For novel ideas about building embedded systems (both hardware and firmware), join the 40,000+ engineers who subscribe to The Embedded Muse, a free biweekly newsletter. The Muse has no hype and no vendor PR. Click here to subscribe.

Review of Keysight's MSO-X 3054A Mixed Signal Scope (February 2011), by Jack Ganssle

What is a pHz?

Wikipedia tells me pHz stands for "picohertz," or 10-12 Hertz. fHz means "femptohertz" - three orders of magnitude smaller. A one fHz signal would take about 20 million years to complete a single cycle. Four cycles ago dinosaurs roamed the Earth.

When confronted with a new oscilloscope the first thing I do is connect a probe to the calibration node and look at the resulting square wave. I tried this with the new MSO-X 3054A from Agilent Technologies and then pressed "Measure." The display showed the expected waveform, with current, mean, min and max frequencies, all at 1.0012 kHz. But it also showed the standard deviation as a few tens of fHz, and later pHz, prompting my dive into Wikipedia.

The same sense of surprise dogged me for most of my evaluation of this product, which was a meta-surprise in itself. I don't expect much novelty in looking at scopes anymore. Until the advent of digital versions the technology evolved very slowly. A World War II version wouldn't be much different, other than bandwidth, than one from the 80's.

Fifteen years ago HP startled me with their 54645D MSO  - Mixed Signal Oscilloscope - which combined both a logic analyzer and a scope in one package. The brilliance of this device was its cross-triggering: it could start a sweep either on a combination of logic conditions, or on a normal scope-like analog threshold crossing. Suddenly we embedded types could see how our digital circuits interacted with the analog components. A new day had dawned for debugging embedded systems.

Later HP spun off their test equipment division into Agilent, a move that always puzzled me, since that group represented both the origins of the company as well as its finest engineering. But perhaps that was for the best since Agilent was spared the slings and arrows of Carly.


The screen updates up to a million waveforms/second on the screen. A demo dramatically shows how a slower rate masks glitches. And in using the scopes I consistently found them to be extremely fast, with the exception of computing an FFT.

USB, GPIB and Ethernet connectivity are available on the scopes, of course, but the Ethernet is LXI-compliant. LXI is a standard aimed at the test and measurement world that specifies what kinds of capability, accessible via a standard LAN, is available in the instrument. For instance, it defines standards for triggering. If you have two instruments at opposite sides of, say, an airplane, the LXI standard insures you can trigger both devices from the same signal.

In the olden days scopes had three basic bits of functionality: a vertical amplifier, time base, and triggering, all controlled by a sea of knobs and buttons. It took a lot of electronics to provide those features, so packages were quite large, yielding plenty of panel space for the controls. Tiny displays freed up even more space. No longer: these models occupy less than half a cubic foot and nearly half the panel is devoted to an enormous (8.5" WVGA) screen. Like many other scopes today soft keys are used to control many of the vast array of features, though all basic scope functionality has dedicated controls.


Agilent sent me two scopes representative of the bottom of the line (the DSO-X 2002A) and the top (MSO-X 3054A), which have identical form factors and look and feel. Both scopes came with all available options, which initially caused me some confusion and then delighted surprise. At first I couldn't find that 1 kHz calibration square wave. These instruments have two scope compensation lugs instead of the usual one, and somehow, due to excessive button pushing (who can resist all of those seductive knobs?) I'd enabled training mode. That generates one of a large variety of signals on these lugs ranging from a simple sine wave to RF bursts, runt pulses, and even streams of serial data like CAN and I2C. This is targeted to the educational market and is accompanied by a quite large book that takes students through from the basics of using an oscilloscope to quite advanced techniques. But at $500 I'd recommend buying the option even if you're an old pro. We've always used that 1 KHz square wave to, among other things, provide a sanity check that the unit was set up properly. These scopes offer so much analysis capability that a similar sanity check is just as essential.

Long ago I worked with an engineer who had applied for a job at Cape Canaveral. The tour seemed to always come back to a panel of beautiful controls just crying out for some tactile interaction. A big bundle of wires coming trailing on the floor was cut, proving the box wasn't connected to anything, and my friend finally succumbed and twisted a knob. Klaxons suddenly blared all over the blockhouse! The panel was a test; management didn't want to hire someone who pressed buttons in a launch complex. He didn't get the job.

I was pretty sure the MSO-X 3054A wasn't connected to a firing circuit, so pressed, twisted and pushed to put the unit through its paces. Next surprise: all of the rotating knobs can be pushed in as well as twisted. Press a vertical or horizontal positioning control and the display returns to the zero position. A similar action on the trigger knob sets the trigger threshold to the signal's 50% level.

For generations vertical and horizontal controls operated in steps of 1, 2, and 5. The first position might be 1 usec/division, the next 2 usec/div, then 5, repeating at 10, 20 and 50. That's true on these scopes as well. But press the vertical or horizontal knobs and the devices switch to a "fine" mode where the detents select variable steps. For instance, at 100 usec/div this mode now clicks in 2 usec increments, letting one finely control how much of the waveform is shown on the screen. Another quick press and it returns to 1, 2, 5 steps.

I got these scopes before their official introduction and presumably the instruction manual had not been complete, as none was included in the package. These are complex products and I was initially disappointed, but the next surprise mitigated my frustration. Hold any button in for two seconds and the screen shows pretty complete help information for that control. And this includes the soft keys.

Triggering works as you'd expect - until you start pushing more buttons. It was no surprise that the unit can trigger off of a particular I2C or CAN packet; that's hardly unusual on an instrument with built-in protocol analyzer. But it's an essential feature for working with serial datastreams. The scope's ability to trigger on a variety of runt pulses, too, was expected and very welcome.

But Agilent has taken the integration of analog and digital channels a step further. Sure, you can trigger on data patterns from the 16 digital inputs, but even with that entire section of the instrument turned off the logic analyzer-like triggering features still exist. With just a few button presses one can start a sweep based on a rise/fall time being greater or less than a user-set value. Similarly, the units can trigger on specific pulse widths, programmable from 2 nsec to 10 seconds.

I once worked on a system that read data from a high-speed tape. What turned out to be a shifting bias problem caused the 900th byte in each frame to be read incorrectly. We saw the symptom but not the cause, and eventually rigged a scope to trigger on the start of a frame, coupled the trigger-out to a bit of logic, tied that to a lab counter, and used the 900th occurrence to trigger another scope. Was that the start of the gray hair or was it the teenagers? Very coolly, these scopes include a trigger mode that looks for the Nth edge in a burst, with N programmable from 1 to 65k. That feature alone would have saved me more than the cost of the fully-loaded high-end scope.

Metastability, the clocking of a latch too close in time to the data changing, is a perennial problem in the world of logic. Normally I'd set a scope to acquire and integrate many thousands of sweeps on the screen and try to pry these setup/hold violations from the increasingly blurry displayed mess. But these new Agilent scopes have a mode that triggers a sweep only when the system being tested has setup and hold times less than a user-entered value.

We trigger scopes because memory is finite; it's impossible to suck in minutes worth of data at 2 nsec/div. But even with triggering finite memory continuously frustrates us. Suppose a second processor becomes bus master once a second but sometimes there's a timing problem. Clearly you'd have to have the time base cranked up to a pretty high rate to see what's going on, but even a deep memory will fill after the first bus exchange. It's impossible to see multiple instances of the problem.

Time for another surprise. An option adds "segmented memory," which is the ability to carve storage into as many as 1000 chunks. Each triggering event can populate one of the segments, so you can capture many events that are widely-separated in time, even at a very fast sweep rate. That's a must-buy extra.

There's a lot more lurking under the hood of these new instruments. For instance, like all digital scopes these new offerings will automatically make a variety of measurements. Press a button and the unit shows the voltage, period and all of the usual parameters about a signal. Nice feature, though hardly novel and something we old-timers had been doing by squinting at the screen and using a grease marker even in the Dark Ages. But the Agilent team must have burned through too many Starbucks as they've included, well, just about everything measurable. 33 different types. Unusual items include the area under a waveform summed over any number of cycles, the time at which a waveform hits its max or min value, the width of a burst, and much more.

An optional waveform generator feeds a front-panel BNC with sine, square, ramp, pulse or noise signals, all, of course, programmable. Then there's the mask test function, which apparently (here the lack of a manual left me a bit at sea) watches an input to see if it falls out of some range. I'm told the test and measurement community uses this feature extensively.

The scopes have a nice built-in storage compartment for probes and the power cord. As is common with these sorts of devices it uses the IEC C13 power cord which does indeed fit in the compartment. Since the rated max consumption of the scopes is 100 watts (I measured 66 watts on the more powerful unit), I'd prefer a less bulky cord; even better, one that retracts into the unit as on some laundry irons.

One annoyance: after pressing the power button it takes a whopping 12 seconds before there is any indication the system is booting up. No lights, no beeps, no nuthin'. I'm told this has been chopped to 4 seconds in production versions. From hitting the switch to a useable state you'll wait 35 seconds, not far off the old vacuum tube boot times of yore.

Without a manual it took a bit of time to get comfortable with the various menus and modes. It's easy at first to get lost and forget how to access a function. But after the initial learning experience these units are a dream to use and very intuitive.

With a 70 MHz bandwidth the bottom of the line model is probably too slow for most of us in industry, unless you're working with PICs or other slowly-clocked devices. But I can see filling a university lab with them. The cost is less than that of the Macbooks that seem to grow on most campuses, and the scopes will have a far longer lifetime. The higher-end models meet the needs of anyone doing 8 and 16 bit work, and as long as 16 digital channels are adequate, will replace the logic analyzer, scope and waveform generator even in a lot of 32 bit applications.

Check out for more information. Bottom line: this is a breakthrough product line that makes a mixed-signal oscilloscope even more essential for embedded developers.

The scope is dead. Long live the MSO.