Choosing and using the right tools for high-speed serial data analysis - Embedded.com

Choosing and using the right tools for high-speed serial data analysis

Proven first generation serial bus technologies are now widelydeployed. Many new consumer products are arriving with PCI Express,Serial ATA, and other serial implementations built in. These busesdeliver higher data rates than their parallel counterparts, while atthe same time simplifying circuit layout.

At the same time, faster serial bus technologies, including secondand third generation serial bus architectures such as HDMI 1.3, SATAIII and PCI-Express 2.0, are now beginning to appear.

Serial bus technology has raised the bar in performancerequirements for high-performance test and measurement instruments usedin design, validation and troubleshooting. High data rates are only thebeginning. Designers need tools that can support critical signalintegrity measurements and eye diagram analysis.

Signal integrity
Signal integrity measurements on serial bus technologies, such as PCIExpress 2.0, Serial ATA III and HDMI 1.3, have become an important partof the designer's measurement work. Many of the issues encountered incompliance testing stem from waveform imperfections such as noise,jitter and timing aberrations.

Serial standards have brought with them narrower timing tolerances,requiring ever more bandwidth and accuracy from the measurement tools.With those demands, an additional requirement has emerged — a method tominimize the impact of the measurement itself.

Figure1. The latest serial analyzers have 20 GHz of bandwidth to handle thefull range of serial buses in use today.

Eye diagram measurements
An oscilloscope's bandwidth affects the all-important eye diagram ofthe signal. This standard oscilloscope image, the cornerstone ofcompliance and validation tests, displays one bit or unit interval ofdata with all possible edge transitions and states superimposed in onecomprehensive view.

The resulting screen shows numerous waveform traces surrounding aroughly hexagonal open area in the center. This is the eye, and itsdegree of openness is a measure of signal quality (the more open thebetter). Serial logic devices must distinguish a clear 1 or 0 statewithin the eye area to respond to the data correctly. A graphical maskis often used to define pass/fail areas of the eye.

An oscilloscope with insufficient bandwidth can roll off as much as1 dB of (vertical) amplitude when it acquires the signal.Unfortunately, this loss tends to fall in the open area of the eyediagram — exactly where the binary decisions are made. Thereforesufficient bandwidth is crucial in eye diagram measurements as well asedge measurements.

Fortunately, instruments are now available which can deliver thebandwidth required for the most critical timing and edge measurementson these standards. The latest serial analyzers have 20 GHz ofbandwidth which, when paired with a 50 GS/s sample rate on allchannels, embraces the whole range of serial buses in use today (Figure1 above ).

At these rates, the input rise time over the 10 percent to 90percent portion of the waveform edge is a mere 22 ps. As discussedlater in this article, DSP-based bandwidth enhancement technologyprovides flat frequency performance up to the fifth harmonic range ofthe fastest first generation serial standards. This same feature alsoensures optimal frequency and amplitude response matching across allchannels.

Figure2. This illustrates how the higher bandwidth 20 GHz scope (a) has moreamplitude margin compared to a 13 GHz scope (b).

.

Fifth harmonic measurements
Most first generation serial bus architectures have data ratesclustered between 2.5 and 3.125 Gbit/s — rates which would seem to bewell within reach of today's 4 and 5 GHz oscilloscopes. However, signalfidelity measurements demand much higher bandwidth, and most standardsbodies have recognized this need by specifying instruments that arefast enough to capture the fifth harmonic of the clock signal.

Capturing the fifth harmonic provides the accuracy required for thecharacterization and analysis of fast rise-time signals, as well asproviding more margin to ensure accurate results.

Table 1 below shows how higher bandwidth provides additionalrise-time accuracy, while Figure 2 above shows how theamplitude margin is enhanced. In particular, it demonstrates thedifference between a 20 GHz oscilloscope and a 13 GHz oscilloscope,showing that the higher bandwidth instrument offers much more margin ona 6.25 Gbit/s data stream PCI-E Gen2 signal.

At 20 GHz, the fifth harmonic is captured and plenty of margin isobvious in the nice clean eye. On the 13 GHz oscilloscope there isinsufficient margin to capture the fifth harmonic, and the signal failsthe mask.

Some standards groups, notably the PCI SIG, are going so far as tocite the specific bandwidth that must be available to capture the fifthharmonic.

Table1. Higher bandwidth provides additional rise-time measurement accuracy.

Multiple lane architectures
Faster serial bus technologies, including second and third generationserial bus architectures such as HDMI 1.3, SATA III and PCI-Express2.0, offer increased performance in single-lane applications but arealso implemented as serial data bus architectures that utilize multiplelanes to achieve even greater data exchange rates.

In a multi-lane configuration, serial data packets are firstdecomposed, and then transmitted over four, eight or more lanes atessentially the same time (Figure 3below ).

Designers doing validation or debug work on a multi-lane serial busneed test solutions to capture a real-time data set across four or morelanes simultaneously with the performance necessary to service thelatest generation of serial bus technologies. For validation, the dataneeds to be captured simultaneously and time-correlated across allsignal lanes.

Figure3. In a multi-lane configuration, serial data packets are firstdecomposed, and then transmitted over four, eight or more lanes atessentially the same time.

Instruments are now available that provide real-time sampling ratesof up to 50 GS/s for superior time resolution across all inputchannels, with up to 4 ms of time-correlated serial data traffic on allfour channels. This combination of real-time capture and deep memorygives a designer the ability to analyze offending events or errorscorrelated with bus traffic that preceded and followed it across everylane.

Jitter testing
Jitter is another concern for serial bus developers. In somecases its effect on individual signal edges must be quantified, butmore importantly jitter plays a role in eye diagram measurements. Itcan decrease the width of the eye where rising and falling edges cross,potentially causing mask violations.

If jitter is observed, is it really coming from the devicebeingtested, or is it actually coming from the measurement instrument? Boththe trigger jitter and the jitter noise floor (JNF) of the oscilloscopecan contribute to the measured jitter, and both can narrow the eye toproduce misleading mask failures.

Elaborate software correction schemes have been developedtominimize the oscilloscope's trigger jitter: an approach thatpotentially improves the performance in equivalent-time acquisitionmodes that must re-trigger for each sample point. However, triggerjitter is not a factor in jitter measurements based on a singlereal-time acquisition. In this situation, it is the JNF figure that canaffect both equivalent-time and real-time capture.

A better approach is to use an instrument with sufficientdynamicrange (10 vertical divisions, for example) in its verticalanalog-to-digital converters to achieve a high signal-to-noise ratioand an exceptionally low JNF (400 femtoseconds RMS typical). Such anoscilloscope can deliver jitter measurements without significant jitterfrom the instrument itself (Figure 4 below ).

Fig.4 An oscilloscope with sufficient dynamic range can deliver jittermeasurements without significant jitter from the instrument itself.

Measuring low frequency jitter is a separate challenge. Itmakes twoconflicting demands on the oscilloscope: asking it to capture tinytiming details and to do so over long spans of time. To capture jitterdetails with sufficient resolution, it is often necessary to enter themaximum sample rate (for example, 50 GS/s, which equates to a sampleinterval of 20 ps).

Sample data accumulates in the waveform memory quickly atthat rate.However, low frequency jitter trends develop in milliseconds. Thus, ifthe instrument is to capture enough operational cycles to determine theimpact of low frequency jitter on the measurements, a very deep memoryis needed.

For an instrument with up to 200 Megasample memory perchannel, upto 4 ms of sample data can be stored at the full sample rate. Thisallows engineers to look at close-up jitter details on individualedges, and in the same waveform record jitter changes that occurmillions of cycles later.

Subtracting measurement effects
Emerging standards in the serial bus environment require thetransmission effects of the measurement channel to be removed from themeasurement results. One way of achieving this is to apply digitalfilters, allowing calculations that previously required a separate stepand separate application software to be performed within theoscilloscope.

Signal filtering is a process that goes back to theearliest days ofanalog electronics, when a filter was a circuit made up of discreteresistors, capacitors and inductors. In today's DSP realm, a filter isa mathematical procedure that modifies the shape — i.e., the frequencycontent — of a waveform, much as its analog predecessors did. But itdoes so by processing signals in digital form, executing functionsranging from simple multiplication to differential equations.

This process consists of entering FIR (finite impulseresponse)filter parameters into the instrument's mathematical system. Thedigital filters remove the calculated loss characteristics from themeasurement, and the oscilloscope displays the waveform trace after the”clean up.” The result is an eye diagram unaffected by the measurementconnections, and an eye that more accurately reflects the operation ofthe device in its end-user application.

With FIR filters included as an integral part of the signalpath, itis possible to implement a broad range of arbitrary filtercharacteristics as well as advanced high-speed serial measurementtechniques, including “virtual test points” that enable analyzinginaccessible signals.

DSP filters can deliver much higher precision, linearityandstability than their analog predecessors. FIR filters are stable andcannot oscillate, and they may exhibit linear phase response, i.e., allfrequencies pass through them with the same amount of time delay,minimizing distortion. Moreover, the impulse response of a FIR filterhas a finite, quantifiable duration; its effect is predictable andcontrollable.

There are many oscilloscope measurement applications forwhichfiltering is useful, and it can be used to address several increasinglycommon high-speed measurement problems. For example, filtering enablesengineers to limit bandwidth to reduce noise while maintaining hightiming resolution. Filtering can also be used to control the manner inwhich the oscilloscope's response rolls off beyond the high end of itsfrequency range.

Figure5. The latest generation of high-performance test instruments usesarbitrary FIR filters with characteristics that are easily changed byloading user-specified coefficients developed.

User-specified characteristics
The latest generation of high-performance test instruments usesarbitrary FIR filters with characteristics that are easily changed byloading user-specified coefficients developed in Matlab or a similarapplication (Figure 5 above ).

This user-definable filtering provides powerful tools tosupporthigh-speed measurements. For example, the observation of signals hasbecome a key concern for designers measuring serial data deviceperformance, particularly with regard to the receiving elements. Thereare some signals that simply cannot be reached by any probe or testpoint.

Receiver input tolerance measurements have proven to beparticularlychallenging for this reason. Under normal circumstances, the receiverinput in a serial device is an almost meaningless access point forviewing signals.

This is because the signal of interest is processed by afilterwithin the device to offset the degradation that occurs duringtransmission through cables, PCB traces and connectors. The signal thatgoes into the active portion of the receiver — where eye diagrams andother characteristics must be evaluated — is encapsulated within thedevice and inaccessible using conventional techniques.

The solution is to use a DSP filter to mimic the effect ofthereceiver's internal filter. The user can load the same coefficientsinto the oscilloscope as were used to design the filter in the devicebeing tested.

With the filter applied, the oscilloscope user can probethe inputpin, and yet view the signal as it would be seen if the device could beprobed internally. This “virtual test point” reveals the receiver'spost-filter signal, even though the physical test point is a pin on thedevice package. This process is known as “de-embedding.”

DSP-based filters can be used to implement today'spreferredsignal-filtering techniques, including decision feedback equalization(DFE). Proprietary DFE filters are the intellectual property thatunderpins many of today's advanced serial transceivers. Digital filtersin an oscilloscope can accept arbitrary FIR filter coefficients thatallow a snapshot of DFE coefficients to be loaded into the scope forpost-processing of DFE signals.

DSP filtering can also be used to minimize the effects offixturesand cables connected to the device being tested. By characterizing ormodeling the fixtures and folding the information into appropriatefilter coefficients, the oscilloscope user can develop filters thattune out phase shifts and signal degradation imposed by the externalelements.
Using DSP to enhance acquisition performance
Digital signal processing can deliver benefits across the wholeoscilloscope acquisition system, including the enhancement of frequencyand phase response, channel matching, probe system performance,signal-to-noise behavior and other key features.

DSP-based channel performance enhancement can be used toachieveexceptionally flat magnitude response and phase linearity. Ideally, anoscilloscope's magnitude response stays constant over the fullfrequency range covered by its bandwidth — with no peaks or dips.

In conventional oscilloscope acquisition systems, thisideal isunattainable, but by using DSP it is possible to smooth anyirregularities and to level the response throughout the bandwidth. Thebenefit of this approach is superior measurement accuracy out to thelimits of its specified bandwidth.

For example, with a 12 GHz oscilloscope, a signal with afrequencyof 10 GHz can be captured with fundamentally the same degree ofaccuracy as a signal running at 100 MHz. Signal fidelity is consistentthroughout the range.

The instrument's frequency roll-off characteristic alsobenefitsfrom DSP treatment. Here the goal is to control the rate at which theresponse declines to provide the optimum balance between preservingtransient response and reducing out-of-band noise.

A roll-off that is too gradual allows more high frequencynoisecomponents into the measurement band. A roll-off that is too steep canattenuate the high frequencies that are needed to support accurate,smooth transient response. DSP makes it possible to very accuratelycontrol the slope of the roll-off and achieve an optimized balancebetween noise rejection and transient response, contributing to a veryhigh level of signal fidelity.

DSP can also be used to provide extremely accurate channelmatchingin which each channel is calibrated to the same ideal responsecharacteristics. Having a virtually identical step response acrossseveral channels is valuable when making pseudo-differentialmeasurements or channel-to-channel measurements on multi-lane serialtechnologies. The same techniques can be used to ensure accuratechannel matching between multiple instruments.

DSP can also be used in the probing signal path, enablingtheoscilloscope to account for the characteristics of the appropriatedifferential probe and its high bandwidth detachable tip. Here, the DSPsection acts as a nominal equalization filter that is dedicated to theprobe path, effectively integrating the probe into the oscilloscopesystem more tightly than ever before. This ensures the flattestfrequency response from the probe and oscilloscope working together .

Conclusion
The importance of signal integrity in overall systemperformance has increased with the advent of multi- gigabit per secondserial bus standards. A new generation of high-performance testinstruments meets the needs of serial bus developers by offering a highenough bandwidth and sample rate to support clean, accurate capture ofhigh-speed serial waveform characteristics and eye diagrams.

With ultra-low internal jitter, these tools can measuresignaljitter while minimizing the measurement contribution. In addition, newbuilt-in filtering tools enable the oscilloscopes to eliminatemeasurement path effects from the results — a requirement that isbecoming more common in serial bus standards.

These DSP filters have become an indispensable factor inthemeasurement of high-speed digital devices, especially the serialcomponents used in today's computing and network platforms. As thisarticle has shown, DSP tools can be used to counteract influences ofprobes and fixtures and allow designers to use “virtual test points” tosee signals as they would appear at inaccessible nodes in theirdesigns.

Trevor Smith is EMEA Market Development Manager,Oscilloscopesand Signal Sources, at TektronixInc.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.