0 Voturi pozitive0 Voturi negative

10 (de) vizualizări6 paginiThe paper shows possible problems with the current method for statistical evaluation of RFI measurements as well as the advantages and disadvantages of a new proposal in CISPR for an evaluation in frequency subranges instead of at fixed frequencies. Main item is the presentation of a new method to check the adherence to the 80%/80% rule, based on the use of a general margin to the limit. An application example is given: the use of the new method for appliances within the scope of CISPR 14-1.

Jun 16, 2016

© © All Rights Reserved

PDF, TXT sau citiți online pe Scribd

The paper shows possible problems with the current method for statistical evaluation of RFI measurements as well as the advantages and disadvantages of a new proposal in CISPR for an evaluation in frequency subranges instead of at fixed frequencies. Main item is the presentation of a new method to check the adherence to the 80%/80% rule, based on the use of a general margin to the limit. An application example is given: the use of the new method for appliances within the scope of CISPR 14-1.

© All Rights Reserved

10 (de) vizualizări

The paper shows possible problems with the current method for statistical evaluation of RFI measurements as well as the advantages and disadvantages of a new proposal in CISPR for an evaluation in frequency subranges instead of at fixed frequencies. Main item is the presentation of a new method to check the adherence to the 80%/80% rule, based on the use of a general margin to the limit. An application example is given: the use of the new method for appliances within the scope of CISPR 14-1.

© All Rights Reserved

- Neuromancer
- The E-Myth Revisited: Why Most Small Businesses Don't Work and
- How Not to Be Wrong: The Power of Mathematical Thinking
- Drive: The Surprising Truth About What Motivates Us
- Chaos: Making a New Science
- The Joy of x: A Guided Tour of Math, from One to Infinity
- How to Read a Person Like a Book
- Moonwalking with Einstein: The Art and Science of Remembering Everything
- The Wright Brothers
- The Other Einstein: A Novel
- The 6th Extinction
- The Housekeeper and the Professor: A Novel
- The Power of Discipline: 7 Ways it Can Change Your Life
- The 10X Rule: The Only Difference Between Success and Failure
- A Short History of Nearly Everything
- The Kiss Quotient: A Novel
- The End of Average: How We Succeed in a World That Values Sameness
- Made to Stick: Why Some Ideas Survive and Others Die
- Algorithms to Live By: The Computer Science of Human Decisions
- The Universe in a Nutshell

Sunteți pe pagina 1din 6

- 517 -

STATISTICAL EVALUATION OF RFI MEASUREMENTS

Frank Deter, Bauknecht Hausgeraete GmbH, Schorndorf, Germany

Lutz Dunker, Regulatory Authority for Telecommunications and Post, Berlin, Germany

Wilhelm Kleppmann, Hochschule fuer Technik und Wirtschaft, Aalen, Germany

Abstract:

The paper shows possible problems with the current method

for statistical evaluation of RFI measurements as well as the

advantages and disadvantages of a new proposal in CISPR for

an evaluation in frequency subranges instead of at fixed frequencies. Main item is the presentation of a new method to

check the adherence to the 80%/80% rule, based on the use of

a general margin to the limit. An application example is

given: the use of the new method for appliances within the

scope of CISPR 14-1.

1. Introduction

In many countries the CISPR standards are a legal requirements. The manufacturers use the rules in these standards for

the approval of new types of equipment. For mass produced

appliances the regulatory authorities base their decisions about

the banning of sales or the withdrawal of a type approval only

on a statistical evaluation of the RFI measurements. A wrong

decision may have a big economic impact. Therefore it is

important to consider the shortcomings of the current methods

and to think about possible improvements.

measurements, according to the current

CISPR standards

The CISPR standards require that on a statistical basis at least

80% of the mass produced appliances comply with the limits

with at least 80% confidence.

For this evaluation three methods are given [1]:

random from the production,

and

t-distribution.

Subject of this paper is the statistical assessment. Based on the

binomial distribution it can be shown that the required confidence level is reached, if seven products are tested and all of

them are good (below the limit) or if only one of 14 is bad. No

further assumptions are required for this conclusion.

However, the disadvantage of this procedure is the large sample size necessary for this test.

on k-values calculated using the non-central t-distribution may

be used. This procedure uses the measured data themselves,

rather than just the information good or bad. Therefore it

is more efficient and allows a test decision with a smaller

sample size and gives a steeper operation characteristic of the

test.

This test is passed, if

x + k s < L

(1)

where

x is the mean and

s the standard deviation of the data,

k is a constant factor derived from the non-central tdistribution (e.g. k = 1.52 for sample size n = 5).

Since the measurement procedure is prone to outliers, especially because the evaluation is done at fixed frequencies, the

assumption of normality is violated in some cases. Then this

test may lead to nonsense-results, see the following typical

examples. The values in both examples have been chosen at

will, but they are based on practical experience with lighting

equipment and household appliances.

Example 1:

The EUT produces broadband disturbances with a maximum

at a certain frequency, which varies from appliance to appliance. All appliances produce disturbances well under the

limit but the test is failed.

(see Table 1)

A resonance of the disturbance voltage in the frequency range

between 5 and 15 MHz as shown in example 1 is typical for

many appliances, e.g. motor driven appliances or switched

mode power supplies.

Example 2:

The EUT produces narrowband disturbances, the frequency of

which varies from appliance to appliance.

All appliances produce disturbances well over the limit, but

the test is passed due to a very low arithmetic mean on each

fixed frequency.

(see Table 2)

Values as shown in example 2 have been observed e.g. with

lighting devices, producing narrow band disturbances, the

frequency of which depends on the parameters of different

components.

- 518 -

Table 1: Values of the disturbance voltage and statistical evaluation of the results

based on the non-central t-distribution for example 1

Limit in the investigated frequency range: 60 dB(V)

Frequency Measured result with the quasipeak detector [dB(V)]

[MHz]

Appliance 1

Appliance 2

Appliance 3

Appliance 4

10,0

56

52

50

45

10,3

55

56

52

50

10,6

49

55

56

52

10,9

43

49

55

56

11,2

40

43

49

55

Statistical evaluation at 10 MHz:

48,6

Arithmetic mean

x=

Standard deviation:

s=

8,58

Factor for sample size 5:

k=

1,52

61,64

> 60

x +k s=

Appliance 5

38

45

50

52

56

Fail !

Table 2: Values of the disturbance voltage and statistical evaluation of the results

based on the non-central t-distribution for example 2

Limit in the investigated frequency range: 46 dB(V)

Frequency Measured result with the average detector [dB(V)]

[MHz]

Appliance 1

Appliance 2

Appliance 3

Appliance 4

1,000

50

0

0

0

1,015

0

50

0

0

1,030

0

0

50

0

1,045

0

0

0

50

1,060

0

0

0

0

Statistical evaluation at 1,0 MHz:

10

Arithmetic mean

x=

Standard deviation:

s=

22,36

Factor for sample size 5:

k=

1,52

43,99

< 46

x +k s=

in frequency subranges instead of

at fixed frequencies

CISPR recognized the above described problems and the

subcommittee A is working on a proposal to make the statistical evaluation with the maximum value in certain frequency

subranges instead of at fixed frequencies [2].

The problems described above can thus be overcome.

Appliance 5

0

0

0

0

50

Passed!

Problem 1:

If the frequency of a maximum fluctuates around the boundary between two frequency subranges (see figure 1), the test

may be failed because of a too high standard deviation in

both subranges, even though in fact no disturbances above

the limit are to be expected. The problem diminishes, if the

whole frequency range is divided into fewer subranges or no

subranges at all.

U

Limit

However, new problems arise. One of them is, that the RFI

limit in the investigated frequency subrange may be not constant. This can be solved easily by evaluating not the measured absolute values, but the distance to the limit. As basis

for the evaluation in the latest CISPR proposal [2] instead of

the maximum value the shortest distance to the limit in a

certain frequency subrange has been taken.

But the most important question is, how to define the number

of subranges in the investigated frequency range. Here we

face basically two problems:

fluctuates between two frequency subranges

96 O4

- 519 -

Problem 2:

If the amplitude of one maximum fluctuates, while another

maximum near the limit in the same frequency subrange is

relatively stable (see figure 2), an inappropriately low standard deviation can be calculated. The problem diminishes, if

the frequency range is divided into many subranges.

Therefore, it is suggested to be conservative and use an estimate for the maximum realistic standard deviation max,

based on previous experience with a particular type of product. As common in acceptance sampling and acceptance

control charts [3], using this conservative estimate max, the

following simple test procedure can be used:

1) a representative sample of size n < 7

(frequently n = 5) is chosen

(in dB), xmax is the highest (worst) of them

Limit

xmax + k E max < L

(kE from Table 3, L = Limit):

maximum is concealed by another more stable

maximum in the same frequency subrange

So the number of subranges has to be defined very carefully.

For many appliances all of the above described problems can

be solved with minimum effort by applying the following

new method of statistical evaluation.

80%/80% rule, based on the use of a general

margin to the limit

For a certain product group usually it is known, what standard deviation is realistic.

non-central t-distribution, it is based on the normal distribution, but under the assumption, that the standard deviation of

the population is known. Using max ensures that the results

are conservative (i.e. the EMI-test is passed only if the

80%/80%-rule is satisfied). However, it avoids that outliers

particularly far below the limit can blow up the standard

deviation s of the sample to unrealistic values by using the

realistic limit max. The test procedure based on the binomial

distribution (n = 7 and acceptance, if all items are good) is

equivalent to the procedure described here with kE = 0. For

comparison, table 3 gives 0.02, the slight difference being

due to approximations during the calculation of the values for

the test based on the binomial distribution. But again, the

new test method is on the safe side.

sample size

kE

1

2

3

4

5

6

7

1,68 0,97 0,63 0,41 0,24 0,12 0,02

method:

has to be below - this is the case, if the mean

of the population is at * = 0.8416

80% of its area has to be over the limit,

this corresponds to 80% confidence

distribution of the

population

-2

* = 0.8416

-1

(2)

based on the binomial distribution with a sample of 7 appliances, the acceptance limit AL is identical to the interference limit L

- 520 -

Interference limit L

distribution of the highest of 5 values the area over the limit L is less than 80%

distribution of the

population

-2

* = 0.8416

-1

The confidence in the evaluation result is less than 80%.

interference limit L

acceptance limit AL

for the highest value of 5

at 0.24 below L

this is exactly kE in table 3

distribution of the

population

-2

80% of its area is over the AL,

this corresponds to 80% confidence

* = 0.8416

-1

new test method with a sample of 5 appliances the acceptance limit AL is 0.24 below the interference limit L

the binomial distribution: If the highest value out of 7 is with

80% probability over the limit, a confidence of 80% is

reached. In figure 3 the population has a normal distribution,

but this works also with every other distribution.

Comparing figure 3 and figure 4, first it can be seen that the

distribution for the highest value in a sample of 5 has a similar shape as for the sample of 7, but the center is moved to

the left. The area above the interference limit L is less than

80%, therefore the confidence level of 80% is not reached, if

all appliances in a sample of 5 are under the normal limit.

96 O4

- 521 -

the normal distribution of the population is a necessary assumption. Exactly 80% of the area of the distribution of the

highest value of 5 is over the corrected limit AL. Therefore

the adherence to the 80%/80% rule is assured again, if all

appliances in a sample of 5 are under the corrected limit.

With a standard deviation of the population = 1,0, the limit

has to be corrected exactly by the factor kE, given in table 3,

e.g. 0.24 for a sample of 5.

If instead of = 1.0 a maximum realistic standard deviation

max is used, the factor kE has to be multiplied by max and

the same conditions of adherence to the 80%/80% rule are

given.

The advantage of this new method is that, since the acceptance is based on the largest value of the sample, the value

for kE is much smaller than for the method based on the noncentral t-distribution (starting from the sample mean). Therefore the method is not very sensitive to errors in max. It is

only a slight extension of the method based on the binomial

distribution to allow acceptance with a smaller sample size of

5, if after 5 it is already clear that the product is good.

value is determined and the statistical evaluation has to be

done on the basis of the non-central t-distribution, as defined

in the new proposal for a modification of CISPR 16-3. The

main difference is, that as basis for the evaluation not the

value itself can be taken, because the limit may be not constant in the frequency subrange. Instead the evaluation has to

be done with the distance to the limit. [2]

Step 3

If step 2 has been failed due to a high standard deviation and

not due to measured values over the limit, it has to be investigated, whether this high standard deviation is due to fluctuations of the frequency of a maximum between two

subranges. In this case the values of this maximum can be

evaluated again on the basis of the non-central t-distribution

in a new frequency subrange fmax +/- 20%.

Step 4

If step 3 has been failed, or at once if step 2 has been failed,

more appliances may be measured and the evaluation is performed on the basis of the larger sample.

6. Derivation of Table 3

5. Application example for the new method:

household appliances.

Based on samples of 5 appliances each for 25 different types,

measured by the German authority for market surveillance,

all of which were in the scope of CISPR 14-1, max = 7,5 dB

has been estimated as a conservative value of the expected

standard deviation. This was three times the average standard

deviation in selected subranges.

On the basis of max = 7,5 dB the following table can be calculated (see Table 4):

voltage for appliances within the scope of CISPR 14-1, based

on max = 7,5 dB

sample size

1

12,6

margin

to the limit

[dB]

2

7,3

3

4,7

4

3,1

5

1,8

6

0,9

7

0,15

size 7 has no practical meaning , because the application of

the binomial distribution leads to a margin of 0,0 dB. This

value just shows that the calculated margins to the limit are

on the safe side.

To keep the expenditure at a minimum, the statistical evaluation of RFI measurements for household appliances can be

carried out in a 4 step procedure, where the next step is necessary only in case the previous step has been failed.

Step 1

5 appliances are measured. If the maximum value of all appliances over the investigated frequency range is more than

1,8 dB under the limit, the test is passed. It can be assumed,

that this will be the case for the majority of appliances.

Step 2

If step 1 has been failed, the frequency range can be divided

into subranges, as defined by a product committee or as rec-

the derivation of the factors kE in Table 3 is given.

L = limit for RFI (compliance to be shown using the

80%/80%-requirement)

AL = acceptance limit

AL = L kE

(3)

(all data in the sample have to be below this value)

= standard deviation of the normal distribution of the population, use max for it

Using the formula for converting a normal distribution with

and into a standardized normal distribution, we get that

80% of the population are better than L,

if

L*

= u 0.8 = 0,8416

(4)

where:

= maximum acceptable mean of the normal distribution of

the population

U0,8 = 80% quantile of the standard normal distribution (80%

of the area for the population (=full line in figure 3, 4 and 5)

are below this value).

On the other hand, 80% confidence is achieved, if the probability that the highest of the values x1, x2, x3, xn in the

sample is above the AL is 80%. The dashed line in figure 3, 4

and 5 shows the distribution of the highest of 7 and 5 values

respectively, as examples. This conditions in figure 3 and 5

are equivalent to a probability of all values below AL of

20%, i.e.

P((x1 AL) and (x2 AL) and (x3 AL) and...(xn AL) ) = 0,2

(5)

Since the individual values are independent of each other and

from the same normal distribution, this is equivalent to

(P( x 1 AL ) )n = 0,2

(6)

P( x 1 AL ) = n 0,2

(7)

or

or

- 522 -

AL *

= un

0.2

(8)

AL = L u0,8 + un 0,2 = L kE

(9)

k E = u 0,8 u n 0,2

(10)

with

The factors kE are tabulated in Table 3.

7. Conclusion

The proposed new method for the statistical evaluation of

RFI results is based on the use of a general margin to the

limit, calculated on the basis of a known conservative value

of the standard deviation.

It is faster and has fewer shortcomings or problems than the

current methods. Because the new method is a conservative

one, it should not be used as a full replacement of the other

methods. Failing the new method does not mean fully failing

the statistical evaluation of RFI test results. The use of the

with the 80%/80% rule.

Therefore a step by step procedure for the statistical evaluation of RFI measurements is recommended.

8. Bibliography

[1] CISPR 16-3:2000 Specification for radio disturbance and

immunity measuring apparatus and methods - Part 3:

Reports and recommendations of CISPR

[2] CIS/A/337/CD Amendment to CISPR 16-3, Clause 2:

Statistics; new subclause 2.5: Rules for applying the statistical 80/80 rule and use of partial frequency ranges;

2001

[3] Wilrich, P-Th.: "Qualitaetsregelkarten bei vorgegebenen

Grenzwerten", Qualitaet und Zuverlaessigkeit 24 (1979)

S. 260-271, Muenchen-Wien: Carl Hanser Verlag

[4] Edward R. Heise, Robert E. Heise: "Uncertainty Rationale For Compliance Factors", IEEE International Symposium on Electromagnetic Compatibility 2000, Washington pp. 669-673

- Ee132b Hw2 SolÎncărcat deAhmed Hassan
- QCÎncărcat dePraneethGoverdhana
- Statistical Inference for EveryoneÎncărcat deOscar
- PSGARCHzÎncărcat deAmesh Parmekar
- ECON1203 Hw Solution week07Încărcat deBad Boy
- Lecture1_2013Încărcat deSteven Yoon
- rr220105-probability-and-statisticsÎncărcat deSrinivasa Rao G
- 9709_s10_qp_71Încărcat deroukaiya_peerkhan
- 30247877 DS en MS-TS Analytical LR Original 38829Încărcat deAchmad Huda Fauzi Adzima
- West Lake Groundwater Monitoring Report July 2013 12-1-13Încărcat deEPA Region 7 (Midwest)
- temporal stability of soil.pdfÎncărcat deTakashi Hara
- 1. Sample MeanÎncărcat dedeneke
- Bland-Altman_Plot_and_Analysis.pdfÎncărcat deMade Helen Virginia
- 12-confint (3)Încărcat deNovica Petkovic
- cps330cÎncărcat deIrwandi Irwandi
- 2003 Mathematical Methods (CAS) Exam Assessment Report Exam 2.pdfÎncărcat depinkangel2868_142411
- Chapter 10Încărcat deSudipta Ghosh
- Seminar of Probabilistic and StatisticsÎncărcat dehajriyanti yatmar
- POM Lecture (27)Încărcat demuneerpp
- Reliability of Jack-up Platforms Against OverturningÎncărcat deAbdulrahman
- Econometric sÎncărcat devarunragav85
- Bueno, M. - Geometallurgy applied in comminution to minimize design risks.pdfÎncărcat deMarcos Bueno
- Estimation Guideline- GeneralÎncărcat deaflgunaseilan836
- Garch_ApplFinEcon2Încărcat dePalwasha Malik
- EM561 Lecture Notes - Part 3 of 3[1]Încărcat deAbel Carr
- 267276414 STA301 Final Term Solved Subjective With Reference by MoaazÎncărcat deRaza Malick
- {u^N-en6-Nha}chaptÎncărcat deFunda Badem
- MAT 102Încărcat deApam Benjamin
- 22.bech09.22Încărcat deSaher
- Lect 07Încărcat deVivon Grendel

- Cryogenics&ApplicationsÎncărcat deM.Saravana Kumar..M.E
- A a 12148Încărcat deilianadd
- Welding DetailsÎncărcat dearup11nv
- MineStar Health Connectivity MatrixÎncărcat deX'mix Đreamer
- Stay Silv White Flux SpecÎncărcat deAnte Ninić
- Time Efficient VM Allocation using KD-Tree Approach in Cloud Server EnvironmentÎncărcat deEditor IJRITCC
- TSYEAR2 201213Încărcat deChen Inn Tan
- ISO CodesÎncărcat deJulian Bobb
- sony_hcd-lx7_8 v1.1~9-929-077-12Încărcat dejosetantonio
- Ppt on Draft Dcr 2034 for IcaiÎncărcat deSonia Dogra
- stream 180Încărcat deLuky Yana
- 263882604-EvoC-RNC-BSC-Workshop.pdfÎncărcat demebratuld
- 646 Tambasen v. People Coloquio.docxÎncărcat deAaron Thompson
- Stanford SR830Încărcat dedmcgibb
- WIP8Încărcat deakmalrahman
- Bolt Torque Chart _ Portland BoltÎncărcat dekNdash
- Interview YMDÎncărcat deMohammad Zafar
- ISTE Standards and ArtifactsÎncărcat deLauren Zizwarek
- Plaintiffs' Notice of AppealÎncărcat deBen Sheffner
- DB2 UDB and WebSphere V5 Performance Tuning and Operations GuideÎncărcat deBupBeChanh
- Bayblend T85 XF ISO EnÎncărcat deryan
- lady-gaga WorksheetÎncărcat deLaura Gómez
- NITOSEAL MS100Încărcat deBalasubramanian Ananth
- Air India vs Spice JetÎncărcat deGayathri Atiriya
- Bajaj 131224084925 Phpapp02 ConvertedÎncărcat deAnuj Jaiswal
- Aawe AmerikaÎncărcat deArda Agam Tamtomo
- Basra LightÎncărcat deDhanny Miharja
- Mechanisms of Drug ActionÎncărcat deMarky Bitonio
- UMS(Uniifiied Messagiing Serviice)Încărcat dePramodh Kj
- Human Development Report 2005Încărcat decakalil

## Mult mai mult decât documente.

Descoperiți tot ce are Scribd de oferit, inclusiv cărți și cărți audio de la editori majori.

Anulați oricând.