Improving Measurement Confidence During Power Supply Test

Routine measurements of common power supply parameters, such as instantaneous switch element power, and line current harmonic content have been difficult to make accurately. Some parts of circuit operation, such as monitoring loop response in a pulse-width modulation (PWM) supply, have been virtually impossible to measure. Part of the problem results from the complex demands these measurements make on general-purpose oscilloscopes.

In addition, many of the initial measurements generate intermediate data that requires additional manipulation or calculation to arrive at the final result. The manual calculation process is tedious and prone to error.

The use of proper accessories expands the measurement capabilities of scopes to handle the specific needs of power supply test. The need to obtain accurate voltage and current waveforms is a necessity before any waveform analysis can occur.

Voltage Measurements

The ground lead on an oscilloscope probe connects to the chassis through the shell of the BNC connector. For safety and EMI containment, the chassis of the scope connects to earth ground through the third wire of the power cord.

This places a significant limitation in power supply characterization since many of the signals of interest are referenced to voltages other than ground. Power supply designers try several techniques to overcome this measurement limitation.

The most frequently used—and probably one of the worst—is floating the scope. To float the scope, the protective ground wire is cut in the power-line cable. This allows the chassis to float to the potential to which the probe ground lead is connected.

The most obvious danger is electrical shock. When a scope is floated to hazardous voltages, accidental contact with any metal components of the scope chassis can seriously injure or even kill the operator.

Because all of the input channels connect to the same ground reference, additional channels can only monitor waveforms referenced to the same floating potential. When floated, the scope cannot be triggered externally with ground-referenced control logic, a powerful tool for characterizing larger switch-mode supplies. External printers and computers connected through RS-232 or GPIB cannot be used because these ports will float along with the scope as well.

Most users are unaware of the waveform distortion that happens with floating the scope. This occurs when high slew rates appear on the ground lead, such as when measuring the gate drive of the upper device in a bridge. With the ground lead cut, components in the scope-line filter present a large capacitive load between the chassis and the input power leads, which appears as a virtual ground.

Users who employ this technique usually add an isolation transformer in the power line to reduce the effect. But even the best isolation transformers only lower the capacitance to about 150 pF.

The corruption is caused by the inductance of the ground lead acting on this capacitance. This forms a series resonate circuit that rings when stimulated with a fast rise-time signal, just like a probe with too large a ground lead will do when measuring a fast digital circuit. Only in this case, the ring is caused by the common-mode signal which the scope is floating on and should not be part of the measurement. Because the common-mode signal can be several times larger than the signal of interest, the resulting ring can totally obscure the details of the signal of interest.

Another technique used for measuring voltages not referenced to ground is quasi-differential or channel A minus channel B. Two channels and probes are used to measure one signal. Using waveform math, the reference channel is subtracted from the channel connected to the signal of interest.

The ground leads are not used in the measurement. This technique is safe since the scope remains grounded. However, it is limited to measurements where the differential mode (signal of interest) is approximately the same amplitude or larger than the common mode (signal being rejected).

For example, it can measure the V_CE of the upper insulated gate bipolar transistor in a bridge, but not the V_GE. Two conditions cause this limitation. For the math to work, both input signals must be acquired without clipping. The input attenuators must be set to 50 V/div. It is difficult to see much detail in the gate signal at 50 V/div. The more significant restriction is the limited common-mode rejection ratio (CMRR) caused by gain mismatches between the two input channels.

In some situations, probe isolators and battery-powered, hand-held scopes can be used for these measurements. They are safe for the operator. However, they still generate waveform distortion when measuring signals with fast slew-rate common- mode signals, such as the gate drive of the upper device in a bridge. This is because isolators are similar to a floating scope and lack balanced inputs.

The capacitance to ground is reduced, but not eliminated. A good test to see how much common-mode corruption is present is to connect both the reference lead and the probe tip to the common-mode source in the circuit, such as the upper device emitter.

The best solution for measuring voltages that are not referenced to ground is to use a differential amplifier. A true differential amplifier is similar to using channel A minus channel B with two important differences.

First, the gain, or attenuation, in both input paths, including the probes, is precisely matched from DC to very high frequencies. This gives very high CMRR, often 10,000:1 or more.

Secondly, the common-mode component is subtracted at the first stage before gain is added to the differential component. This gets around the common-mode range problem and allows you to measure small signals at high sensitivity, even when they are riding on common-mode signals hundreds of times larger.

Both input probes isolate the input capacitance from the DUT with the input attenuating resistors. This greatly reduces the effects of parasitic inductance in the input paths, reducing the likelihood of adding ring to the signal.

Measuring the dynamic saturation characteristics of a switching transistor is another challenge. In this situation, the culprit is amplifier overload recovery. To measure the turn on saturation dynamics, the scope must be set to a sensitivity of 1 V/div or less.

Before the device turns on, the collector to emitter (V_CE) or drain to source (V_DS) will be hundreds of volts. This voltage overloads the scope’s input amplifier. When the switching device turns on, the scope amplifier requires a finite amount of time to recover from the overload. Before it is fully recovered, a considerable offset will be superimposed in the waveform, altering the true dynamic saturation characteristics. The best laboratory oscilloscopes require about 800 ns to recover from an overload. Lower-cost oscilloscopes may require several microseconds. Moderate power switching transistors can reach steady-state saturation as fast as 500 ns, before the best oscilloscopes have reached usable accuracy.

Maintaining precise input matching for high CMRR along with high-speed clamping in these amplifiers requires complex circuitry, which drives up their cost. Lower-cost differential power probes do not contain the necessary clamp circuits and cannot make dynamic saturation measurements.

Measuring Current

Current can be measured with shunt resistors or current probes, but several drawbacks limit the usefulness of shunts. Adding the shunt requires cutting the current- carrying conductor. Shunts add a resistance to the circuit that can affect its operation. It is difficult to obtain accurate resistors with low resistance and inductance values necessary to measure large dynamic currents.

Current probes overcome these problems. Some models have jaws that can be opened to install around the conductor, eliminating the need to cut it. Current probes come in two basic types: AC and DC. The DC probes really are a hybrid and can measure DC to higher-frequency AC with relatively flat response.

Keep a couple of points in mind when using current probes. All current probes have an insertion impedance that creates some additional voltage drop and can reduce peak currents in the conductor on which they are clamped. The insertion impedance has resistive and inductive components, so its effects are more prevalent at higher frequencies.

The second concern is bandwidth. AC current probes have both a high-frequency and a low-frequency bandwidth specification. When using the probes to measure power, be aware of the phase shift that occurs as the input signal approaches the bandwidth specification.

At the -3 dB bandwidth, the phase shift is 45°. When multiplied with the voltage waveform, this much phase shift will result in a significant error. Many AC probes have low-frequency passband cutoffs of 40 Hz or higher, which eliminates their ability to measure power at line frequencies.

Avoiding Measurement Errors

The math capabilities in modern digital oscilloscopes can save tremendous time and effort. Both scalar measurements (those which produce a single numerical answer) and waveform math provide direct answers for measurements that, in the past, required considerable computation and analysis.

This functionality can cost time when erroneous results lead the user astray. To avoid this, keep track of both the measurement and analysis process. The cause of erroneous results can be grouped into three areas:

Errors in conditioning the input signal, such as clipping or bandwidth limiting.

Limitations in the acquisition process, such as sample rate, resolution, and record length.

Limitations in the computational algorithms.

Signal-conditioning errors are easy to miss when the input signal is not being displayed. For instance, when measuring power-line harmonics using the fast Fourier transform (FFT) function, it is common to turn off the display of the time-domain current waveform. If the amplitude of the input signal exceeds the full scale range of the scope input, the resulting distortion will be faithfully captured and computed by the scope, showing up as even-ordered harmonics in the display.

In addition to errors caused by amplitude inaccuracies of the input waveforms, dual waveform math results will be affected by skew in the relative time between the two input waveforms. Dual waveform math is computed point-per-point using the two input waveforms. Constant delay (skew), group delay distortion, and phase shift in one or both of the input waveforms can produce large errors in the computed waveform.

The results from acquisition errors can be more subtle. Nyquist limiting from an inadequate sample rate will generate an aliased waveform that often resembles the correct version in wave shape, but with the incorrect period. Other manifestations of inadequate sample rate are a filtering of the waveform that reduces slew rates and may decrease the peak amplitudes of narrow pulses.

Resolution limitations reduce the precision of scalar results. Most DSOs have 8-bit digitizers that can resolve to one part in 256 or about 0.4%. It is easy to be fooled into thinking you have more measurement resolution, when the scalar results are displayed with four or more decimal places.

Just as digitizer resolution limits amplitude resolution, measurements made with inadequate record length will lack timing resolution, resulting in greater error. The need for long record length is especially prevalent in power supply test due to the wide dynamic range of time. Viewing PWM control-loop response requires measuring each pulse width over a time window of several dozen milliseconds.

The last class of error sources involves the algorithms used by the scope for the calculation. Most oscilloscope manufacturers list the algorithms used in the built-in math functions in their manuals. A few minutes studying this section can save hours of frustration.

One common trap is the time window in which a scalar result is calculated. Usually, these measurements are calculated over the entire acquired record, which may include a lot of the waveform that is not visible on the graticule. If this is not desired, the use of a gating function that limits the window of calculation to the portion of the waveform contained within a pair of cursors will generate the desired results.

Switch Element Power

Dual waveform math can be used to calculate instantaneous power. This is a powerful tool for pinpointing losses when attempting to improve efficiency. The multiply function is used with a voltage waveform and a current waveform to generate the power waveform.

Some scopes automatically read the scale factor from the probes and display the proper result in watts. When using a scope that lacks this capability, be sure to account for additional scaling from external differential amplifiers and probes to come up with the correct watts/division factor (Figure 1).

The most common source of error in power waveforms results from the time skew between the voltage and current waveforms. The propagation delay through the current probe and the voltage probe plus differential amplifier are almost never equal.

During the dynamic portions of either waveform, the resulting power waveform will be incorrect. To eliminate this error, it is necessary to deskew the input signals. Some scopes have a deskew function that can be used to shift the time reference of one waveform relative to another.

Calibrate the deskew by monitoring the current and voltage of a fast rising step. Use a calibration loop or a fast signal from the circuit-under-test. Set the scope to 5 ns/div or faster. With both the voltage and current waveforms on screen, adjust the skew delay until the rising edges line up exactly.

The other error to be concerned with is phase shift in the probes or instrument. As the rise times of the input signals approach the rise time of the current probe or differential amplifier, the phase shift will create an amplitude error in the power waveform.

PWM Control-Loop Response

Recently, scopes have been introduced with a feature called jitter time analysis (JTA). This measurement feature is intended for analyzing the jitter in data communications systems.

If the scope has adequate record length, JTA can be used to measure closed-loop response in a PWM supply. The technique works with both small and large signal responses and does not require breaking the loop or coupling external stimulus into the circuit, commonly used techniques that can alter the operating point.

JTA generates a math waveform of time vs time (Figure 2). The horizontal axis is a time scale long enough to show the entire response time constant(s). This usually is several dozen milliseconds. The vertical axis can represent one of several timing parameters from the source waveform.

For PWM applications, the width function is selected. The input channel for the JTA function is driven from the gate signal of the switching transistor. The resulting waveform is gate pulse width (in microseconds) vs settling time in milliseconds.

A current probe can measure and display the load current and serve as the scope trigger source. With a step change in the load, the feedback loop response can be measured. The JTA function effectively demodulates the PWM gate drive signal to a linear plot that can be easily understood.

Line Current Harmonics

By using a high-quality current probe and a scope equipped with an FFT math function, conducted power-line harmonics can be measured. The FFT transforms the time-domain waveform into the frequency domain, generating a display that resembles that of a spectrum analyzer.

To measure the actual current at each harmonic, the display mode must be set to magnitude rather than rms. Properly set up, a scope with the adequate acquisition capability can accurately measure 20 or more harmonics of the fundamental line frequency (Figure 3).

The results you get the first time you try the FFT function may not instill a sense of measurement confidence. Changing the source waveform time/division setting by one click of the horizontal knob can greatly alter the appearance of the frequency domain waveform.

All of the key acquisition parameters directly affect the frequency-domain display. The sample rate determines the highest frequency component that can be measured and should be several times greater.

The digitizer resolution determines the dynamic range of amplitude that will look very small when set to a log scale. The resolution bandwidth in hertz is equal to the sample rate divided by the record length.

Greater accuracy is obtained by narrowing the resolution bandwidth to reduce the effect of random noise on the measurement. However, increasing the number of points measured also increases the computation time.

For measuring power-line harmonics, a resolution bandwidth of 2 Hz usually is optimum. It is very difficult to make accurate power-line harmonic measurements with record lengths less than 50,000 points.

These are just a few of the power supply measurements that can be made using the math features found in modern digital oscilloscopes. By keeping tabs of the accuracy of the raw input waveforms and understanding the limitations of the algorithms, you can use this capability with confidence.

About the Author

Steve Sekel is the engineering manager of the Power Measurement Business Unit at LeCroy. He has more than 20 years of experience developing electronic test and measurement instrumentation. Mr. Sekel has a B.S.E.E. degree and holds three U.S. patents. LeCroy, 8880 S.W. Nimbus Ave., Suite C, Beaverton, OR 97008, (503) 646-2410, e-mail: [email protected].

Copyright 1998 Nelson Publishing Inc.

June 1998