As supply voltages get lower it becomes more of a challenge to accurately measure power supply ripple, or “noise,” on the power bus. Complicating factors include faster-switching power converters, higher bandwidth requirements for today’s circuits, and more of a focus on overall power integrity. Today, we deal with dynamic loads with fast transients, increased crosstalk, and coupling and switching regulators with faster rise times. From an EMI point of view, the days when 100 mV of ripple was acceptable is no longer true, because this can couple common mode currents to I/O and power cables and cause radiated emissions.
It is important to look at each DC power rail to see if the power supplied is within the tolerance band of a target system or device. This includes the nominal DC value of the line, as well as any AC noise or coupling present. The AC noise in a power rail can be broken down further into broad-band noise, periodic events, and transient events.
All three of these noise sources impact the quality of power that reaches a device, and it is important to reduce these noise sources to the point that the target device can operate correctly. Before you can minimize these noise sources, you need to be able to see and measure them accurately. But power rail measurements present several unique measurement challenges, so there are several things to consider:
- Bandwidth requirements
- System noise and additive probe noise
- Tradeoffs of AC or DC input coupling
- Loading challenges for power rails
When considering many power delivery systems, measurement bandwidths of a few MHz might seem reasonable as most power converters are switching from 100s of kHz to 3 MHz at most. However, the rise times of these switched signals are just a few nanoseconds, or faster, and these fast edges tend to produce broadband EMI up into the low GHz. Add fast transients or other coupled digital switching and the power rail can easily contain harmonic energy and transients up to 1 GHz.
DC offset and DC/AC coupling
Most oscilloscopes have a limited DC offset of plus or minus just a few volts to get the displayed waveform into the center of the screen and this also depends on the vertical setting. The raw voltage powering a product could range from 12 to 48V, depending on the application. This could easily fall outside the DC offset range of most oscilloscopes. You might think AC coupling would be the answer, but then you sacrifice low-frequency bandwidth. One example of this would be missing measurements like voltage droop with transient loads.
Some microprocessors and power management ICs employ power saving features, such as dynamic frequency and voltage scaling, that change the DC supply voltage based upon the work load. These features are difficult to analyze with the instrument in AC coupled mode as the low frequency information is not shown by the instrument.
10X passive probes
High attenuation passive probes offer great dynamic range when looking at a wide variety of signals, but due to the attenuation, often introduce more measurement noise compared to low attenuation probes. This is because the signal is divided by the attenuation factor, driving it closer to the noise floor of the measurement system. Thus, the usual 10X passive probe will have a much higher signal-to-noise ratio (SNR) than a lower impedance probe. For example, a 10X passive probe might have only a 5:1 SNR, while a power rail probe would have a 40:1 SNR.
Noise performance of an instrument is proportional to the vertical sensitivity setting, with higher sensitivity ranges offering better noise performances than lower sensitivity ranges. Maximizing the displayed signal on screen will provide more resolution and a more accurate representation of the signal by the instrument. Lower vertical sensitivity ranges can often make signals appear to have more peak noise on them than they really do.
Power rail probes
Most companies who manufacture oscilloscopes have now added special power rail probes designed specifically to measure the low-impedance power rail voltages accurately (see the references section below for some examples). The ideal probe would offer a very high impedance at DC and transition to a 50 Ω impedance at higher AC frequencies (typically above 10 to 100 kHz).
To be able to accomplish this, the probe will have two paths; a low-frequency amplifier with DC offset control for DC and lower frequencies, and a separate capacitively-coupled path for higher frequencies to GHz (Figure 1). For example, the Tektronix TPP1000 (1 GHz BW) and TPR4000 (4 GHz BW) power rail probes both have a +/- 60V DC offset and can handle 42V peak measurements. The Tektronix probes start at 50 kΩ and transition to 50 Ω as the frequency increases. Other manufacturers, including Keysight and Teledyne LeCroy, have very similar probe specifications. Figures 2-4 show example connecting accessories.
Figure 1 This typical power rail block diagram shows the high-frequency and low-frequency paths.
Figure 2 Here’s a look at the Tektronix TPR4000 power rail probe with an assortment of connecting cables and a handheld probe.
Figure 3 Several connecting accessories come with the power rail probe. These allow you to either use a handheld probe or make various direct connections to the circuit board under test.
Figure 4 One clever connecting accessory is this “micro SMD” clip, which can be clipped across any decoupling capacitor to measure rail voltages.
To show the difference between regular 10X probes and power rail probes, as well as the importance of minimizing any connection inductance, I’ve made several comparison measurements in both the time- and frequency-domain. In each case, the ringing of one of the switched transitions will be displayed, along with the peak-to-peak ripple measurement and ringing frequency. I’ll also display the resulting frequency domain response showing the peaks at the resonant ring frequency. The frequency span will be zero to 1 GHz.
For these example measurements, I’ll be using the Tektronix Series 6 MSO (8 GHz BW), with the TPP1000 10X passive probe (1 GHz BW) and the TPR4000 power rail probe (4 GHz BW). I’ll use the built-in Spectrum View application to display the frequency domain.
The device under test will be a 1 MHz buck GaN converter with 1.2V output by Efficient Power Conversion (EPC 9101 demo board). The board will be loaded with a 10-Ω 2 watt resistor. We’ll be making the measurement on the board, right at the output.
Regular 10X passive probe
The first measurement will be made using a Tektronix TPP1000 (1 GHz BW, 10 MΩ) with the typical 6-inch ground lead (Figure 5).
Figure 5 This is the voltage ripple measurement setup using a common 10X passive probe. The 6-inch ground lead will add considerable inductance causing large apparent ripple and ringing.
Figure 6 Here are the resulting measurements with the passive probe with long leads.
Referring to the time domain wave (bottom) in Figure 6, we see the ringing is 550 mV p-p with ring frequency of 214 MHz. However, you can observe a relatively large second harmonic trying to get through (secondary ripples in the time domain wave). The frequency spectrum display (top) shows resonant peaks at the ring frequency and second harmonic of 214 and 433 MHz, respectively.
In the next measurement, we’ll use the same 10X passive probe, but use a short ground lead clipped to the probe ground (Figure 7). Note that while we can make this ripple and noise measurement, there’s still the issue if we wish to make this same measurement on a higher-voltage rail, such as 12 to 48V power rails.
Figure 7 This voltage ripple measurement setup uses a common 10X passive probe and a short ground lead.
Figure 8 The resulting measurements with the passive probe with short leads.
Referring to the time domain wave (bottom) in Figure 8, we see the ringing is now reduced to 142 mV p-p and the ring frequency is now 570 MHz. The frequency spectrum display (top) shows resonant peaks at the ring frequency.
This shows why it’s so important to reduce any connecting inductance to a minimum when measuring transient or switched waveforms. Any extra connecting inductance in the measurement probe will introduce very high resonant artifacts in the measurement, providing inaccurate ripple and noise measurements.
Power rail probe
Comparing the above measurements with the Tektronix TPR4000 power rail probe, we should see a considerable difference in the induced ripple and noise measurement. We’ll connect to the sample converter board in two ways; using the handheld probe with short probe pins (Figure 9) and with the solder-in coax cable accessory (Figure 11).
Figure 9 The voltage ripple measurement setup using the handheld power rail probe with short probe tips.
Figure 10 The resulting measurements with the power rail probe with short leads.
Referring to the time domain wave (bottom) in Figure 10, we see the ringing is now reduced even further to 70 mV p-p and the ring frequency is now 520 MHz. The frequency spectrum display (top) shows resonant peaks at the fundamental ring frequency of 520 MHz and the first sub-harmonic frequency of 222 MHz.
Figure 11 This voltage ripple measurement setup uses the solder-in coax cable accessory. It plugs onto the measurement cable using a MMCX connector.
Figure 12 Here are the resulting measurements using the solder-in coax accessory.
Referring to the time domain wave (bottom) in Figure 12, we see the ringing is about the same at 80 mV p-p and the ring frequency is now back to 220 MHz. The frequency spectrum display (top) shows resonant peaks at the fundamental ring frequency of 220 MHz and fourth harmonic frequency of 892 MHz.
This measurement of 70 to 80 mV p-p is still way high for good engineering design practice and could easily couple to product cables and cause a radiated emission failure. Typically, we’d like to see no more than 10 mV, with 1 mV ripple and noise as a good target to achieve.
You can see that when using the common 10X, 10 MΩ passive probes, even with short connecting leads, there is still large inductive resonance and ringing artifacts, which can mislead the designer. The low-frequency bandwidth is also severely limited when using AC coupling. These facts, plus the challenge in measuring voltage rails with levels well outside the ability of the oscilloscope to provide a compensating DC offset voltage, is a compelling reason for using the newer power rail probes.
—Kenneth Wyatt is president and principal consultant of Wyatt Technical Services.