately 1 millisecond. (Note that we were able to artificially slow down this scope s update rate by entering a trigger hold-off value of 1 ms.) Although the repetitive captures of the waveform certainly look lively and responsive at this update rate, the scope is unable to capture and display a very infrequent non-monotonic edge. The screen image on the right shows the scope updating at approximately 300,000 waveforms per second. Now we can see the infrequent non-monotonic edge. Note that the scope used in this measurement is capable of updating waveforms up to 1,000,000 waveforms per second. Capturing random and infrequent events is all about statistical probabilities. It depends on the trigger rate of the signal, the update rate of the scope, the scope s timebase setting, and the rate of occurrence of anomalies. 27

30 Isolate Waveform Ghosts with Advanced Violation Triggering Once a waveform anomaly has been reveal with fast waveform update rates, its then time to isolate it, and then attempt to find the root cause. Isolating it typically means setting up the scope to trigger exclusively on the anomaly. Many of today s newer digital oscilloscopes have advanced parametric triggering capabilities such as rise/fall time trigger, pulse-width trigger, runt trigger, and setup & hold time violation triggering. In the case of the infrequent non-monotonic edge, we could use the scope s rise time trigger capability. But setting up advanced trigger conditions such as this are easier-said-than-done. Most engineers simply use edge triggering. To help out with the complexity of setting up an advanced trigger condition, Agilent provides built-in help screens with detailed instructions and graphics on how to set it up as shown by the screen image on the right. 28

31 Rise Time Trigger: Isolates a non-monotonic Edge Ghost This slide shows an example of triggering on the non-monotonic edge based on triggering a rising edge that is slower than 165 ns. But there is actually an easier way to isolate a waveform ghost such as this 29

32 Zone Trigger Example #1: Infrequent Non-monotonic Edge Agilent s InfiniiVision 4000 X-Series oscilloscope have a unique hardware-based feature called InfiniiScan Zone Trigger. To trigger on just waveforms that have the non-monotonic edge, simply draw a small box (zone) using the scope s capacitive touch-screen in the region of the anomaly as shown in this slide, and then the scope locks-in on the waveform anomaly. Basically, if the scope s fast waveform update rate reveals a ghostly waveform while using standard edge trigger, then InfiniiScan Zone Trigger can easily catch it. 30

33 Zone Trigger Example #2: Infrequent Metastable State Here we show the infrequent metastable state waveform we showed earlier. Again, just draw a small box (zone) in the region of the anomaly and then the scope uniquely captures just the anomalous waveforms. 31

34 Zone Trigger Example #3: Very Infrequent Glitch This is the one-in-a-million glitch waveform. Although we could use the scope s pulse-width trigger mode to trigger on this very infrequent anomaly, Zone Trigger is simply much easier. See it draw a box around it trigger on it! 32

35 Zone Trigger Example #4: Narrow Runt Pulse Here we show a 2-zone trigger qualification in order to trigger on the narrower of two infrequent positive runt pulses. Zone #1 is a must intersect zone, while zone #2 is a must not intersect zone. 33

36 Zone Trigger Example #5: Wide Runt Pulse By simply sliding Zone #2, which is a must not intersect zone, down to the region of the falling edge of narrower runt pulse, the scope now triggers exclusively on the wider runt pulse. Alternatively, we could have left Zone #2 where is was, and then simply redefined it to be a must intersect zone. 34

37 Zone Trigger Example #6: Setup Time Violation Lastly, here we show the infrequent setup time violation signal. Drawing a must intersect zone box in the region of the dim waveforms locks the scope in on just the setup violation data signals. 35

38 Memory Depth Re-visited: Traditional Deep Memory Acquisition Let s now talk about some additional oscilloscope tools that you may have in your arsenal that might help you characterize and debug ghostly waveforms. Previously we talked about how deep memory may actually hide waveform anomalies. Let s now show how you can more effectively use deep memory acquisitions to analyze infrequent anomalies. As we discussed earlier, the maximum time-span of signals that a scope can capture while still sampling at its maximum sample rate is a function of the scope maximum memory depth. Remember, maximum captured time-span is equal to the scope s memory depth divided by the scope s maximum sample rate. In the case of Agilent s InfiniiVision 3000 and 4000 X-Series oscilloscopes, the maximum memory depth is 4 M points, and the maximum sample rate is 5 GSa/s. This means that the maximum continuous time-span that the scope can capture (while still sampling at its maximum rate) is 800 µs. In this example of a signal with two low duty cycle pulses separated by 400 µs, the maximum number of pulses that the scope can capture is just two. You can also think of the widely separated and narrow pulses as two waveform anomalies buried within a stream of normal waveforms. Let s now see how we can expand the number of events and effective timespan - that the scope can capture using the same amount of acquisition memory. 36

39 Segmented Memory Acquisition Segmented Memory acquisition can effectively extend the scope s total acquisition time by dividing the scope s available acquisition memory into smaller memory segments. The scope then selectively digitizes just the important portions of the waveform under test at a high sample rate as illustrated here, but without capturing signal idle-time or unimportant portions of the signal. This enables your scope to capture many successive single-shot waveforms with a very fast re-arm time without missing important signal information. After a segmented memory acquisition is performed, you can then easily view all captured waveforms overlaid in an infinite-persistence display, as well as quickly scroll through each individual waveform segment. And in the case of serial bus applications, the scope also automatically provides protocol decode of each captured packet/segment. Although most of the signal dead/idle-time between each segment is not captured, the scope provides a time-tag for each segment so that you know the precise time between each pulse, each burst, or each serial packet captured. 37

40 Setting up a Segmented Memory Acquisition The screen image on the left shows the scope triggering on any rising edge of the signal. We can also see that the scope is randomly capturing a narrow and very infrequent glitch, which is actually a metastable state. Using the scope s pulse-width trigger mode, the screen image on the right now shows the scope exclusively triggering on just the glitch. 38

41 Step #3: Using Segmented Memory to Capture 1000 Consecutive Ghosts Segmented Memory acquisition goes by many different names depending upon the scope vendor. Names includes Segmented Memory, FastFrame, Sequence mode, and History mode. After setting up the scope to exclusively trigger on the anomaly using pulse-width triggering, we can then select he Segmented Memory acquisition mode in Agilent s InfiniiVision 3000 and 4000 X-Series oscilloscopes to capture up to 1000 consecutive occurrences of the anomaly, which in this case is a very infrequent glitch, which is actually a metastable state. After capturing each selective event, we can then view each captured event individually, or all overlaid on top of one another in an infinite-persistence display as shown in this slide, along with precise time-stamps of each event relative to the 1 st captured event. For this particular example, the last captured event occurred 289 seconds after the 1 st event. Capturing a time-span of 289 seconds using conventional and continuous deep memory acquisition would have required 1.5 Terra Bytes of memory (Memory = time-span x sample rate = 298 seconds x 5 GSa/s = 1.5 TB.) The deepest memory available today in any oscilloscope is 1 GB, which is only offered by Agilent Technologies in our Infiniium Series oscilloscopes. 39

42 Waveform Intensity Modulation Now for our last topic before we wrap up this presentation. Many of today s newer digital storage oscilloscope have the ability to display waveforms with graduated intensity modulation that simulates the look of older analog-technology oscilloscopes. This is extremely helpful for visually interpreting the relative frequency of occurrence of captured events. Portions of signals that occur most often are displayed brightly while portions of signals that occur rarely are displayed dimly. Agilent s InfiniiVision X-Series oscilloscopes are capable of displaying up to 64 discreet levels of trace intensity. This effectively provides you with a 3-D view of waveforms. This ability to display multiple levels of intensity is also greatly enhanced by the scope extremely fast waveform update rate of up to 1,000,000 waveforms per second and automatic MegaZoom deep memory acquisitions. The more data the scope can capture in a given amount time, the better the scope can build up a valid statistical accumulation of waveform data based on frequency-of-occurrence. 40

43 Summary To summarize what we ve learned today, a scope s real-time bandwidth is a prerequisite for capturing the highest frequency content of not only random and infrequent events, but also the highest frequency content of the valid signals in your designs that you need to test. Select a scope that meets your minimum bandwidth requirements. Sampling faster than the minimum required sample rate based on Nyquist criteria does not improve a scopes ability to capture random and infrequent events. Also, although there are many applications today that may require deep memory acquisitions, using deep memory to discover if random and infrequent events are occurring is not one of them. In fact, as we showed earlier, deep memory acquisitions can actually hide signal anomalies. The key to improving a scope s probability of capturing and displaying infrequent and random events is minimized dead-time and faster waveform update rates. If a scope s waveform update rate is fast enough to show waveform ghosts, then there are additional tools available in many of today s new DSOs that can help you debug your designs including: Advance Parametric/Violation Triggering Zone Triggering Segmented Memory Acquisition Waveform Intensity Modulation Lastly, remember, if you can see it (a waveform ghost), then Agilent s InfiniiScan zone trigger can catch it! 41

44 Additional Technical Resources If you are interested in learning more about oscilloscopes and oscilloscope measurements relative to this presentation, you can download these documents at no charge using the URL listed in this slide. Simply insert the publication number in place of xxxx-xxxx. Alternatively, you can go to Agilent s website at and then enter the publication number into Agilent s search engine box.

45 Agilent s InfiniiVision X-Series Oscilloscopes Agilent s InfiniiVision X-Series oscilloscopes come in three different flavors with various performance capabilities. The scope used for all the screen images shown during this presentation was the newest InfiniiVision 4000 X-Series oscilloscope. 43

46 Questions and Answers At this time we would like to open up this presentation for any questions. Thanks for attending today. 44