The UIE Flickermeter Demystified

Figure 1 shows a simplified circuit diagram for a system in which time-varying currents flowing through the mains' system impedance cause varying mains voltage inputs to a lamp connected to the same circuit as the load drawing the varying currents. In this case the flicker source is shown connected to the same low-voltage branch circuit, but observable flicker may be generated by sources elsewhere in the system. Large industrial loads such as welders and arc furnaces connected at the medium-voltage (MV) level may, for example, generate flicker that is observed in low-voltage circuits connected to the same medium-voltage bus.

The scope for EN 61000-3-3 tests is limited to equipment that is connected at the 230 V and 400 V low-voltage level with single-phase circuits modeled as shown in Figure 1. Three-phase 400-V circuits are modeled in a similar manner, but the low-voltage system impedance is split between phase conductor impedances, which are set to 0.24 + j0.15 , and a neutral conductor impedance, which is set to 0.16 + j0.10 .

The voltmeter in Figure 1 represents the flickermeter measuring instrument. Flickermeters are described in IEC 868 (1986) and IEC 868 Amendment 1 (1990). IEC 61000-4-15 (1997) provides an updated description of the instrument and its implementation.

A thoughtful examination of the flickermeter specification reveals the instrument to be a specialized amplitude-modulation (AM) analyzer in which the carrier frequency is the mains frequency rather than the RF frequencies found in more conventional modulation analyzers. Other adaptations include specialized postdetection band-pass filtering to mimic the response characteristics of the lamp-eye-brain system and sensitivity to extremely low modulation levels.

The designation UIE in the instrument's name refers to the International Union for the Application of Electricity, formerly known as the International Union for Electroheat. The UIE is a nonprofit organization that provides a variety of services aimed at promoting effective and informed usage of electrical energy in industry and commerce. In the context of flickermeters, the UIE was pivotally involved in the development of specifications for the first internationally standardized flickermeter. Prior to this effort, various national instruments existed, but these produced somewhat different test results.

A brief review of amplitude modulation should help in understanding the UIE flickermeter. An amplitude-modulated sine wave may be defined using the following equation:

The equation describes amplitude modulation in terms of a parameter (shown as AM above) defined as the ratio of peak amplitudes of the carrier and the modulation envelope expressed as a percentage. This parameter is called the modulation index, and its use to describe amplitude modulation is standard practice in communications. The IEC standards define mains voltage variations as a percentage ratio between the voltage change and the nominal voltage, U/U (%), where both "U" and "U" are rms voltages. Since U is effectively a statement of the peak-to-peak amplitude of the modulation envelope, the relationship between this description and modulation index is exactly a factor of 2.

Figure 2. 50% sine wave modulation.

The relationship between U/U and modulation index can be hard to visualize, but an inspection of Figure 2 should help. In this example, we have a 230-V, 50-Hz carrier with 50%, 5-Hz sine wave modulation. Another look at the equation above reveals that 50% modulation will result in multiplication of the carrier by values ranging between 0.5 and 1.5. Recalling that the peak value for a 230-V rms sine wave is approximately 325 V, it can be seen in Figure 2 that the peak values for the modulated waveform are 487.5 V and 162.5 V. The peak-to-peak amplitude of the modulation envelope therefore is 325 V, while the peak amplitude is 162.5 V, thus 50% modulation.

The calculation is slightly different using IEC definitions. The rms voltages corresponding to peak values of 487.5 V and 162.5 V are approximately 345 V and 115 V. The difference between these rms levels is 230 V, which, expressed as U/U, becomes 230 V/230 V or 100%. Flicker modulations are generally more complex than the example given, but the mathematical principles are the same.

The last piece of the puzzle is recognition that actual flicker modulations occur as voltage drops across mains circuit source impedances. This means that both levels will typically be lower than the nominal mains level. Remembering again the equation for AM modulation, a restatement of U/U in terms of modulation index implies adjusting the stated carrier level to the midpoint between the peak excursions of the voltage change waveform. Considering the manner in which flicker is generated, U/U provides a more natural description than modulation index, but its use also tends to obscure the fact that the phenomenon is amplitude modulation.

Given in IEC 61000-4-5, tables 1 and 2 define test signals for confirmation of flickermeter performance to three decimal places (i.e., with a resolution of 0.001% for U/U), which implies measurement resolution in the instrument to 0.0005% AM or 5 ppm. For those familiar with more conventional modulation analyzers, this will be recognized as extremely fine resolution.

Figure 3 shows a block diagram for the complete flickermeter instrument as described in IEC 868 and IEC 61000-4-15. In the sections that follow each block will be described in some detail with respect both to the signal processing functions provided by the block and to the relationship to corresponding physiological phenomenon. Since the spectral response characteristics of some blocks provide the greatest insight into function, figures showing behavior of these blocks in the frequency domain are included.

Block 1—Input Voltage Adapter. The primary function of the input voltage adapter is to provide a normalized rms voltage to the input of Block 2. An automatic gain control (AGC) circuit with a 10 to 90% step-response characteristic of 1 minute provides the necessary functionality. From a frequency response perspective, the specified time constant corresponds to a first-order high-pass function with the corner frequency set to 0.00583 Hz. Higher frequency voltage fluctuations pass through the AGC unattenuated, but long-term trends are effectively removed.

From a physiological perspective, the AGC circuit mimics a well-known characteristic of human perception wherein moderate-level, constant stimuli to the senses gradually become imperceptible.

Block 1 also includes provisions for a calibration generator. This function is appropriate for ensuring calibration accuracy in the older-style analog instruments described in IEC 61000-4-15, but may be less needed in digital instruments, since it is easy to provide sufficient calibration stability to eliminate the need for on-line verification in such implementations.

Strictly speaking, the input transformer shown before Block 1 is not part of the block; however, its function is closely related, namely, to adapt the instrument input circuit to the nominal level of the measured signal. Once again, modern instruments may differ slightly by virtue of using variable-gain differential amplifiers rather than transformers.

The output of Block 1 is applied to the input of Block 2. It may also be applied to an optional rms voltage-measuring circuit for developing the voltage change characteristic time series used for evaluating the performance of the equipment under test (EUT) relative to the D_max, D_c, and D_t limits specified in
EN 61000-3-3. The required time series is an uninterrupted sequence of successive rms voltage measurements with half-cycle integration periods. The rms measuring circuits are typically included in UIE flickermeters, but are not used to evaluate flicker compliance.

Block 2—Demodulator. Block 2 specifies the use of a squaring multiplier as a demodulator. The purpose of this block is to recover modulating signals while simultaneously suppressing the mains frequency carrier signal. Operation of Block 2 is most easily understood in the frequency domain. As an example, Figure 4 shows the spectrum of a 50-Hz mains signal with 1%, 9-Hz sine wave amplitude modulation. The equivalent U/U level is 2%. For purposes of illustration, the 50-Hz carrier signal is shown normalized to 0 dB.

Figure 4. 50 Hz with 1% AM modulation at 9 Hz.

Figure 5. Demodulated signal from Figure 4.

Leaving aside normalization of the carrier amplitude, the signal shown in Figure 4 would be the input to Block 2, assuming an equivalent modulation of an otherwise unvarying mains voltage. The modulating signal is seen as a pair of sidebands offset 9 Hz above and below the carrier signal. The sideband signals are –6 dB relative to the –40 dB level of the modulating signal.

Applying the signal from Figure 4 as an input results in the output from the squaring demodulator shown in Figure 5. As would be expected from a nonlinear function, frequencies other than those at the input appear at the output of Block 2. The 50-Hz carrier is doubled in frequency and appears as a signal at 100 Hz. The modulating frequency is recovered and appears at 9 Hz and –40 dB (i.e., 1%). A significantly attenuated second harmonic of the modulating signal is also output, as is a dc signal, which is hidden against the left axis of the figure. Finally, upper and lower sidebands offset by the modulating signal frequency and the second harmonic of the modulating signal appear at about the 100-Hz doubled carrier frequency. The second harmonic signals are quite small and barely visible at the bottom of the figure.

The output of Block 2 is applied exclusively to the input of Block 3. The only desired output from Block 2 is the recovered 9-Hz modulating signal, but the work of removing unwanted frequencies is left to be performed by filters in Block 3.

Block 3—Weighting Filters. Block 3 includes three filters connected in series and a ranging circuit. One filter is a first-order high-pass having F_c set to 0.05 Hz. A sixth-order Butterworth low-pass with a corner frequency at 35 Hz is also specified. Despite the label for Block 3, neither filter provides weighting in a strict sense, but instead acts, respectively, to remove the dc component and the 100-Hz doubled carrier, with its associated sidebands, from the signal output by Block 2.

The third filter provides a band-pass response centered at 8.8 Hz. The band-pass filter provides a very specific weighting function within the frequency band of interest between 0.05 Hz and 35 Hz and acts to simulate a portion of the overall filament-eye-brain response for an average human observer. This response peaks at 8.8 Hz. The filter is very precisely specified by means of an equation documenting the required transfer function in the complex frequency domain.

Figure 6. Composite filter frequency response.

Figure 7. Filter output signal.

Figure 6 shows the composite frequency response of the three filters in series, while Figure 7 shows the output from Block 3 given the input shown in Figure 4.

The ranging function shown at the output of Block 3 is required for instruments using certain types of statistical classifiers in Block 5, but is often eliminated if nonlinear classifiers are used (this topic is discussed in more detail in the section on Block 5). In either case, the instrument must provide for measurements of instantaneous flicker sensation within a range of 0.01 to 6400 in units of perceptibility threshold. Full scale ranges corresponding to U/U levels of 0.5, 1, 2, 5, 10, and 20% are defined with a requirement for a minimum resolution of 1 part in 64 within each range.

Because the relationship between flicker perceptibility and U/U varies with frequency, there is no direct relationship between the two sets of ranges. Instead, the U/U ranges are specified to ensure that the desired dynamic range for flicker sensation is accommodated at any modulating frequency.

Block 4—Squaring Multiplier and First-Order Sliding Mean Filter. Block 4 provides functionality to implement the remainder of the filament-eye-brain model for flicker perception. The squaring operator simulates nonlinear eye-brain response characteristics while the first-order filter simulates perceptual storage effects in the brain. The first-order filter is somewhat imprecisely specified as a sliding mean filter having a time constant of 0.3 seconds, but should be implemented as a first-order low pass with a corner frequency of 0.53 Hz.

When the overall gain of the instrument is properly set, modulation levels corresponding to the mean human threshold for flicker sensation will produce values of 1 at the output of Block 4. Various additional processing steps may be performed on the output of Block 4 to assist in selection of instrument range or for research purposes. For EN 61000-3-3 tests, however, these outputs may be optional (assuming use of a nonlinear classifier in Block 5), since the outputs from Block 5 are used exclusively to determine flicker emissions compliance.

Block 5—Statistical Classifier. The statistical classifier in Block 5 models human irritability in the presence of flicker stimulation. Flicker is more tolerable if it occurs infrequently over short intervals. Tolerance decreases in the presence of increasing level intensity, event frequency, or event duration.

The input to Block 5 shows a sampling A/D converter followed by a statistical classifier. In modern instruments, the conversion from the analog to the digital domain typically takes place earlier in the signal processing chain. The classifier itself is dedicated to providing the statistical information required to calculate short-term flicker severity (P_st), as well as long-term flicker severity (P_lt).

The equations defining both parameters may be found in IEC 868 Amendment 1 or IEC 61000-4-15 and are as follows:

Short-term flicker severity is calculated using percentile values obtained from the statistical classifier. The statistical classifier implements a process for calculating percentile values for instantaneous flicker sensation on a "time at level" basis. This is done for integration periods of 1, 5, 10, or 15 minutes, with 10 minutes specified as the integration period for reference-grade compliance measurements. Assuming for a moment that samples are accumulated at a rate of 100 Hz and that the preferred 10-minute integration period is used, 60,000 individual samples of instantaneous flicker sensation will be acquired for each calculation of P_st.

Each sample is accumulated in the statistical classifier by incrementing a counter uniquely associated with one of a number of adjacent "bins." The counter to be incremented for a specific sample is selected by determining if the sample falls within the upper and lower instantaneous flicker sensation bounds specified for the bin. At the end of the integration period, the accumulated total count for all of the bin counters equals the total number of samples taken during the integration period (60,000 in our example). A set of percentiles may then be calculated using standard statistical methods. The overall process is analogous to percentile-based test scoring.

Depending upon the design of the classifier, the bins may be as few in number as 64, may be larger in number, and may be either linearly or logarithmically weighted. Use of a design with the minimum requirement for 64 linearly weighted bins also requires implementation of the ranging circuitry specified for Block 3. Modern implementations typically use logarithmic binning, usually with 1024 bins or more. This approach, assuming adequate dynamic range elsewhere in the signal chain, eliminates the need for the ranging function at the output of Block 3. Logarithmic binning also avoids certain accuracy problems that arise when applying linear interpolation to calculate percentile boundaries falling within bins.

The percentile notation used in the standards is slightly confusing, since the percentile subscripts correspond to percentages of samples for which levels are exceeded rather than to cumulative numbers of samples at lower levels. For example, P_0.1 corresponds to the level exceeded by 0.1% of the samples. This level is more conventionally referred to as the 99.9th percentile. Percentiles with an "s" subscript are smoothed values derived from weighted summations of nearby levels according to equations given in IEC 868 Amendment 1 and IEC 61000-4-15.

Long-term flicker severity is calculated from successive P_st values according to the equation shown on page 69. For reference-grade measurements, P_lt is calculated from a set of 12 consecutive P_st measurements acquired during a 2-hour test using 10-minute P_st integration periods (i.e., N = 12).

It should be noted that the units for P_st and P_lt are changed from "perceptibility" to "irritability" with P_st = 1 being the mean threshold for irritability for short-term flicker severity and 0.65 defined as the corresponding threshold for long-term flicker. Pass/fail limits for P_st and P_lt are set in EN 61000-3-3 at these irritability thresholds, and tests are conducted with an IEC 725 reference impedance placed between the mains source and the EUT, as shown in Figure 1.

Understanding flicker and the way the flickermeter works will benefit most, if not all, compliance and EMC engineers. Armed with this knowledge, they should be able to confidently test products for compliance with EN 61000-3-3 and to explain both the flicker phenomenon and the measuring instrument to their clients.