Complete Guide to Power-Rail Ripple and Noise Testing: From Principles to Practice

Why Do Your Test Results Always Differ from Someone Else's?

Hardware engineers have probably encountered this situation: the same board and the same power supply produce a measured ripple and noise level of 50 mV in one test, 80 mV in another, and 65 mV when a different oscilloscope is used. Which result is correct?

Power-rail ripple and noise testing may appear to be as simple as touching a probe to the test point and reading the value. In practice, however, it is one of the hardware measurements most susceptible to error. Different test methods, probes, and oscilloscope settings can cause the measured result to vary by several times.

The key issues are straightforward: what ripple and noise actually are, the advantages and limitations of six common measurement methods, and a repeatable test procedure. Understanding these points makes it easier to identify why measurements differ and establish a consistent basis for judging results.

Ripple Is Not the Same as Noise

When engineers use the phrase "power-supply ripple and noise,” they often combine two different phenomena. Their sources, frequency content, and measurement methods are not the same. Confusing them can lead to inconsistent or misleading results.

What Is Power-Supply Ripple?

Ripple is generated by the switching power supply itself. Its frequency is related to the switching frequency and is typically in the range of several hundred kilohertz to several megahertz. It is an inherent result of DC-DC converter operation: as the switching circuit charges and discharges the inductor, periodic voltage variation appears at the output.

Ripple generally has the following characteristics:

• It is relatively low in frequency.
• It has a regular, periodic waveform.
• It is synchronized with the converter switching frequency.
• It is primarily observed at the PMU output.

What Is Power-Rail Noise?

Noise has a more complex set of sources. Transient currents generated by high-speed chip I/O switching, parasitic inductance in the power-delivery network, and electromagnetic coupling between power and ground planes can all add high-frequency spikes to the supply rail.

Noise generally has the following characteristics:

• It contains higher-frequency components.
• It is less periodic and less predictable than switching ripple.
• It depends strongly on the operating state of the load.
• It is primarily observed at the power pins of the load device, such as a CPU or DDR memory device.

A practical distinction is that ripple is generated mainly by the power supply itself, whereas noise is added by the load and the PCB. Ripple should generally be evaluated at the PMU output, while noise should be evaluated at the power input of the device being supplied.

This distinction determines the measurement configuration:

• For ripple measurements, use a 20 MHz bandwidth limit and measure at the PMU output.
• For noise measurements, use the full available bandwidth and measure close to the power pins of the load device.

A common measurement error is to use the settings intended for ripple when measuring noise, or to use a full-bandwidth noise configuration when measuring ripple. The resulting values are not directly comparable.

Why Is Power-Rail Testing So Difficult?

Power-rail ripple and noise signals are often very small, typically tens of millivolts or even only a few millivolts. When the desired signal is this small, interference from the instrument, probe, ground connection, and measurement environment can significantly affect the result.

The six major factors affecting measurement accuracy are described below.

1. Oscilloscope channel noise: The oscilloscope itself contributes noise. In general, a wider bandwidth also allows more instrument noise to reach the measurement. If the actual signal is only 10 mV while the oscilloscope channel noise is 5 mV, the measurement accuracy will be limited.

   

   Oscilloscope channel noise with a P6139B passive probe and 200 MHz bandwidth

   

   Oscilloscope channel noise with a P6247 differential probe and 250 MHz bandwidth

   

   Oscilloscope channel noise with an SMA connection and 200 MHz bandwidth

2. Oscilloscope ADC resolution: An 8-bit oscilloscope has a much larger quantization step than a 12-bit oscilloscope. Small signals can be obscured by quantization levels and become difficult to measure accurately.
3. Vertical scale: A larger vertical scale produces a larger effective quantization error. For example, if a 1 V power rail is displayed at 200 mV/div, the quantization error may be approximately 7.8 mV. If the DC component is removed with the offset function and the vertical scale is reduced to 10 mV/div, the quantization error may fall to approximately 0.4 mV.
4. Probe bandwidth: Insufficient bandwidth prevents high-frequency noise from being measured. Excessive bandwidth, however, allows more instrument and environmental noise into the measurement.
5. Probe ground-loop area: This is one of the most frequently overlooked factors. The larger the loop formed by the signal connection and ground connection, the more easily it picks up radiated EMI. The measured noise level can therefore be substantially higher than the actual noise on the rail.
6. Oscilloscope channel configuration: Coupling mode, termination impedance, bandwidth limiting, and other channel settings all influence the displayed waveform and measured value.

Because these factors accumulate, it is not surprising for the same board to produce significantly different results under different test configurations.

Six Common Measurement Methods

Method 1: Passive Probe with DC Coupling

This is the most common configuration: a passive probe, DC coupling, 1 MΩ termination, and the oscilloscope offset function used to remove the DC component from the display.

Advantages: The method is simple, requires no additional hardware, and can measure signal content from DC through the available probe bandwidth.

Limitations: Passive probes have limited bandwidth, typically several hundred megahertz. More importantly, the alligator-clip ground lead creates a large loop area that can pick up radiated noise. The measured value is therefore often higher than the actual power-rail noise.

Improvement: Do not use the long alligator-clip ground lead for a small-signal measurement. Instead, wrap a short ground spring around the probe tip or solder a short enamel-coated wire directly to the probe as the ground connection. The goal is to minimize the loop area. A 1:1 probe is also preferable where the voltage range allows it. A 10:1 probe attenuates the signal by a factor of ten, reducing the signal-to-noise ratio accordingly.

Method 2: Passive Probe with AC Coupling

This configuration is similar to the DC-coupled passive-probe method, except that the oscilloscope uses its internal AC-coupling circuit to block the DC component. The offset does not need to be adjusted manually.

Advantages: The setup is easier because the DC level does not need to be positioned with the offset control.

Limitations: AC coupling introduces a low-frequency cutoff, typically around 10 Hz. Very-low-frequency components are therefore attenuated. For switching-converter ripple, this effect is usually negligible, but it should still be considered when comparing results from different configurations.

Method 3: Coaxial Cable with an External DC-Blocking Capacitor and 50 Ω Termination

This is one of the most accurate methods for power-rail ripple and noise measurement. A coaxial cable replaces the probe, and the oscilloscope is set to 50 Ω termination to match the cable impedance. An external DC-blocking capacitor removes the DC component.

Advantages:

• The impedance is controlled, minimizing signal reflection.
• The coaxial cable provides effective shielding against radiated interference.
• The available bandwidth is high enough to capture high-frequency noise.
• The measurement configuration generally provides the lowest noise floor among the methods discussed here.

Limitations: A small DC-blocking board or adapter is required, so the method is less convenient than attaching a standard probe. The blocking capacitor also introduces a low-frequency cutoff. With a 10 μF capacitor, signals above approximately 2 kHz can be measured accurately according to the stated test configuration.

Method 4: Coaxial Cable with AC Coupling and 1 MΩ Termination

When the low-frequency attenuation associated with a DC-coupled 50 Ω configuration is too large, the coaxial cable can be connected to a 1 MΩ input with AC coupling.

Advantages: The low-frequency response is improved, making the method suitable for measuring low-frequency ripple.

Limitations: The 1 MΩ input does not match the impedance of the coaxial cable, so signal reflections occur. Because the cable is relatively short, the reflection effect may be limited, and the result can still be used as a reference. However, the method is less suitable when high measurement accuracy is required.

Improvement: A 50 Ω termination resistor can be connected at the coaxial-cable input. This preserves the low-frequency advantage of AC coupling while improving impedance matching.

Method 5: Differential Probe with an External Capacitor and DC Coupling

A differential probe enables floating measurements. It isolates the oscilloscope ground from the board ground and can reduce interference caused by ground-loop currents.

Advantages: Differential probes provide strong common-mode noise rejection and are useful when the ground environment is electrically noisy or when a short single-ended ground connection cannot be established.

Limitations: Differential probes are relatively expensive and normally have a higher noise floor than a coaxial connection. An external DC-blocking capacitor is also required in this configuration, so the low-frequency cutoff must be considered.

Method 6: Differential Probe Attenuation with DC Coupling

Setting the differential probe to 10× attenuation expands the available offset range by a factor of ten. This allows a higher-voltage power rail to be measured directly with DC coupling, eliminating the external blocking capacitor.

Advantages: No external capacitor is needed, the low-frequency response is preserved, and the setup is convenient.

Limitations: The 10× attenuation reduces the signal-to-noise ratio and increases the effective contribution of the probe noise.

Measurement Comparison: Which Method Is Most Reliable?

The following table compares six methods when measuring four power rails on a hardware platform. The values are specific to this test case and should not be treated as universal limits for other boards or power systems.

Several patterns are visible in these results:

1. Passive-probe measurements are generally higher. The large ground-loop area allows more radiated noise to couple into the measurement. A power rail that appears to exceed a limit with a passive probe may produce a lower result when measured with a better-controlled method.
2. The coaxial connection with a DC-blocking board and the AC-coupled differential probe produce similar results. These methods can serve as cross-checks for one another when the test conditions are consistent.
3. The AC-coupled coaxial configuration with a 1 MΩ input can show significant deviation. The impedance mismatch creates reflections, and the effect is particularly noticeable on DC-DC converter outputs. This configuration is therefore not recommended when measurement accuracy is the primary objective.
4. The relative noise floors follow the trend coaxial cable < differential probe < passive probe. The influence of the instrument noise floor is especially obvious when measuring small signals, such as the output of an LDO.

For the highest measurement accuracy, the preferred configuration is a coaxial cable with a DC-blocking capacitor and 50 Ω termination. When ease of use is more important, an AC-coupled differential probe is also a practical choice. A passive probe is suitable for rough checks, but its result should not be used as the sole basis for a final compliance decision.

Five Common Measurement Mistakes

Mistake 1: Measuring at the Wrong Location

When measuring power-rail noise, the test point should be as close as possible to the power pin of the load device. The most demanding point is typically the power pin farthest from the PMU. Measuring at the PMU output for convenience captures primarily the supply ripple and may not represent the noise actually seen by the chip.

Mistake 2: Failing to Lock the Frequency and Load

CPU, GPU, and DDR operating frequencies can change dynamically, and the supply voltage may change with them. If the operating frequency is not locked during the test, frequency transitions and transient voltage adjustments may be included in the measured peak-to-peak value, producing an unnecessarily high result.

A practical test condition is to lock the CPU, GPU, and DDR at their highest operating frequencies and run a demanding workload, such as a large application or game. This forces the power system to operate under a severe load condition.

Mistake 3: Using an Excessively Large Vertical Scale

When the vertical scale is too large, the waveform occupies only a small portion of the display. The effective quantization error then becomes significant, and the measured result may not be reliable.

As a practical rule, the noise waveform should occupy at least two-thirds of the oscilloscope display, or the vertical scale should already be set to its minimum usable value.

Mistake 4: Using the Wrong Bandwidth

Use a 20 MHz bandwidth limit for ripple measurements and the full available bandwidth for noise measurements. Although this rule is widely known, it is frequently overlooked during actual testing. If the bandwidth settings differ, the results cannot be compared directly.

Mistake 5: Measuring for Too Short a Time

Power-rail noise contains random components. Capturing only a few cycles may miss the worst-case event. A practical procedure is to measure for more than 1 minute, use an acquisition interval of at least 1 ms, and set the sampling rate to at least 500 MS/s. The peak-to-peak statistical value should be used for evaluation.

A Repeatable Power-Rail Noise Test Procedure

Preparation

• Oscilloscope bandwidth: Approximately 500 MHz to 1 GHz is suitable for the stated procedure. Excessive bandwidth increases the noise floor, while insufficient bandwidth may miss high-frequency noise.
• Preferred measurement method: Coaxial cable with a DC-blocking capacitor and 50 Ω termination. An AC-coupled differential probe is an alternative.
• Test point: The sink side, at the power pin farthest from the PMU and as close as possible to the device ball or package connection.

Oscilloscope Settings

• Coupling: DC coupling with an external DC-blocking capacitor, or AC coupling.
• Termination: 50 Ω for a coaxial connection, or 1 MΩ for a probe input.
• Bandwidth: Full bandwidth for noise measurements and a 20 MHz limit for ripple measurements.
• Vertical scale: Adjust the scale so that the noise waveform occupies at least two-thirds of the display.
• Sampling rate: At least 500 MS/s.
• Time base: Approximately 1 ms/div, allowing a sufficient number of waveform cycles to be observed.

Operating Conditions

• Lock the CPU, GPU, and DDR frequencies at their highest operating frequencies.
• Run a demanding workload so that the power system operates under a heavy-load condition.
• Maintain the measurement for more than 1 minute.

Data Interpretation

• Use the peak-to-peak statistical value as the primary indicator, while also checking the maximum and minimum values.
• Be aware that peak-to-peak and maximum-minus-minimum values may not be identical because the maximum and minimum may not occur within the same acquisition cycle.
• Save the waveform file for later analysis and comparison.

Conclusion

The difficult part of power-rail ripple and noise testing is not operating the oscilloscope. It is understanding why each measurement configuration is used. The same numerical result may represent a final engineering conclusion in one test and only a rough reference in another. The difference lies in the measurement method and the control of the test conditions.

Probe connection, bandwidth, vertical scale, termination, coupling mode, test location, workload, and acquisition time may each appear to be minor details. Together, however, they can make the measured value differ by a factor of two or more.

A reliable measurement therefore requires more than simply touching a probe to the board. Distinguish ripple from noise, select a suitable connection method, minimize the ground-loop area, control the operating condition, and record the oscilloscope settings together with the result. With a repeatable procedure, measurements from different engineers and instruments become much easier to compare and interpret.