Conversion peak-hold to mean value?

[WIM32]

Is there a theoretically correct (or statistically) way to convert (afterwards) peak-hold values into mean (average) value?
I would think this is impossible, since data is being averaged and therefore reduced.
Many thanks
Wim

 

[nbucska]

Peak-hold DOESN'T average -- it holds the max.
It is possible to get the main with small error IF the P/H
holds the max. of sufficiently short intervals and
is updated often enough. <nbucska@pcperipherals.com>

 

[WIM32]

P/H in my case holds max over 20 blocks of FFT data of 1 second each. The approximate difference between mean-value and peak-hold (I think both are form of averaging, mean-value being: Sum(ni)/n and peak hold: Max(ni)/n), but never mind) is in this case a factor 3 to 5, varying over frequency.
Any research on this???

 

[electricpete]

From your definitions, I would agree with your original statement. If all you know is the maximum value over selected blocks (and you don't have any info about the nature of the vibration, then I don't think you have any hope of estimating the mean.

If you knew that the vibration was very periodic and repeatable (non-random), then I would assume that the rms could be computed as the peak over the crest factor. Probably not much help here.

 

[nbucska]

If your equipment holds 20 one sec. samples, you must
digitize them -- why can't you simply measure the main values ?

Can you tell me what you have, and what do you need ? <nbucska@pcperipherals.com>

 

[WIM32]

I already measure mean value, as well as peak hold value, I also save the original time-data. So I can find an empirical relation between peak-hold and mean value. What I need to know is whether or not there is a theoretical relationship between both values? My opinion, and from the responses I've had, I would there is no theoretical deduction possible... I just wanted to verify, to see if I did not overlook something.

 

[IRstuff]

That's not totally true nor not-true.

A standard rule-of-thumb with oscilloscopes is that the peak-to-peak of a noise signal is about 6 or 7 times the standard deviation. Obviously, peak in this context is not the same as a peak during an arbitrarily long period.

TTFN

 

[WIM32]

IRstuff,

this rule of thumb, can it be found in any textbook, or (semi)-official document?
thnx

 

[IRstuff]

Just standard statistics.

Assuming a normal distribution, +/-3 standard deviations around the mean encompasses over 98% of the distribution. So for those us too lazy or unable to do the statistics with recorded data, you can eyeball the oscilloscope trace and get a quick answer.

TTFN

 

[WIM32]

The peak-to-peak value of a signal is not the same as the peak-hold average value. What do you mean by peak-to-peak of a noise signal? Do you mean peak-to-peak=2*0-to-peak (amplitude)?

 

[MikeyP]

The only references to &quot;peak hold averaging&quot; that I have seen is in relation to spectra. ie. the peak hold averaged spectrum is one where the value at each frequency line is the highest value of that frequency component during the measurement period.

Is this what you have?

In that case, for a narrow band (FFT type) analysis on a statistically stationary signal, the value at each frequency can be considered as a sine wave. The mean will of course be zero and the rms value at each frequency line will be 1/sqrt(2) of the peak hold average.

M

 

[WIM32]

I have the following: 20 blocks of amplitude spectra. These blocks are averaged using to methods: mean-value and peak-hold value. So for example at 25 Hz I have values ranging from 1 to 100 (steps of 1), mean-value-average gives 50, peak-hold-value-average gives 100...
Therefore I am presuming there is no theoretical relationship..

 

[GregLocock]

There IS a theoretical relationship, but not one that will be much help

peak hold value>= average value

anything else requires knowledge of the signal.

Cheers

Greg Locock

 

[WIM32]

haha, thanks Greg...