Probing for the Truth

With multigigahertz signals, selecting the right probes and ensuring that your pc boards accommodate them can be just as important as picking the right scope. Indeed, your probe choice may decide which scope you buy.
If your signal is distorted beyond recognition even before it reaches the input amplifier of your new 6GHz-bandwidth, real-time-sampling digital scope, your management may decide that investing US$60,000 or so in the scope was a really, really dumb idea. Before you recommended the purchase of such a scope, you should have received answers to serious questions about how you would get signals into the instrument. If you failed to do that groundwork, the people you used to work with will probably remember you as the person responsible for the dumb idea. In short, successfully acquiring signals that contain components at 5GHz and above is a lot trickier than probing slower signals.

Inattention to probing can necessitate redoing high-speed pc-board layouts. Continued com-ponent miniaturization, shrinking lead spacing, and the rapidly growing use of low-voltage differential techniques to transmit digital data at ever-higher rates requires you to keep probing in mind from the start of your board designs. Failure to do so can cost your project-and your career.

Fortunately, scope manufact-urers are keenly aware of the importance and growing difficulty of probing. Their new probe products aim to reduce the burden on board designers who must view multigigahertz signals, but the reduced burden comes at significant cost: A set of four differential active probes, each with 5 or 6GHz bandwidth, usually costs nearly half as much as an entire 6GHz-bandwidth scope without probes. If you add any of a large number of options that are common in high-end scopes and order the instrument with four wideband differential probes, the total amount of your purchase can top US$100,000.

The differential difference
Probing wideband differential signals can make using active differential probes a virtual necessity (Reference 1). At lower frequencies, using single-ended probes and taking the difference between signals on pairs of channels may prove acceptable, but at higher frequencies, that approach almost guarantees failure. Unless you can limit the length of the single-ended probes' ground leads to a few millimeters, the relatively large high-frequency currents that flow through the ground-lead inductances add significant ac voltage drops that distort the signals enough to make them meaningless. Taking the difference between the outputs of a pair of single-ended probes offers little hope of rescuing this situation. Neither the ground leads' ac-voltage drops nor the scope channels' high-frequency characteristics are likely to be adequately matched. In addition, this approach uses two scope channels for each differential signal, in effect reducing your four-channel scope to a two-channel instrument.

Wideband differential probes eliminate or greatly reduce these problems. Considering the frequencies involved, the probes' CMRR (common-mode rejection ratio) is surprisingly good, often exceeding 26dB (20-to-1 CMRR) at several gigahertz. Moreover, the probe's two signal connections are frequently adequate, so you can do without a separate connection from the probe ground to your circuit ground. When you do need such a connection, the current it carries is usually small enough that the lead can often be more than an inch long and need not have an inconvenient geometry; that is, the connection need not, for example, be a wide, flat plate.

The major disadvantages of differential probes are their size and cost, both of which are somewhat greater than those of single-ended probes. A perhaps unexpected advantage of several wideband differential probes is that when you use them for single-ended probing, you can simply leave one of the inputs unconnected; you need not ground or otherwise terminate it.

Ohms are where the art is
A complication with using scopes that have bandwidths greater than approximately 3GHz is that such instruments generally do not provide as a standard feature narrower-bandwidth scopes' familiar 1M½ input impedance-actually, the resistive component of the parallel R-C input circuit. Wideband scopes have a truly resistive 50W input impedance. Those that offer 1M½ do so by means of optional adapters that mate with the scopes' wideband input connectors. Installing these adapters usually reduces the scope bandwidth to 500MHz. Some scopes with bandwidth of 3GHz and below allow you to switch between internal 50½ and 1M½ input circuits, still with 500MHz bandwidth at 1M½. The 3GHz bandwidth limitation is a consequence of the more convoluted signal paths that must accompany the switchable impedance. These signal paths preclude the faithful acquisition of signals containing significant energy beyond approximately 3GHz.

In large numbers of wideband-scope applications, the 50½ input necessitates the use of some sort of probe. In many such applications, passive attenuator probes are acceptable. Such probes are always single-ended. Ultra-wideband passive probes usually have dc input resistances of 500½ with 10-times signal attenuation or 1000½ (with 20-times attenuation).

A key parameter is the probe's input capacitance. The best wideband passive probes have an input capacitance of less than 0.25pF. At 6GHz, this capacitance presents a reactance of a little more than 100½. This reactance appears in parallel with the 500 or 1000½ resistance, swamping it out, for most practical purposes, and creating the possibility of series resonance with the inductance of leads that you may add to help the probe and its ground reach your test points. In some wideband active probes, the input capacitance does not appear alone in parallel with the probes' input resistance. This feature represents yet another advantage of certain active wideband probes.

Different philosophies
The three major wide-band-digital-scope manu-facturers-Agilent, LeCroy, and Tektronix-make probes as well as scopes. The companies view the combination of the scope and its probes as one system. Although you can sometimes use a wideband probe from one manufacturer with a scope from another, the combination almost invariably involves com-promises and inconveniences. These problems don't occur if you use probes that the manufacturer designed for use with your scope. For example, although the primary analog signal connections may use compatible connectors, the con-nectors that enable the scope to power the probe and that carry ancillary information, such as attenuation ratio, from the probe to the scope are generally incompatible among products of different manufacturers.

Because the manufacturers have recently adopted markedly different philosophies of designing wideband probes, it is not necessarily an overstatement-or a case of the tail wagging the dog-to suggest that your choice of probe can determine your choice of scope. The most significant recent development is Agilent's and LeCroy's recognition that designing active probes that possess all of the attributes that high-speed-system designers need requires resistively damping the resonance between the probe's input capacitance and the inductance of the wiring between the signal source and the probe input. The damping resistor must be near the signal source; you can't realize the desired benefits if you place it-as some probe designs have tried-at the end of the connection nearest the probe body.

This realization, combined with a host of other design requirements, led Agilent to design the Infiniimax probe family (Figure 1). These probes place a tiny passive damping resistor and resistive divider within a few millimeters of the signal source and use flexible coaxial cable of as long as several inches to connect the minuscule assembly to the probe body's active circuits. Tektronix insists that Agilent didn't invent resistive damping in wideband probes and that Tek probes have used the technique for years. But Agilent counters that Tek hasn't always placed the damping resistor near the signal source and that, so far at least, the company has not been willing to separate the damping resistor from the probe body. According to Agilent, the resulting one-piece designs force users to try to get the probe body so close to the test point that the probe won't fit, especially if the probe must contact differential signals on closely spaced pins.

Figure 1

Figure 2

Tek's latest differential active probe, the 5GHz-bandwidth P7350, addresses this problem through extreme miniaturization of its active electronics (Figure 2). The probe's body is amazingly small for a unit that contains such sophisticated electronics, even though the probe design doesn't-as do Agilent's InfiniiMax probes-allow several inches of flexible cable between the portion that contacts the test point and the part that contains the amplifier.

Probing styles
Anyone who has used a traditional narrowband scope probe-even a passive one-is familiar with the browser configuration, which is intended for quickly contacting and viewing large numbers of signals. Historically, you held the probe in your hand as you browsed, but the tiny geometries of the high-speed components that you must now probe often require holding probes with something steadier than a human hand. Now, EEs often browse with probes mounted on small manipulators or stands (Figure 3).

Figure 3

Probes with plugs that you solder directly to the pc board under test or that plug into a socket on the board are newer develop-ments. These configurations are especially convenient if you can identify a limited number of signals to which you will frequently connect and your probe uses series damping resistors and attenuators to drive the wiring between the board and the amplifiers in the probe body or bodies.

Besides making multiple solid connections to ultra-high-speed boards, these approaches minimize the probe's input capacitance. Of course, your board must incorporate solder-down pads for the probe plug or a socket that mates with the plug. In addition, the board's circuit design must anticipate not only the require-ments for driving the traces that carry signals from the real sources (usually IC pins) to the solder pads or socket, but also-at least for the socket version-problems from reflections caused by unterminated stubs. The stubs can be a problem if you want to use your board without plugging in the probes.

LeCroy's newest line of differential active probes-the WaveLink family, which includes a unit having nominal bandwidth of 7GHz-focuses on browsing. A thumb wheel enables you to set the spacing between the nickel-titanium-alloy probe tips to any distance between zero (probes touching) and 0.3in. The spacing holds its setting, and you can twist and bend the tips without undue concern about damaging or permanently deforming them. Contrast this design with that of Tektronix's P7350, in which the probe-tip spacing is fixed. LeCroy says that the WaveLink probes' mechanical and electrical architect-ure is conducive to additional designs for different styles of probing and that such versions will probably debut during 2003.

Lower input resistance
In designing the WaveLink probes, LeCroy made an unusual decision to make the resistive component of the input impedance relatively low-nominally 4k½ between the noninverting and inverting inputs. Because this resistance is substantially less than that of Agilent and Tektronix differential probes, you might think it too low. But remember that, at merely 159MHz, an input capacitance of just 0.25pF-with no impedance in series-places a reactance of 4k½ in parallel with the 4k½ input resistance and drops the input impedance magnitude to 2.8k½. With no resistance or inductance in series with the input capacitance, the input impedance's reactive component drops by half each time the frequency doubles. Thus, even a resistance as low as 4k½ becomes a matter of primarily academic interest throughout more than five octaves of the probe's passband.

Besides making the input impedance more nearly constant over a wider band of frequencies than would a higher resistance, the low resistance allows inserting a smaller value of attenuation ahead of the probe's input amplifier, thus improving the probe's SNR. Compared with probes having higher input resistance, the lower SNR adds less jitter to low-amplitude signals, such as those common in high-speed differential signaling, thereby improving the accuracy of jitter measurements. You might think that the low input resistance would make the WaveLink probes more prone to overvoltage damage than units with higher resistance, but LeCroy says that no damage results from inadvertent application of 25V dc.

When Agilent introduced the InfiniiMax probes, the company had a problem with the way it expressed the probes' input capacitance. The capacitance value, which can be as great as 0.7pF, depends on the whether the probe is single-ended or differential and on the probe's physical style. Because the data sheet did not include an equivalent circuit, you could easily reach the incorrect conclusion that the 0.7pF appeared by itself across the probe's input, as does the input capacitance of most scopes and probes. If this were the case, at 6GHz, the input impedance would consist mainly of a capacitive reactance of slightly less than 38½.

However, unlike that of conventional probes, the input capacitance of the InfiniiMax probes does not appear alone across the probe input. The simplified equivalent circuit of Figure 4 shows an approximately 200½ resistor, R1, in series with the input capacitance. In the socket and solder-in configurations, this resistor resides in the tiny assembly that contacts the probe point on the unit under test. This series resistor not only places a lower bound on the probe's input impedance, but also damps the unavoidable resonance between the series inductance and the distributed capacitance of the leads that connect the probe point to the input attenuator. The attenuator protects the probe's input amplifier and matches the Z0 (characteristic impedance) of the coaxial connection to the amplifier, which resides in the probe body. The amplifier's input impedance is also Z0, so the coax is properly terminated at both ends, enabling several inches of flexible cable to separate the probe point from the probe body. In fact, were it not for the cable's dielectric losses, which can become significant at and above several gigahertz, this cable could be many inches long.

Figure 4

Interaction between the probe and the unit under test is unavoidable. But if connecting the probe alters the nature of the signal, how do you determine what the signal looks like in normal operation when the probe isn't present to load the signal source? The answer is that you have to use modeling and work backward. With good models of the probe and the circuit that drives it, you can mathematically infer the appearance of the signal at the source when the source isn't driving a load.

If the manufacturer supplies a sufficiently detailed circuit model of the probe, deriving the signal source's unloaded waveforms can be a purely mathematical exercise. However, some probe manu-facturers also offer characterization fixtures to help signal-integrity engineers, who often have special probe-characterization require-ments. Signal-integrity engineers use these fixtures in conjunction with instruments such as vector-network analyzers to construct their own probe models. LeCroy provides characterization fixtures at no extra charge with its WaveLink probes.

Probe manufacturers also argue over whether they should separately characterize probes' loading and transmission effects. According to the argument against separate characterization, if you cannot observe the signal source's output waveform without loading the source, then all you should want to know are the probe's input and output waveforms. The opposing argument is that the loading effects depend upon the signal-source characteristics-particularly the source's internal impedance, which is a property of the source but not of the probe. If you lump together the source-related and probe-related effects, you can become seriously confused about your unit under test's true behavior.

A rather similar argument relates to the combined behavior of scopes and probes (Reference 2). Unfortunately, you can't use the rules that apply to narrowband scopes and probes to determine the combined rise time and bandwidth of ultra-wideband scopes and probes. For these devices, where TR° Ã (Ts2 1 TP2), TR is the 10 to 90% rise time of the scope plus the probe, TS is the scope rise time, and TP is the probe rise time. For example, Agilent says that the 10 to 90% rise time of its 6GHz-bandwidth Infiniium scope is 70psec and that, without the scope, the InfiniiMax probe intended for use with this scope has a bandwidth greater than 7GHz. Both bandwidths are at the -3dB point. Yet, when you connect the probe to the scope, the bandwidth from the probe tip to the screen is still 6GHz. Presumably, the scope's bandwidth without the probe is somewhat greater than 6GHz, and the scope displays rise times somewhat longer than 70psec for signals that come through the probe. Of course, with or without the probe, you must measure the scope's rise time with a signal whose own rise time is significantly less than 70psec. The signal must also exhibit little or no overshoot. Such sources are difficult to find.

Connectors: oh, the pain! Narrowband scopes use BNC coaxial connectors, which are ubiquitous and whose bayonet locks make them secure when mated yet convenient to mate and unmate. Standard BNCs are unreliable beyond approximately 4GHz, however. The problem is that, with repeated insertion and withdrawal, the spring fingers in a panel-mounted BNC can lose some of their force and produce a high-resistance shield connection whose characteristics vary as you wiggle the cable or as you mate and unmate the connector. At freq-uencies of 4 to 26.5GHz, the industry-standard coaxial connector is the SMA, which owes its superior reliability to the use of a screw-down shield as its locking mechanism.

Both LeCroy and Tektronix use modular input-connector systems on their ultra-wideband real-time-sampling digital scopes. The modules that provide an SMA interface are the most popular because they allow the use of the scopes' full 6GHz bandwidth. Both companies also offer BNC modules. You can obtain such modules with built-in networks that convert the scopes' 50½ inputs to 1M½. Converting to the higher impedance reduces the bandwidth to 500MHz, however.

In its 6GHz-bandwidth Infiniium scope, Agilent uses a different approach. The input connector is BNC-compatible but is not a true BNC (Figure 5). Instead, it is an ingenious, proprietary, 17.9GHz design with an air dielectric and a lockdown mechanism that works with bayonet-locking BNCs but can also accommodate a proprietary mating connector that uses both bayonet and screw locks. Once the screw mechanism locks together the cable- and panel-mounted connector halves, you cannot move the connector halves with respect to each other. Thus, the design defeats the principal cause of conventional BNCs' unreliable high-frequency operation.

Figure 5

For some of Agilent's scope customers, this connector probably proves superior to the SMA. Agilent points to the large number of BNC-equipped accessories it offers. Many potential customers for the 6GHz scope already use these accessories with narrower-bandwidth Agilent scopes. Although the BNCs on those accessories mate with the new scope's input connectors, the scope still doesn't work properly-or at all-with many of the accessories. The reason is that, with the exception of cables, which work but don't reliably allow the use of the scope's full bandwidth, many of the accessories are intended to drive 1M½ inputs and hence won't drive the 6GHz scope's 50½ inputs. Agilent says, however, that it will provide a 1M½ adapter if enough customers request it.

Author Information
Senior Technical Editor Dan Strassberg got the idea to write an article on wideband scope probes while he was preparing for his Feb 6, 2003, EDN article 'The scopes trial.' He holds a BSEE from Rensselaer Polytechnic Institute (Troy, NY) and an MSEE from the Massachusetts Institute of Technology (Cambridge). He belongs to the same engineering and scientific honor societies (Tau Beta Pi, Eta Kappa Nu, Sigma Xi) as do most EDN readers and is a member of the IEEE and the National Society of Professional Engineers (NSPE). He also holds three patents and is a registered professional engineer in Massachusetts. You can reach him at 1-617-558-4405, fax 1-617-558-4470.

References
1. McTigue, Mike, 'Single-ended or differential? That is the question,' EDN, Feb 6, 2003, pg 77.
2. Weller, Dennis, 'Relating wideband DSO rise time to bandwidth: Lose the 0.35,' EDN, Dec 12, 2002, pg 89.

At a glance

• Ultra-wideband digital scopes' 50½ inputs often make probes essential.


• All manufacturers of ultra-high-bandwidth scopes also make probes and all now offer differential active probes appropriate to scopes with bandwidths as high as 6GHz.


• Probe-design philosophies differ, and wideband probes work best with the probe manufacturer's scope; thus, your probe choice can determine your scope choice.


• If you can afford and have room for them, differential active probes simplify high-fidelity acquisition of multigigahertz signals.


• Layouts of pc boards that carry wideband signals must anticipate probing requirements.

Sidebar: Designing pc boards for maximun signal fidelity at multigigabit-per-second data rates

As the insatiable desire for performance in the digital domain continues to increase data rates and circuit densities, engineers have adopted design rules for signal integrity, such as allowing for transmission-line effects and minimizing parasitics.

Even though many EEs have internalized these rules, they haven't yet begun to strike the necessary balance between signal integrity and signal fidelity. Without signal fidelity-how faithfully the measurement system acquires and represents the signal under test-you can't determine the performance of your circuits.

Now, more than in the past, the balance between the design goals of signal integrity and signal
fidelity is critical to validating, characterizing, and debugging a design. To achieve the proper balance, answer these questions before finalizing a circuit layout or board design:

What types of signals will you measure?

Determine whether the signals are single-ended (ground-referenced) or differential and consider other characteristics, such as amplitude and rise time. Will you need to probe the board to acquire the signals at the transmitter, at the receiver, on traces or vias, or on a connector? An industry standard, such as PCI Express, may specify the signal type and location. Alternatively, the design's most critical performance requirements, such as timing and jitter margins, may govern where and how you probe.

What probes will you need?

Choose the appropriate probe-single-ended or differential-to match the signal requirements. Differential probes are best for measuring differential signals but are often good for single-ended measurements. For differential measurements, you can sometimes also use a pair of single-ended probes in combination with the scope's subtraction capability. However, this approach can create problems, such as measurement errors and excessive cost. Be certain that the probes you choose meet the signals' requirements for sensitivity and dynamic range.

What probe bandwidth will you require?

Because you use oscilloscope probes primarily for time-domain measurements, a probe's rise time is often more important than its bandwidth. However, a probe's most common banner specification is bandwidth, and the signal's rise time dictates the probe's bandwidth. To convert rise time to the equivalent bandwidth, a useful rule of thumb is that the signal's bandwidth is equivalent to approximately 0.4 divided by the 10 to 90% rise time. To maximize signal fidelity, the probe bandwidth should exceed this calculated signal bandwidth by the widest possible margin.

Another way to achieve good signal fidelity, which standards such as InfiniBand and FibreChannel recommend, requires that probe bandwidth be at least 1.8 times the digital signals' bit rate. Also, the key to meeting overall test-system-bandwidth requirements is selecting a probe whose bandwidth is appropriate for the oscilloscope with which you use it.

What test-point spacing do you need?

Make sure test points are accessible and compatible with the probe you will use. Make the points available for probing without compromising signal integrity or signal fidelity. Traces or vias should be accessible on the board's surface. Place grounds symmetrically on either side of a differential signal pair, especially if you plan to use single-ended probes. Space differential lines and test points to best match the spacing of your probe tip. If matching the board to the probe spacing is impractical, be sure to use the shortest probe-tip adapters and relieve the solder mask around the intended test points or runs. For single-ended measurements, route a ground point as close as possible to the test point and make sure that the ground routing introduces no excessive inductance between the ground point and the true ground.

How does the probe load the circuit under test?

Connecting any probe to a modern high-speed circuit inevitably affects the circuit to some degree. To enable simulating the loading during circuit design, some vendors now supply probe-load models. Others supply fixtures that facilitate the development of loading models.

Author's Biography
Bill Hagerup is a hardware-design engineer in the Measurement Accessories Division of Tektronix (Beaverton, OR). He has been designing high-speed analog circuits at Tektronix for more than 10 years and has received two patents on differential probing. He holds an MSEE from Oregon Graduate Institute (Beaverton), a BSEE from Oregon State University (Eugene), and a BS in Physics from the University of Portland (Oregon).