Analyzing the lack of repeatability is often considered only when measuring radiated emissions. In practice, however, repeatability problems also occur in the measurement of conducted emissions.¹ This article reproduces realistic situations that are known to hinder repeatability of conducted-emissions measurements, and that can lead to differences of >20 dB on subsequent tests of the same piece of equipment. Identifying the causes of this problem would enable industry to adopt solutions that improve the repeatability of measurements.

A series of measurements using an actual prototype was designed to illustrate that repeatability problems appear in practical situations and to show that these problems are not merely hypothetical, nor do they occur only in unusual circumstances.

The method used to obtain measurements of conducted emissions is described in EN 55022:1998, a standard normally applied to the measurement of electromagnetic emissions.² The setup adopted for each test is described. The instruments consisted of a line impedence stabilization network (LISN) Electro-Metrics ANS-25/2, a Tektronix EM-7600/TRL-30 transient limiter, a Tektronix EM-7600/TRL-30 preselector filter, a Tektronix 2712 spectrum analyzer, and in-house-developed software for the automatic acquisition of measurements. Each instrument was calibrated by its manufacturer's technical service. Furthermore, the laboratory had both vertical and horizontal ground reference planes as described in EN 55022:1998. The availability of both ground planes enabled determinations to be made of how changes in an installation could affect test results.

The device under test (DUT) was a panel thermometer based on a switched power supply. Such a device is very noisy (because of the switched power supply), but commonly used. The prototype enabled tests to be made using different configurations and situations to highlight problems commonly encountered. Although the DUT was an uncalibrated, arbitrary disturbance source, previous testing showed that it was stable enough to perform the repeatability analysis presented in this article.

The thermometer was fitted with a thermocouple as a temperature sensor. The manufacturer placed no limit on either the length of the thermocouple or on the power supply cable. This flexibility allowed researchers to ascertain how cable variation could modify measurement results.

The prototype could be placed either in a plastic housing (in which case it was not necessary to fit an earth connection) or in a box with user-accessible metal parts (which involved providing an earth for the unit). In the latter case, it was possible to demonstrate how the position of metal parts referred to the ground plane can affect the readings.

One might assume that a careful reading of EN 55022:1998 and compliance with its requisites would be sufficient to eliminate variations introduced by a test installation. However, even though the standard establishes a set method for taking measurements, the standard is primarily a general description. Some aspects are not fully defined, which leads to differing interpretations. It is important to be aware that it is a general standard designed for all products that do not conform to more-specific requirements.

The repeatability problem is exacerbated when performing precompliance tests if these tests are not carried out with extreme care. Precompliance tests are usually performed on prototypes at several design stages to approximate the unit's emission levels and incorporate any necessary corrections. Because such tests are aimed at obtaining fast, although approximate, results to compare different designs, they are often performed with the knowledge that the test conditions do not fully comply with the requirements of the standards. Under these circumstances, it is extremely critical to be aware of how omitting portions of the standard can influence measurement results.

In practice, the discrepancies between tests can be attributed either to variations in the test installation itself or variations directly related to the DUT, namely those caused by the position of the unit, the power supply, and other cables.

Most installations for the measurement of conducted electromagnetic interference (EMI) are based on a vertical reference plane and a horizontal one on a nonmetallic table of 80 cm in height. According to the standard's instructions for distances, this configuration enables tests to be performed on either desktop or floor-standing units. Nevertheless, the position and connections to the LISN are not precisely defined, which is the main source of variations caused by the installation itself .³

Some companies that perform their own testing consider an EMC testing facility as an auxiliary laboratory that is set up only when required, unlike the fixed installations for other laboratories that companies usually have. Under such temporary conditions, changes in the installation between one test and another make it practically impossible to provide the required consistent conditions to ensure repeatability.

To illustrate how this affects the position and earth connection of the LISN, the same reading was taken with two different layouts. In the first layout, the LISN was placed on the test table and connected to the vertical plane by a 1-m long cable with a 1 sq mm cross section. In the second layout, the LISN was placed on the floor securely connected to the horizontal plane. In both cases, the DUT was placed 40 cm from the vertical plane and with the thermocouple lead stretched taut and parallel to this plane.

The results obtained in the two experiments are shown in Figure 1. To compare those results easily, Figure 1 includes a plot with the envelope of the levels measured in both experiments. The significant differences in the high-frequency range are due to the different inductance of the earth connection of the LISN. Furthermore, changes in the position of the LISN referred to the earth plane, together with the capacitive coupling, are normally the cause of differences in the frequency range under 1 MHz.

Figure 1. Example of variations in the measurements due to changes in the test installation.

To avoid these problems, the LISN is often placed on the horizontal plane with its housing also connected to this plane. The inductance of this connection can be reduced by using a flat copper braid or metal brackets so that the electrical connection between the LISN and the plane has the largest possible surface area. Furthermore, brackets hold the LISN firmly in position and prevent any changes from one test to the next.

The main cause of DUT-related variations is the unit's distance from the vertical reference plane. This phenomenon is especially relevant for units with no earth connection, because the emission levels depend on the capacitance between the DUT and the earth plane. To resolve this, the standard clearly states that the unit must be placed 40 cm from the vertical plane. The following experiment demonstrates the effect of failing to comply with this condition. The DUT was placed as shown in the photograph in Figure 2, with the sensor lead parallel to the vertical plane. The distance from this vertical plane was then varied. Figure 3 shows the results obtained at distances of 20, 40, and 60 cm. A power supply cable of 0.8 m in length was used in each case. The graphs in Figure 3 depict how the distance between the DUT and the vertical plane mainly affects the low-frequency range (between 150 kHz and 1 MHz).

Figure 2. Layout used to show the influence of the distance of the DUT from the vertical plane.

 

Figure 3. Differences due to the distance of the DUT from the vertical plane.

The DUT's position also influences variation. For small units with a housing made of only one material (whether a conductor or not), the position of the DUT was not especially relevant. However, for larger units with housings that combine metallic surfaces with nonconducting surfaces, the alignment of the unit to the reference plane could cause discrepancies among tests.

To reproduce results for a unit with at least one metallic surface, the prototype was placed inside an aluminum dihedral as shown in Figure 4. This setup is similar to placing the unit in a housing where only the bottom and rear sides are metallic. These sides should be grounded; therefore, an earth connection was added to the aluminum dihedral.

Figure 4. Setup simulating a unit with two metallic sides (a). A slightly different setup also simulating a unit with two metallic sides (b).

This setup enabled easy comparison of readings when the metal side was closest to the vertical plane as in Figure 4a, and when the unit was reversed, with the metal side furthest from the vertical plane, as in Figure 4b. In both cases, the part of the DUT most enclosed was placed 40 cm from the plane. Figure 5 shows the results obtained in both cases. The variations in the frequency range under 1 MHz were caused by the difference in capacitive coupling in the two situations.

Figure 5. Differences due to the position of the DUT.

 

Figure 6. Different ways of folding a cable so that it is not inductive.

Variations: The Power Supply Cable

The DUT's power supply is a major source of variation, especially at high frequencies.⁴ The position of the power supply cable compared to the reference plane causes considerable deviations among tests. Random or careless cable distribution (for example, cables that hang down touching the reference plane) hinder repeatability. This effect can be corrected easily by placing the cables at the distances defined in the standard, or by documenting their exact position for later tests (by photographs, for example).

Nevertheless, even if the specifications in the standard are followed scrupulously, variations are still possible due to cable length and the way in which it is folded. The standard recommends using cables of 80 cm in length. If this length is exceeded, extra cable length should be eliminated by folding it (according to the standard) in a noninductive way. However, even when these instructions are followed, the cable length and the way it is folded can still affect repeatability because cable can be folded in several ways so that it is noninductive (see Figure 8). The graphs in Figures 7, 8, and 9 compare the results obtained with a cable of 0.8 m in length and results obtained with a 5-m cable folded in different ways. The main influence of the cable length manifests itself above 5 MHz and can even result in differences of >25 dB in certain cases.

Figure 7. Measurement with a power supply cable of 0.8 m in length.

 

Figure 8. Measurement with a power supply cable of 5 m in length folded so that it is inductive.

 

Figure 9. Measurement with a power supply cable of 5 m in length folded so that it is not inductive.

Variations: Other Cables

For units that include different cable types to connect various subsystems or to supply a signal from a sensor, the necessary cable length and layout may be unclear. The standard states that the length and layout should reproduce real operating conditions. Quite often this is impractical because it is impossible to reproduce real conditions or because the number of possible conditions is too large.

To illustrate the importance of cable position, the results obtained when placing the 1-m length of cable of the temperature sensor 40 cm from, and parallel to, the vertical plane were compared with those obtained with the cable wound into a 10-cm-diam spiral.

Although the dc signal traveling along the lead of the temperature sensor should not cause any emission problems, this lead could be the source of other unwanted signals. Its position in relation to the reference plane varies the coupling of the unit with this plane and, therefore, the measurements of the emission levels. Furthermore, coiling the cable changes its impedance to high-frequency signals, thereby changing the distribution of the unit's current flows. Figure 10 shows the readings obtained when the cable was stretched taut and when it was coiled.

Figure 10. Difference due to the position of the temperature sensor lead cable.

When exact cable position is not clearly defined by the standard, measurement variations can be reduced by defining the cable layout during the test. The requirements defined in the standard must be met, but adjustments should be made to the test bench to resolve unclear areas. When defining cable distribution, an attempt should be made to resolve an easily reproducible worst-case scenario, which would require clear and precise documentation of cable distribution.

Small changes in the configuration of conducted-emission tests can give rise to considerable variations in measurement readings, especially at higher frequencies. The results obtained from experiments described here show that these variations can easily occur in real, everyday situations. Errors introduced by the human factor, as well as the latitude provided by the standard, are two major causes of poor repeatability in these tests.

Human error can be minimized by carefully planning all tests, paying close attention to the geometrical requirements of the measurement setup as described in the standard. As for those areas that the standard does not clearly define, variations should be minimized by limiting cable length and layout, for example, or by specifying the position of the DUT compared to the ground planes. To ensure repeatability of the tests, a detailed description of such aspects should be included in test reports.