Introduction
This page gives measurement methods and limits to be generally applied to a LPC converter, regarding noise emitted by the converter (strong focus on conducted noise vs radiated noise). CERN applies specific gabarit in line with its core activity, and proposes pragmatic measurement methods.
Some fundamental measurements should always be performed using a high frequency current probe. These measurements can easily inform on the radiated emissions as well as the conducted ones. One concerns the input/output current cables, when the other one refer to the direct radiated noise emitted by the D.U.T. They are quick and don't request high-cost additionnal material (except the spectrum analyser).
Measuring the current leaving/entering a D.U.T can give better information than a standard dissymmetrical measurement, since, for example it can give access to the real common mode current. Moreover, measurement can be done up to hundreds of MHz and radiated emissions can roughly be deduced without a test with antenna and anechoid chambers.
A test using a very low-cost Moebius probe can give some interesting information on the radiated noise on a D.U.T, indicating up to the area which is the most polluted if the case. It is then a nice measurement for a designer which want to improve the radiated performances of his D.U.T.
Noise conducted on AC lines can affect any other user powered by the same mains network. This requirement is strongly defined in the standards an is a basic requirement for any system delivered to CERN. A computer is measured vs level proposed here
Recommended measurement methods vs frequency range
CERN gabarit is historically based on IEC-478-3 Curve C (replaced by IEC-61204-3), with extension down to 2kHz.
- [2kHz..10kHz] range is out of standard EMC equipement range (impedance and measurement capability) and should be treated following CERN recommandations.
- Quasi-Peak Measurement required, even if peak could be used (saving time) initially to point at possible issues to be later studied more carefully. A Peak measurement below limit confirms entirely the compliance with Quasi-Peak limit. The more noise signal "duty cycle" reduces (its frequency of apparition) the more Quasi-Peak Measurement result differs strongly from Peak one. Converter noise (by nature) typically generates less than 6dB of difference between Q-Peak and Peak.
Using
CERN level
CERN level
-
CERN level vs CISPR11
CERN level vs CISPR11
-
Frequency Domain measurement Test Set Up .vsd
CERN Input Side Gabarit .xls Predefined Plot versus standards .xls
Values for spectrum analyser are given below.
Resolution Bandwidth to be set on Spectrum Analyser vs frequency range
This measurement is well documented in standards, with some freedom (high power constraints) described below.
- LISN provides "dyssimmetrical" measurement, from ground to active measured lines, getting common and differential mode.
- LISN provides known & balanced impedance (50 Ohms line to ground f>fo) on all lines (3x for 3-phases) giving meas. reproducibility.
- LISN provides also reduction of existing noise on power network not to pollute the measurements.
- A 1500 Ohms EMC probe measurement is possible when LISN is not available
...
- Connect generally the ground part of the probe in less than a 30 cm area from the measurement point.
- Line Impedance is not guaranted anymore (being then given by network), jeopardizing measurement reproducibility.
- No decoupling exists between other clients on network and D.U.T. A measurement is mandatory with D.U.T OFF to identify the initial quality of the network.
- Sensitivity of the measurement is highly reduced (30..35dB attenuation from the probe, due to its natural high impedance required not to un-balance measured line).
- A 1500 Ohms EMC probe probe takes 1-2 hours to be built home-made
home-made 1500 Ohms 9khz-30Mhz probe
home-made 1500 Ohms 9khz-30Mhz probe schematics
home-made 1500 Ohms 2khz-100khz probe schematics (specific low frequency CERN measurement probe)
home-made 1500 Ohms 9khz-30Mhz probe attentuation measurement
- A frequency domain measurement can be done using modern oscilloscope
...
- Nyquist criteria is taken into account, with correct matching between sampling rate and max frequency in signal measured
→ fmax in signal of interest < 0.5 x (Samples / s) Hz (if not, aliasing to be expected)
- Sufficient number of points allows decent frequency range FFT measurement (signal frequency is "contained" in oscilloscope memory)
→ fmin measured safely > 2/ [ Nb.Samples / (Samples / s) ] Hz
- Adequate passive filters can be placed to limit anti aliasing phenomena, if frequency range dynamic is too high
- Adequate vertical sensitivity is used (some mV of noise are not correcly computed if a too large (1V/div) setting is used).
- AC coupling is used correclty when spying low frequencies
- Simplified Procedure:
- Choose the highest sampling rate + maximum memory size oscilloscope you own
Exemple: TDS3014 100MHz: 1.25GS/s + 10ksamples.memory
- Identify maximum frequency f.signal.max-freq in measured signal.
(this can easily be done sampling at highest sampling rate of the oscilloscope)
Exemple: 50kHz switching converter : f.signal.max-freq = 20MHz
- Caclulate the minimum safe sampling rate Ch.sampling.rate to be set on oscilloscope
→ Ch.sampling.rate > 2(min) x f.signal.max-freq samples/sec
→ Set it adjusting Channel horizontal time scale
→ Exemple: Ch.sampling.rate = 50 MS/s → Channel horiz. scale = 20µs
- Caclulate the minimum safe measurable frequency
→ f.signal.min-measurable-freq = Ch.sampling.rate / Nb.Acquisition.Points x 2
→ Exemple: f.signal.min-measurable-freq = 50 MS/s / 10kS x 2 = 10 kHz
- Adjust channel vertical sensitivity to optimum, keeping signal maximum in windown
- Make FFT Result analyze:
→ Safe frequencies measured range = [f.signal.min-measurable-freq .. Ch.sampling.rate]
→ Never touch Channel horizontal scale setting for zooming on FFT, use FFT one.
→ Exemple: Frequencies measurables in range = [10kHz..50MHz]. Measuring lower frequency would require or:
- a lower sampling rate with the addition of external additionnal sharp low-pass filter to avoid aliasing phenomena.
- better oscilloscope with higher nb of points memory type
- Typical Settings on TDS3014 Oscilloscope:
- 1500 Ohms 2k..100kHz CERN EMC probe with 100kHz low pass filter activated, on 50 Ohms selected Input
Time Scale = 400µs (2.5 MS/s)
Measurement Range = [2kHz..100kHz] (flat part of the pass-band)
Measurement Gain = 35dB (56.2x) in range [2kHz..100kHz] only
Minimum measurable signal amplitude = 110dBµV (0.3Vrms) on 500mV amplitude channel
Recommended & practical measurement methods vs frequency range
Using
LISN
LISN
-
1500 Ohms Probe
1500 Ohms Probe
-
Oscillo + Single-Ended Probe
Oscillo + Single-Ended Probe
Frequency Domain measurement Test Set Up .vsd
Output side (user side) is a key parameter for general CERN applications, since many often connected to magnet controled at the level of ppm with ultra precise electronics, potentially sensitive to noise.
CERN pay high attention to noise below 1MHz, especially in differential mode, even if randomly distributed and with very low ocurrence, or coming from specific conditions: 0V crossing in 4 quadrant power converters, a fan or contactor switched on... For this reason, time-domain measurement are required, in addition to frequency-domain measurement, to catch these unexpected signals.
Recommended & practical measurement methods vs frequency range
CERN gabarit is based on IEC-478-3 Curve C, extended for low frequencies (magnet current ripple constraints).
- [10Hz..10kHz] range is out of standard EMC equipement range (impedance and measurement capability) and should be treated with adequate material (bandwidth issue on 1500 Ohms probe for example, with a 35dB rejection for frequency higher than 10kHz usually).
- A time domain reference level Vpeak-peak measured with 100 MHz bandwidth capability should always be performed in parallel to this test.
- Reference gabarit is given (strong line) for a DC output voltage of less than 50V. Dashed line shows an hypothetic level for a higher range / purpose / topology application. This limit level is defined from final user requirements.
Measurement
Using
Peak
Peak
-
Quasi-Peak
Quasi-Peak
CERN Output Side Gabarit Peak-preferred (<50V dc output) .xls
Values for spectrum analyser are given below.
Resolution Bandwidth to be set on Spectrum Analyser vs frequency range
An oscilloscope running on a battery can measure differential mode noise very easily. Use of modern proble or a basic additionnal capacitor to cut the DC voltage can be used (1uF..100uF) in series with simple 50-ohms cable can be used.
- Some spectrum analyser with differential input exists, but FFT module available with modern oscilloscope fits well the frequency range.
- Avoid using differential probes, with common mode rejection ration not sufficient when reaching high frequencies.
- Better use a batterry powered oscilloscope than a insulation transformer on AC powered oscilloscope. (coupling capacitor)
- Always keep in mind that an ultra-basic measurement with a Fluke voltmeter (Series 79 or similar), used in AC.mode gives you interesting indication expressed in true rms value, in the range [45; 1 0000] Hz, regarding the noise being measured.
Output differential mode measurement methods .vsd
Recommended & practical measurement methods: frequency range vs Frequency / Time Domain
This "dyssimmetrical" measurement gets both common and differential mode (not a pure "common mode" meas.) and has to be performed on all outputs. (common mode often referred as asymetrical, when differential mode is symetrical noise).
- For practical reason, dyssimmetrical measurement on output side is done at the level of the standard DC connections (bolts), with ground reference connected locally to the busbar (30cm area). Electrical high frequency conductiviy of the rack is assumed compliant.
- Measurement has to be made on both polarities even if often identical (high frequency diff. capacitors usually being used).
- A 1500 Ohms EMC probe or a standard modern oscilloscope probe provide sufficient impedance and is not intrusive.
- A 1500 Ohms EMC probe measurement is possible when LISN is not available
...
- Connect generally the ground part of the probe in less than a 30 cm area from the measurement point.
- Line Impedance is not guaranted anymore (being then given by network), jeopardizing measurement reproducibility.
- No decoupling exists between other clients on network and D.U.T. A measurement is mandatory with D.U.T OFF to identify the initial quality of the network.
- Sensitivity of the measurement is highly reduced (30..35dB attenuation from the probe, due to its natural high impedance required not to un-balance measured line).
- A 1500 Ohms EMC probe probe takes 1-2 hours to be built home-made
home-made 1500 Ohms 9khz-30Mhz probe
home-made 1500 Ohms 9khz-30Mhz probe schematics
home-made 1500 Ohms 2khz-100khz probe schematics (specific low frequency CERN measurement probe)
home-made 1500 Ohms 9khz-30Mhz probe attentuation measurement
- A frequency domain measurement can be done using modern oscilloscope
...
- Nyquist criteria is taken into account, with correct matching between sampling rate and max frequency in signal measured
→ fmax in signal of interest < 0.5 x (Samples / s) Hz (if not, aliasing to be expected)
- Sufficient number of points allows decent frequency range FFT measurement (signal frequency is "contained" in oscilloscope memory)
→ fmin measured safely > 2/ [ Nb.Samples / (Samples / s) ] Hz
- Adequate passive filters can be placed to limit anti aliasing phenomena, if frequency range dynamic is too high
- Adequate vertical sensitivity is used (some mV of noise are not correcly computed if a too large (1V/div) setting is used).
- AC coupling is used correclty when spying low frequencies
- Simplified Procedure:
- Choose the highest sampling rate + maximum memory size oscilloscope you own
Exemple: TDS3014 100MHz: 1.25GS/s + 10ksamples.memory
- Identify maximum frequency f.signal.max-freq in measured signal.
(this can easily be done sampling at highest sampling rate of the oscilloscope)
Exemple: 50kHz switching converter : f.signal.max-freq = 20MHz
- Caclulate the minimum safe sampling rate Ch.sampling.rate to be set on oscilloscope
→ Ch.sampling.rate > 2(min) x f.signal.max-freq samples/sec
→ Set it adjusting Channel horizontal time scale
→ Exemple: Ch.sampling.rate = 50 MS/s → Channel horiz. scale = 20µs
- Caclulate the minimum safe measurable frequency
→ f.signal.min-measurable-freq = Ch.sampling.rate / Nb.Acquisition.Points x 2
→ Exemple: f.signal.min-measurable-freq = 50 MS/s / 10kS x 2 = 10 kHz
- Adjust channel vertical sensitivity to optimum, keeping signal maximum in windown
- Make FFT Result analyze:
→ Safe frequencies measured range = [f.signal.min-measurable-freq .. Ch.sampling.rate]
→ Never touch Channel horizontal scale setting for zooming on FFT, use FFT one.
→ Exemple: Frequencies measurables in range = [10kHz..50MHz]. Measuring lower frequency would require or:
- a lower sampling rate with the addition of external additionnal sharp low-pass filter to avoid aliasing phenomena.
- better oscilloscope with higher nb of points memory type
- Typical Settings on TDS3014 Oscilloscope:
- 1500 Ohms 2k..100kHz CERN EMC probe with 100kHz low pass filter activated, on 50 Ohms selected Input
Time Scale = 400µs (2.5 MS/s)
Measurement Range = [2kHz..100kHz] (flat part of the pass-band)
Measurement Gain = 35dB (56.2x) in range [2kHz..100kHz] only
Minimum measurable signal amplitude = 110dBµV (0.3Vrms) on 500mV amplitude channel
Using
Oscilloscope
Oscilloscope
-
Spectrum Analyser
Spectrum Analyser
Dyssimmetrical Measurement .vsd
Recommended & practical measurement methods: frequency range vs Frequency / Time Domain
A 600A-10V converter rack was installed on a copper ground, and DC output noise was measured on top of the rack (directly on busbars), and at the level of the load (warm magnets of 1.5 tons), 10 meters far away, connected with standard 240mm² DC cables (not shielded). An active earth system puts the DC outputs at a DC common mode of 10V, but level of noise of the converter is very low (differential and common mode).
- Measurements give same level whatever the location of the measurement on a low output noise converter.
Common Mode Measurement | Differential Mode Measurement
Common Mode Measurement | Differential Mode Measurement
Example of output noise measured on a LHC converter (LHC600A-10V).
- On 600A-40V converter rack on a copper ground, DC output common mode noise was measured (gnd to pol+).
- Measurements show the huge difference between frequency and time domain measurement of pulse-type noise (low duty cycle apparition),which appears to be very visible on time domain, as expected.
|