You Have Anything That Can Measure Ripple And Noise On Voltage Rails?

Being an oscilloscope probe design engineer I get the chance to get out of the cave several times a year and talk with our users so that I can better understand the measurements they want to make and what they need from us to make their lives easier.  In a typical conversation we would be discussing the next type of DDR memory or CPU (or whatever) that the users need to probe/measure and inevitably the question would come up—you got anything to measure ripple and noise on my power supplies? Initially the answer was no. These users wanted to measure mV ripple and noise riding on top of their 1.8 V, 3.3 V…24 V supplies. This is kind of a specialized measurement. To turn that answer into a yes I would need to design a specialized probe. But before I could do that I had to understand the application and measurement need better. Here is what I learned and what we came up with.

Thanks to Moore’s law, the doubling of gates on an IC every 18-24 months, the electronics that we encounter everyday are packed with more functionality in ever smaller, denser packaging. Consider your mobile phone. It wasn’t that long ago they were a brick that performed a single function—they made a phone call, and now they are elegantly thin machines that can give turn-by-turn directions, shoot high-definition video, monitor some of my biological functions and respond to my voice control. With this increased functional density comes some power related problems—power density and power supply induced jitter (PSIJ—power supply induced jitter can be the single biggest cause of clock and data jitter in a digital system). Folks learned that if they put tighter tolerances on the ripple and noise on their supplies and reduced the supply voltages where possible, they could reduce their power and jitter issues. It is not uncommon today to see supplies with tolerances of 1-3%. I saw an LPDDR with a 0.6 V supply with a 1% tolerance—this means measuring 6 mVpp ripple and noise.

Based on what our users shared with me I distilled the measurement challenges down to this: the need to measure a small AC signal riding on top of a large DC signal. If the AC signal exceeds the tolerance, the design failed to meet its requirements. This illuminated the challenges we needed to overcome. Users needed a very low noise probe & oscilloscope combo so that their ripple and noise were not overshadowed by the noise of the measurement system. They also needed a way to remove the large DC offset so they could put the signal in the center of the screen and zoom-in on it (get down to 1 mv/div if needed). The measurement tools also had to have enough bandwidth to capture the high frequency noise caused by the switching of the digital circuits. Since this is a function of clock speeds many users needed up to 2 GHz of bandwidth (frequencies above this are attenuated quickly by the circuit board fairly close to the noise source). And in addition, they had to have a probe that would not load the supply. For example, some users would connect a 50 Ω cable to the 50 Ω input of the scope to measure ripple and noise—for a 3.3 V supply the scope would sink 66 mA which can change the behavior of the supply.

Here is what we came up with, the Keysight N7020A Power Rail probe. The first and only probe designed specifically for making ripple and noise measurements on supplies. The probe has 1:1 attenuation ratio so that full size signals make it to the oscilloscope. This creates a very favorable signal:noise ratio. There is ±24 V of probe offset. This means the probe can remove up to 24 V of DC content from the signal so that signal can be placed in the center of the screen and be scrutinized at high sensitivity settings. Desperate users had been making use of DC blocks/AC coupling/DC reject to remove the DC content. They told us they disliked this because if filters the signal. A DC block is a ‘big’ capacitor so it also blocks low frequency supply drift and supply compression from being seen on the scope. Since the probe is active it has a DC input impedance of 50 kΩ which means it won’t change the behavior of the supply when it is attached. Finally the probe has 2 GHz bandwidth so that users can capture high frequency digitally induced noise on their supplies. Not everyone needs this much bandwidth so for those that don’t we recommend using the oscilloscope’s built-in bandwidth limiting capabilities so as to minimize oscilloscope/probe noise. And, I almost forgot, the probe comes with a lot of cool connection accessories so that you can easily probe a variety of locations on your target.

If you are curious to learn more, there is a great teardown video of the probe at The Signal Path: https://www.youtube.com/watch?v=d_Ybe6xnMIg

 

Splurge, buy an active probe, you’ll be glad you did.

Why, you may ask, would I want to spend money on an active probe for my oscilloscope when it came with free passive probes? As an oscilloscope probe designer I’m going to share with you a reason why you should consider upgrading to an active probe—and it’s not about bandwidth. I think a lot of folks who upgrade to an active probe do it because they need more bandwidth. Most passive probes top out at about 500 MHz so if you need more bandwidth than that you’ll need to buy an active probe. An active probe offers other benefits that should be considered even if you only feel you need a 100 MHz of so of bandwidth. I’m going to point out one that I think is most often overlooked.

Consider this, an active probe will provide significantly less probe loading than a passive probe. Probe loading is the effect that the probe has on your circuit when it comes in physical contact with it. Excessive probe loading will change the behavior of the signal being probed. With excessive probe loading, the signal that you see on the scope will be an accurate image of the signal being probed but it won’t be an accurate image of the real signal—the signal when the probe is removed. Here is how it works. When you look at the probe you are using, the label will say something like 10 MΩ:10 pf for a passive probe and 1 MΩ:1 pf for a general purpose active probe. What this is describing is a simplified circuit model for what the probe will look like to your circuit when the probe is connected to it. When in contact with your circuit the probe will appear as a resistor and capacitor, in parallel, connected to ground. It is easy to focus on the resistor value and overlook the contribution of the capacitance of the probe to probe loading. Considering only the resistor would lead one to conclude that a passive probe, with its 10 MΩ impedance will have much less loading than the active probe with its 1 MΩ impedance. Remember though that the impedance of a capacitor is inversely proportional to frequency. This means that the liability of the passive probe lies in its large input capacitance. Comparing the two probes, their input impedance (the impedance to ground when connected to your circuit) will be equal at 10 kHz. Therefore the active probe will produce less loading to any signal you are probing that has frequency content above 10 kHz. This is shown in figure above.

Now we will put this to the test. I’ve got a circuit that contains a 1.1 ns edge. Traditional guidance would suggest that I need about 300 MHz of bandwidth in my measurement system (scope and probe) to measure this signal. This is well within the capabilities of our free passive probe. I first probe the signal with my active probe and I measure the rise time. Just like I expected, 1.1ns. Now I remove my active probe and probe the signal with my passive probe. Oops, I’m measuring 1.5 ns. Is my measurement wrong? No, my measurement is correct. That is what the edge speed of my target signal has become due to the loading of my passive probe. The large capacitance to ground of the passive probe is creating a low impedance path to ground for the higher frequency content of my signal and my target cannot drive this load so the actual signal on my target is distorted.

What you can conclude is that a passive probe is good for making qualitative measurements and an active probe is good for making quantitative measurements. Qualitative measurements are things like: is the patient’s heart beating, is the 5V up, is the clock toggling..? Quantitative measurements are things like: what is the patient’s heart rate, how much ripple/noise is on the 5V supply, what is the rise time..? Do yourself a favor, next time you get a chance, splurge and buy an active probe.

Do yourself a favor, read this.

I hope to impart on you a bit of wisdom I have learned from my years of travel and talking with well over 1,000 oscilloscope users. If you do yourself the favor of reading through this you will have gained enough insight to not shoot yourself in the foot like so many of the scope users I have visited. They are not to be judged, they were doing the best they knew how at the time, and we all make mistakes or could do better—I know this to be true for me. Once I pointed out the mistake and the solution to these users they usually all had the same reaction—“oh, that makes sense”, followed by the classic palm slap to the forehead.

These users I speak of all made the same mistake. They spent valuable time and money selecting the best oscilloscope to buy or use for their measurement task. Then they connected a high quality probe to their scope. In some cases the probe was the nice passive probe that came with their scope, other times they had sprung for a snazzy active probe (smart move going for the active probe upgrade, more on this in another blog post). Then, and this is the crux of the matter, they put a bunch of long, dangly connection accessories onto the end of the probe. Maybe it was something innocent looking like a nice convenient long ground lead or one of those super helpful looking long red input wires that make it easy to connect the probe to a grabber that they could clip onto a part on their board.  In the end, the result was the same, the signal on screen “looked bad” or the device they were testing started to misbehave. This is usually when they grabbed me and said “Hey, you designed this probe. It’s not working right”.

The Weakest Link

What these users were experiencing was what I like to call the “weakest link” phenomenon. There are three links in the typical oscilloscope measurement chain—the scope, the probe and the physical connection to the target. You can have the best scope and probe that money can buy but if you put some crazy long wires on the end of the probe to make the connection to the target easier you have limited the performance of the measurement system to be equal to the performance of those crazy long wires. The connection accessories are the weakest link. They will limit the measurement bandwidth and they can excessively load your target.

Think of those long connection accessories as inductors that are being placed in series with the probe. If they are connected to the signal pin of the probe they are going to limit the bandwidth of the signal that can pass through to the probe because an inductor’s impedance increases proportional to frequency. Additionally, since there is an impedance mismatch between the long inductive connection accessory and the probe input, the signal traveling up the wire will create a reflection that will show up on the scope.  If that nice long ground wire is connected to the probe similar results will follow. The long inductive ground creates a higher impedance path for the ground return currents flowing on the shield of the cable. This will also limit the bandwidth of the probe.  Additionally, the impedance resulting from the inductance of the long ground wire can create a voltage potential between the ground on the target and the ground point at the tip of the probe resulting in measurement error and poor common mode rejection. If all that isn’t bad enough, those nice long connection accessories act as an antenna and can pick up noise from your surroundings and couple that noise into your measurement. Finally, there is loading. These long wires that are touching your circuit are now part of your circuit and their inductance and capacitance can change the way your circuit behaves. We call this probe loading.

Shorter is Better

At this point I can almost hear you saying “if those connection accessories are ‘bad’ why do you include them with your probes?” We include those accessories for convenience. The idea is that you use those accessories for qualitative measurements, things like “is the clock toggling”, is there “data on the bus”, “is the 5V up”. They make it easy to poke around your circuit quickly to check for functionality. If you want to make quantitative measurements like rise-time, over-shoot, noise levels, et cetera, then we intend for you to remove the convenience accessories and use the shortest connection possible. That’s it, that’s the punch-line, shorter is better.

Consider this example. I take my fancy 2 GHz active probe and I configure it three different ways, long wires connected to grabbers, long wires only and short input pin and ground contact. Notice how the bandwidth increases as the length of accessory in front of the probe gets shorter. By the way, we try to make it easy for you and we publish these bandwidth limitations in the product manuals.

Notice too how the probe loading (how the physical presence of the probe changes the way your circuit functions) decreases as the length of the connection accessory decreases. In this example the original circuit is producing a rising edge with a rise-time of 1.1ns (the green trace). Connecting the probe to the circuit using the long wires and grabbers loads the circuit and the rise-time changes to 1.7ns. When I remove the grabbers and just use the long wires the rise-time gets better, 1.5ns, though you can see the connection accessories are still affecting the circuit. Finally, I remove all the wires and go with the shortest connections for this probe and the circuit rise-time is back to its original 1.1ns.

I Hope This Was Helpful

Don’t feel bad if you’ve been making the mistake of using long connection accessories when making important measurements. You’re in good company, a lot of oscilloscope users have made this mistake and to be honest, I have too. Just remember, it’s ok to use those long, convenient connection accessories for a quick peek but if the signal looks strange or you are not getting the answer you expect, you’d do best take them off and go with the shortest connection possible. Shorter is better.

 

See all of the Keysight oscilloscope probes.