标签: probe

My cup runneth over. Vendors keep sending nifty tools for measuring the current consumed by low-power embedded systems. Recently the intriguingly-named 3flareslab  delivered their Ampere analyser. It’s one of the few I’ve looked at where no claims are made about power measurements; instead they correctly state that the device tracks current.   Housed in a beautiful small metal box the thing looks like it could be run over by a tank and survive.    The Ampere analyser   The USB port on the side is used only to power the analyser; there are no comms with a host computer. The BNC goes to your scope, and a connector (not shown) on the side opposite the USB is for two wires that go to your embedded system. They run across a 0.1 ohm sense resistor inside the box.   Internally (I did take it apart) it’s basically that sense resistor feeding a multi-stage amplifier with a gain of 298. If your system draws one mA the scope shows a DC signal of 29.8 mV.   Sounds a little boring, doesn’t it? But one must understand the application. The analyser was designed to compete with Keysight’s N2820A current probe, a $4000 probe. The Ampere, and the N2820A, are used to watch dynamic current. For instance, what is the current profile of a system that goes from a deep sleep to awake, and taking data, and that then bursts a bit of Bluetooth for a handful of ms? A scope shot from their web site gives the idea:     It’s critical to understand your system’s current profile. Those short Bluetooth squirts may very well cause the board’s brown-out detection circuits to fire; or, if battery powered, the cell’s internal resistance may not be able to source the required current, so your system crashes.   Other specs are interesting. It can take plus or minus 10V. Few other units handle positive and negative inputs. That feature could be important when, for example, working with charging circuits.   The sense resistor is ½ watt. With a ±400 mA range it will read down into the microamps.   A resistor and an op amp? What’s the big deal? The most important spec is the unit’s 800 KHz bandwidth. That can be hard to achieve, especially with the stated 1% accuracy. Yes, you could build something like this. But you probably won’t.   It doesn’t like long sense wires going to the target; I found a lot of 60 Hz pickup using multiple clip leads. That went away with short point-to-point wiring.   The only thing I didn’t like about the Ampere is the price, at $475 CAD, or about $375 US.   Low power has always been important in the embedded space. I remember watching a Motorola processor run from the power of a pair of lemons back in the late 70s. But there has been a sea-change in the industry recently. Today MCUs will run off the energy of a few free-floating electrons, practically. Microchip was one of the pioneers. Now ARM MCUs are everywhere; Digi-key lists over 6000 part numbers for Cortex-M parts. Many of these will run for years from a battery. The so-called IoT is demanding ever more systems that run from batteries or from energy harvested from the environment. If you’re building low-power systems, do equip your lab with tools to measure current consumption.   After all, engineering is about holding our aspirations and designs to the cold hard truth of real numbers.

In a series of columns recently, Guide to probing and Another guide to probing , I discussed some of the issues that arise in using scope probes as frequencies increase. Alas, too few engineers do a real analysis. Turns out, a decent (think over $300) 10 pF 10X probe at 100MHz looks like a 160Ω load! That's 14 mA at 2.3V, which is more than many gates (not to mention CPUs) can drive. In other words, putting a probe on a perfectly good circuit may cause the system to stop functioning. But it gets worse as the frequency goes up. In the following graph probe impedance in ohms is shown on the vertical axis and frequency along the horizontal. Impedances are shown for three different probes. Note that 10 pF is pretty common for decent probes; better ones, like Tektronix's very nice 5 pF TPP1000 approach a thousand dollars.     Agilent recently introduced the Infinium 90000, a 33GHz scope. ( I want one of those! ) But how do you probe a signal that fast? A 5 pF probe will look like a one ohm load at that speed. That's over 2 amps at 2.3V. The answer, of course, is to buy special probes, which run about $30k a pop. A set of four will buy a house in the Midwest. For those of us working at more modest frequencies, say 50 to a few hundred MHz, one can steal an idea from High-Speed Digital Design by Howard Johnson and Martin Graham. I briefly mentioned this in the second of the two referenced articles, but quite a few people asked for more details.   The probe is simplicity itself. Note in the figure above that a typical quarter watt resistor has about 0.5 pF of capacitance. Get a meter of RG-58/U coax. On one end install a decent BNC connector ( or, just buy cable with a preinstalled BNC ) and solder the resistor (with short leads) to the inner conductor at the far end. Then as show in the figure below solder the other end of the resistor to the node being probed, and solder the braid ( very short ) to a ground near the node. The result looks like this:   ( Note the SOT-23 package I'm probing is so small you can't really see it in this picture, or in real life if you have the slightest myopia ).ÿ It's very important to set your scope's input impedance to 50Ω, since most scopes default to 1 M? or so. If your scope doesn't have a selectable impedance buy a 50 attenuator ( Agilent's N5442A or Test Products International's 120082, which is $56 from Digi-key ). Now you have a 0.5 pF probe which divides the input by a factor of 21 ( i.e., this is a 21X probe ). And its performance, shown in the red line in the figure below is pretty stunning:   The downside is that this probe isn't as easy to use as one from a vendor. After all, you will have to solder it in place every time it's moved. An alternative is a commercial low-capacitance probe, which will set you back about five grand each. Probably the most important take-away here is to understand that everything is part of your circuit. Even humidity can affect sensitive analogue designs. And your test equipment is part of the circuit. No scope, logic analyser or any other device is perfect; they will all interact in lesser or larger ways with your board. Always do an engineering analysis to understand how things will behave.  

I didn't have any X1 probes around, so put a 100-pf capacitor on the node to simulate a really crappy probe. Rise time spiked to 5.5 nsec, more than a five times increase, and the signal was delayed by almost a nsec. I suggest immediately combing your lab for X1 probes and donating them to Goodwill. And be very wary of ad hoc connections—like clip leads and soldered-in wires—whose properties you haven't profiled.   But 100 pf is a really crummy probe. I soldered a 30 pf cap on the node to simulate one that's somewhat like an ad hoc connection or a moderately-cheap probe. In Figure 4 , the orange trace is the gate's output with no load—just the 21X probe. The green is with the additional 30 pf. The distortion is significant. So a 30-pf probe grossly reshapes the node's signal. What effects could that cause? First, everything this signal goes to will see a corrupt input. If it goes to a flip flop's clock input the altered rise time could cause data to be incorrectly latched. Or, if the flop's data input(s) are changing at roughly the same time, the flop's output could become metastable—it'll oscillate for a short time and then settle to a random value. If it goes to a processor's non-maskable interrupt input the leading-edge bounce could cause the CPU to execute two or more interrupts rather than one. (Generally this is not a problem for normal maskable interrupts since the first one disables any others). But wait, there's more. Note that the signal extends from well below ground (about -600 mV) to 3.7V (be sure to factor in the attenuation of the 21X probe), which is much higher than the 2.5V Vcc. Depending on the logic family this signal goes to, those values could exceed the absolute maximum ratings. It's possible the driven device will go into SCR latchup, where it internally tries to connect power to ground, destroying the device. I have seen this happen: the chips explode. Really. It's cool.     So far I haven't shown any signals acquired by the N2890A. The yellow trace in Figure 5 , is the gate's output using that probe. It's pretty ugly! The distortion is entirely in the probe, and not on the board, so does not represent the signal's true shape. In this case the probe is grounded using the normal 3-inch clip lead. Using the formula from last month, that loop has 61 nH of inductance. In orange the same signal is displayed, but in this case I removed the probe's grabber and connected a very short, about 5 mm, ground wire to the metal band that encircles the tip. The signal is still not displayed correctly—it extends below ground and has a total magnitude of about four volts, much more than the 2.5 Vcc. But the better grounding did clean up the shape. The point is that poor grounding can cause the scope to display waveforms that don't reflect the node's real state. Electronics matters Many in the digital world find themselves divorced from electronics. We think in ones and zeroes, simple ideas that brook little subtlety. A one is a one, a zero a zero, and in between is a no-man's land as imponderable as the "space" that separates universes in the multi-verse. But electronics remains hugely important to digital people. Ignore it at your peril. Power supplies have crawled below a volt so the margin between a one and a zero is ever-tighter. On some parts the power supply must be held ±0.06V or the vendor makes no promises about correct operation. On a 74AUC08, typical fast logic, at 0.8 Vcc there's only a quarter volt between a high and a low. Improper probing can easily skew the node's behaviour by that much. And, as we've seen, capacitance and inductance are so vital to digital engineering that we dare not ignore their effects when troubleshooting. Reactance, impedance, and electromagnetics are big subjects that I've only lightly touched on. They're pretty interesting, too! I highly recommend the book High-Speed Digital Design for a deep and dirty look at working with high-speed systems. 1 The ARRL Handbook from the American Radio Relay League is possibly the best introduction to electronics available. 2 It doesn't skimp on the math, but never goes beyond complex numbers. The focus is decidedly on radios, since this is the bible of ham radio, but the basics of electronics are covered here better than any other book I've found. There's a new edition every year; my dad bought me a copy in 1966, and since then I've "upgraded" every decade or so. Endnotes 1. Johnson, Howard and Martin Graham. High-Speed Digital Design ,1993 PTR Prentice-Hall Inc, Englewood Cliffs, NJ. 2. The ARRL Handbook , American Radio Relay League. Published afresh every year. www.arrl.org .  

In my previous article , I gave a mostly theoretical overview of the effect probes—like scope and logic analyser probes—have on the nodes being tested. The most important effects stem from the capacitance of the probe tip. To reiterate, the reactance, or resistance to AC, at the tip is:   This reactance loads the node and can alter a device's operation—or worse. To explore this, I built a circuit on a printed circuit board with ground and power planes, keeping all wires very short. A 50MHz oscillator drives two AND gates. The 74AUC08 is spec'd with a propagation delay between 0.2 and 1.6 nsec at the 2.5V I used for the experiment. The second gate is a slower 74LVC08 whose propagation delay is 0.7 to 4.4 nsec. Still speedy, but slower than the first gate. I was not able to find rise-time specifications but assumed the faster AUC would switch with more alacrity and thought it would be interesting to compare effects with differing rise times. Alas, it was not to be; the LVC wasn't much slower than the AUC. So I'll generally report on the slower gate's results. These parts are in miniscule SOT-23 packages, which keeps inductances very low but means one solders under a microscope, sans coffee. I wanted to see the effect that probes have on nodes, but that posed a meta-problem: if probing causes distortion, how can one see the undistorted signal? Thankfully there's a simple solution. I made a pair of metre-long probes from RG-58/U coax cable. A BNC connector on one end goes to the scope. A short bit of braid is exposed and soldered to the ground plane very close to the node being probed, and a ¼-watt 1K resistor goes from the inner conductor to the node. I used an Agilent MSO-X-3054A scope with selectable input impedance, set to 50Ω. This is critical for the shop-made probe; the normal 1 MΩ simply will not work. If your scope doesn't have a 50-Ω mode, use a series attenuator such as the 120082 from Test Products International (this part doesn't seem to be on their web page, but Digikey resells them). Agilent's N5442A is a more expensive but better-quality alternative. RG/58U is 50-Ω cable; add the resistor and the total is 1,050Ω. The scope's 50-Ω input forms a 21:1 divider, but the resistor's very low capacitance (remember, a ¼-watt resistor runs only 0.5 pf) means the probe's tip looks extremely resistive, with little reactance. The scope thinks a 1X probe is installed, so to accommodate the oddball 21:1 ratio one multiplies the displayed readings by 21.   The first experiment showed Fourier at work. The blue trace in Figure 1 shows the output of the fastest gate using a 21X probe. Note that it's far from perfect since the circuit had its own reactive properties. The rise time (measured with a faster sweep rate than shown) is about 690 psec (picoseconds). "About" is the operative word, as the scope has a 500MHz bandwidth (though samples at 4 GS/sec). I found that having the instrument average readings over 128 samples gave very consistent results. The pink trace is the Fourier Transform of the gate's output. Unlike the blue trace, this one is not in the time domain (e.g., time across the horizontal axis) but is in the frequency domain. From left to right spans 2GHz, with 500MHz at the centre. The vertical axis is dBm, so is a log scale. Each peak corresponds to a term in the Fourier series. Point "A" is exactly 50MHz, the frequency of the oscillator. Most of the energy is concentrated there. Peak "B" is 48 dBm down from "A." That's on the order of 100,000 times lower than "A." "B" is at 900MHz. Remembering that little energy remains in frequencies above with F=900MHz the rise time is 555ÿpsec, close enough to the 690 measured. The same experiment using the slower 74LVC08 gate yielded 48ÿdBm down at 450MHz, or a rise time of 1.1ÿnsec. That's close to the 0.95ÿnsec reported by the scope. Next, I connected a decent-quality $200 Agilent N2890A 500MHz probe (11-pf tip capacitance) on the 74LVC08's output. The 21X probe saw an additional third of a nanosecond in rise time due to the N2890A's capacitance. In other words, connect a probe and the circuit's behaviour changes.   In Figure 2 the orange trace is the gate's output measured, as usual, with the 21X probe, although now there's 10ÿinch of wire dangling from it. That trace is stored as a reference, and the green one is the same point, with the same probe, but the N2890A is connected to the end of that 10 inch of wire. Note that the waveform has changed—even though that other probe is almost a foot away—and the signal is slightly delayed. This is probably not going to cause much trouble.   Gates typically have a very low output impedance, so it's unsurprising there's so little effect. Often, though, we're sensing signals that go to more than one place. For instance, the "read" control line probably goes from the CPU to quite a few spots on the board. To explore this situation, I put the 21X probe five inches down that wire, captured the waveform into the reference (orange in the figure above ), and then connected the same N2890A at the end of the 10 inch of wire. The signal (green) at the 5-inch point shifted right and was distorted. Consider the clock signal: On a typical board, it runs all over the place. The impedance at the driver is very low, but the long PCB track will have a varying reactance. Probe it and the distortion can be enough to cause the system to fail. The ringing is caused by an impedance mismatch. The N2890A has changed the node's impedance, so it no longer matches that of the driver. Part of the signal is reflected back to the driver, and this reflection is the bounciness on the top and bottom of the pulses.    

Fourier fits But it gets worse. In 1822, Joseph Fourier released his Théorie analytique de la chaleur ( The Analytic Theory of heat ). That seminal work has tormented generations of electrical engineering students (and no doubt others). Discontinuous functions—like square waves—are very resistant to mathematical analysis with calculus unless one does horrible things like use unit step functions. But Fourier showed that one can represent many of these periodic functions as the sum of sine waves of different amplitudes and frequencies. The Fourier Series for a square wave is:   The series goes on forever, so all square waves have frequency components going to infinity. However, the amplitude of these decrease rapidly due to the division by an ever-larger odd number. The point, though, is that the "frequency" of a square wave is composed of many frequencies higher than that of the baseband. Pulses, like the ones that race around every digital board, the ones we probe with our scopes and logic analyzers, are square-wave-ish. The good news is that they're not perfect square waves: obviously, with the exception of clocks, they rarely have a 50% duty cycle. Pulses are also, happily, imperfect. Fourier's analysis assumed that the signal transitions between zero and one instantaneously. In the real world every pulse has a finite rise and fall time. If T r is the rise time, then the frequency components above F in the following formula will be so far down they're not important:   This does mean that, assuming a 1-nsec rise time, even if your clock is ticking along at a leisurely rate about the same as a Florida old-timer's speedometer, the signals have significant frequencies up to 500MHz. Those unseen but very real frequency components will interact with the scope probe. Ad hoc formulas Long troubleshooting sessions often see a board covered with connections to test equipment, data loggers, etc. Long lengths of wire-wrap wire get soldered between a node and an instrument. These connections all change the AC properties of the nodes by adding inductance and capacitance. Here are some useful formulas with which one can estimate the effects. These are derived from book High-Speed Digital Design (Howard Johnson and Martin Graham, 1993 PTR Prentice-Hall Inc, Englewood Cliffs, NJ), and there's much more useful data in that book. Most of us use multilayer PCBs that have one or more ground and power planes. Solder a wire to a node and drape it across the board as it runs to a scope or other instrument, and you'll add capacitance. If d is the diameter of the wire in inches, h is the height above the PCB, and l is the length of the wire, then the capacitance in pF is:   (A better solution is to run the wire straight up from the node, perpendicular to the PCB.) AWG 30 wire-wrap wire is 0.0124 inches in diameter. Typical hook-up wires are AWG 20 (0.036 inches), AWG 22 (0.028 inches), and AWG 24 (0.022 inches). The inductance of the same wire in nanohenries is:   The inductance of a round loop of wire (for example a scope probe's ground lead) in nH is, if d is the diameter of the wire and x is the diameter of the loop:   Next month we'll look at some real-world data.