The Quest for Fast and Accurate Ripple Noise Test
by Mike O'Connor, Schaffner EMC, Inc.
Accurate ripple and noise testing can be elusive, but these suggestions and guidelines can help ease the frustration.
•A Real Test Problem
•Locating the Source
•Careful Fixturing
The ripple and noise component of a power supply's output waveform is almost certainly the most difficult parameter to verify in an automated, production-style test environment. Manufacturers require accuracy with speed, yet techniques to solve this problem can easily generate a poor or even inadequate measurement and are continually losing ground as switcher frequencies increase.
The consequences of excess ripple and noise vary from degraded performance-of analog or telecommunications circuitry-to outright failure by causing logic circuitry to falsely transition. Such faults usually are unpredictable and are very difficult to trace. This has led to the situation where accurate ripple and noise testing has become a key specification in ATE acquisitions as manufacturers push for higher and higher quality and reliability levels.
Ripple and noise consists of hum caused by the low frequency of the AC input; ripple related to the turn-on/turn-off cycle of the switching transistors; and noise caused by the fast rising or falling edge of the switching voltage reacting with parasitic inductances and capacitances in the circuitry of the power supply. The idealized waveform in Figure 1 illustrates these components.
In the design lab, a power supply invariably is characterized using a good scope. On the production floor, test engineers typically work to tighter equipment budgets and require faster test speeds. If they do consider using a scope to duplicate the lab measurement, the need for a sophisticated instrument with a differential probe to eliminate common-mode noise and/or the slowness of the measurement often rules out the instrument.
Two measurement techniques are commonly employed by ATE system integrators to test ripple and noise. The first is the true RMS voltmeter. This has the advantage of producing an unambiguous figure that can provide the ideal go, no-go result to the test department.
But, it suffers from limitations. When quoting or specifying RMS noise figures, the measurement bandwidth is rarely mentioned. Since the bandwidth of most RMS converters usually is only a few hundred kilohertz, this could significantly affect measurement. And the bandwidth of a RMS converter is rarely flat, typically causing additional variation in measurement at higher frequencies.
A crucial aspect is just what the measurement value actually represents. Because the switching spikes are quite narrow, it often is wrongly assumed that the RMS value is a measure of the ripple portion of the waveform. The measurement actually depends on the mean of the squared amplitude, and so is substantially affected by spikes regardless of how narrow they actually are.
Looking at Figure 1b, it can clearly be seen that a true RMS meter would yield a highly inaccurate result. When considered carefully, most ATE measurements using this technique are compromises.
An alternative approach is the use of a wideband peak-to-peak detector. This instrument responds to the overall peak-to-peak value of the signal, shown as "A" in Figure 1b. By adding a high pass filter, low-frequency hum may be eliminated and the amplitude B measured. The technique responds to the spike amplitude and coveys little information on the ripple component. This typically applies to advanced scopes/waveform analyzers capable of taking automatic peak-to-peak measurements.
Some users attempt to improve the measurement by using further filtering. Invariably, this is unsuccessful because it is not possible to discriminate between the ripple and the spike components since they have overlapping frequency spectra. Fourier analysis shows the noise has the same fundamental period as the ripple, imposing a barrier to any such approach.
The accuracy of these approaches is suffering as switching frequencies increase. The only way to get an accurate measurement is to perform the task in the time domain using, for example, a differential scope.
One technique, developed by Schaffner, provides the means to duplicate scope quality measurements in the production environment. It uses sample and hold methods to allow independent measurement of noise and ripple.
The waveform is first sampled from top to bottom. A sample at the top of the noise spike will have a very narrow width. As the sampling threshold is moved down the noise, the width of the sample increases gradually.
At the point where the threshold meets the ripple waveform, the sample width increases very quickly. This difference is sample width is used to discriminate between noise spikes and ripple. The decision sample width can be programmed to measure spikes as narrow as 10ns, a specification providing a very good margin for evaluating and testing state-of-the-art FET-based power supplies.
Figure 2 is a block diagram of the circuitry used to achieve this measurement. The input signal is compared with a variable threshold using a high-speed comparator.
The digitized output of the comparator is then fed to a pulse-width discrimination circuit that observes the mark-space ratio of the pulse train and indicates if the allowed spike width is exceeded. This achieves excellent correlation with scope techniques and allows fast automatic testing without manual intervention.
The waveform shown is idealized and does not mimic some practical waveforms that may have multiple narrow spikes. To accommodate this situation, the pulse discriminator does not, in fact, look at the width of the individual pulses but defines the maximum allowable mark space ratio from the comparator.
This peak-to-peak detector, coupled with true RMS measurement capability, allows the ATE to automatically determine the values of the underlying peak-to-peak ripple voltage and the RMS value of the waveform. Figure 3 shows a programming screen for this test, a system with a Windows 3.1 graphical user interface programmed by selecting and entering the required test conditions and limits in program boxes.
Contents
Accurately measuring ripple noise is only part of the problem. Once a fault is discovered, the ATE user needs to diagnose the cause, ideally down to the component level.
One technique yielding good diagnostics for power supplies is the fault directory database. Fortunately, power supplies usually are built using well-understood circuit ideas and functional blocks, and this greatly eases the problem of generating diagnostic routines to feed a fault database rework/repair system.
It is very easy for a test engineer to start a ripple noise fault directory using a number of simple core premises:
- Increased hum could indicate problems with the input filtering circuitry or the feedback loop.
- High ripple could indicate problems in the output filtering.
- Changes in the height or width of switching spikes indicate problems in the transistor switching times.
Using the latest premise as an example, the problem may well be associated with the transistor switching times or snubbing circuits. This suggests investigation of the waveform with a scope. To speed the rework loop in such cases, fault finding software allows the incorporation of instructions in the form of images derived from a video source or scanner input, or a PC graphics package
On detecting a switching spike failure within the ripple noise test, for example, the ATE could automatically generate a faultfinding step such as a guided-probe instruction. Using this premise, it might switch the user into a diagnostic routine that details the point at which to apply a scope probe and the expected waveform to look for on the scope.
The results of actions like this are used to create a probability table of causes of failures and likely repair solutions. If no fault message exists for a particular test failure, the operator is asked to generate a new entry with a repair code or diagnostic instruction.
Over time, a powerful heuristic algorithm is generated. This allows users to quickly characterize and trace the common failure modes of any supply type.
Contents
Adding ripple noise checks to a power supply test schedule may prompt rework of the fixturing process, since standard wiring practices can impact measurement accuracy and integrity. Here are some basic guidelines for ripple noise testing:
Good Practice
- Use 50 Ohm coaxial balanced channels with two wires/channel
- Use only balanced differential measurement
- Measure between the cores; tie off the shields at the power supply output to each other.
- Keep coaxial lines as short as possible; try to minimize unshielded wire at the UUT
- Keep high-frequency scanning as simple as possible.
- Physically separate UUT input and output wiring.
- Shield control lines (e.g., power relays) since they can act as antennas.
- Ensure that the ATE noise platform (e.g., on load lines) is very low; eliminate switching power supplies and processors if possible.
- Ensure that there is no oscillation on DC inputs to the UUT as well as UUT output power lines (decouple input lines if necessary).
- Use shielded cabling for DMM test points with all shields tied together.
What Not To Do
- Put capacitors on the output.
- Break the coaxial path from the UUT to the measurement unit with any non-coaxial connector or unmatched impedance unit
- Bundle power and signal lines in the harness.
- Put a switching power supply in the fixture near the UUT.
- Use a switch-mode or pulse-width modulated ac or dc source (AC sources carry 5% harmonic distortion)
Contents