S/N Determination using Empower Software

[danko]

PF 34(5) includes the following definition of signal-to-noise ratio:

S/N = 2h/hn, where h is "the height of the peak corresponding to the component concerned" and hn is "the difference between the largest and smallest noise values observed over a distance equal to at least five times the width at the half-height of the peak and, if possible, situated equally around the peak of interest."

I think it’s just a ridiculous modification of the genuine S/N (i.e. H/h).
What is the benefit from multiplying the ratio with 2 anyway – other than it produces a larger value?
Does it make my LOD any better? I don’t think so!

Best Regards

[Wendy38]

Hello,

Thank you to everyone who responded. We have figured out how to use the standardard deviation of the noise/slope *3.3 to get LOQ and have gotten some consistant results by setting the noise in Empower 2 to Average Peak to Peak. These seem to be our most promising avenues.

I wonder if anyone would know what the difference between S/N and EP S/N would be in Empower 2 and which one is the more accurate to use?

Wendy

[aceto_81]

I think it’s just a ridiculous modification of the genuine S/N (i.e. H/h).
What is the benefit from multiplying the ratio with 2 anyway – other than it produces a larger value?
Does it make my LOD any better? I don’t think so!

Best Regards

The ratio is intended to give a peak which is 2 times higher than your baseline noise. If you don't multiply by 2, you end up with peaks equal to your baseline, and you can't detect them (so you are below Limit of Detection).

There has been an article on this in LC/GC some time ago:
http://chromatographyonline.findpharma. ... rticle.pdf


Ace

[danko]

Hi Aceto

Please read the beginning of the thread and in fact the actual proposal in Pharmacopeial Forum.
The latter states: “The S/N is calculated as follows: S/N = 2h/hnâ€

[aceto_81]

[quote="danko"]Hi Aceto

Please read the beginning of the thread and in fact the actual proposal in Pharmacopeial Forum.
The latter states: “The S/N is calculated as follows: S/N = 2h/hnâ€

[M Azmi]

Dear Ace,
I'm also get single number which is 999.5547717. Need more explaination.

Tq.

M Azmi

[aceto_81]

Hi M Azmi,

Your peak is 999.5547717 times more than your noise, so your S/N ratio is surely above LOD/LOQ.

Maybe an explanation for the factor 2:
If you calculate the noise as the maximum in a certain range minus the minimum in the same range, you get the noise.
But actually this is the double of the real deviation of the baseline (above + below).
This can be an explanation for the 2, but it's just my thinking no references for this.

Ace

[danko]

Hi Ace,

I realize that the following:

Maybe an explanation for the factor 2:
If you calculate the noise as the maximum in a certain range minus the minimum in the same range, you get the noise.
But actually this is the double of the real deviation of the baseline (above + below).
This can be an explanation for the 2, but it's just my thinking no references for this.

is just a speculation and not any kind of dogmatic assumption, so please don’t regard my comments to your thoughts as a heated debate.

It’s important to keep in mind that the signal (peak height) is measured from the averaged noise (the designated zero baseline and not from the minimum) to the very highest point of the peak (including the positive part of the noise amplitude at the peak apex). So the signal is by no means “cheatedâ€

[danko]

Correction.

(the designated zero baseline and not from the minimum)

Make the last word maximum.

Best Regards

[aceto_81]

Hi Ace,

I realize that the following:

Maybe an explanation for the factor 2:
If you calculate the noise as the maximum in a certain range minus the minimum in the same range, you get the noise.
But actually this is the double of the real deviation of the baseline (above + below).
This can be an explanation for the 2, but it's just my thinking no references for this.

is just a speculation and not any kind of dogmatic assumption, so please don’t regard my comments to your thoughts as a heated debate.

It’s important to keep in mind that the signal (peak height) is measured from the averaged noise (the designated zero baseline and not from the minimum) to the very highest point of the peak (including the positive part of the noise amplitude at the peak apex). So the signal is by no means “cheatedâ€

[HW Mueller]

aceto, again: multiplying the S by 2 doesn´t distinguish the peak from noise any better than not doing this multiplication. The only discernible advantage is that the S/N will look twice as good as that of someone who doesn´t use this factor of 2.
(This is late, because I got the "forbidden" again).

[aceto_81]

Hello,

You're absolutely right about that multiplying isn't better than not doing this multiplication.
But it was the only reason I could come up with for the multiplication with a factor of 2, but as I said before, it was just my (poorly) thinking and not a found referenced statement.

Ace

[jskjei]

The factor of two is apparently being included in this calculation in order to harmonize USP with the European Pharmacopoeia (see 5.0, Volume 1, page 79), although the noise value in the EP calculation seems to be determined from a blank injection. This doesn't answer the question of the source of or reason for the factor of two, which ultimately would complicate the system suitability calculation for S/N in Empower, as well as LOD and LOQ definitions if the change is adopted by the USP.

Does anyone who is familiar with the EP have any information or insight into this?

[sdegrace]

Hello,

Thank you to everyone who responded. We have figured out how to use the standardard deviation of the noise/slope *3.3 to get LOQ and have gotten some consistant results by setting the noise in Empower 2 to Average Peak to Peak. These seem to be our most promising avenues.

I wonder if anyone would know what the difference between S/N and EP S/N would be in Empower 2 and which one is the more accurate to use?

Wendy

EP "cheats" by multiplying the peak height by two, making your LOQ and LOD calculated in this fashion look twice as good. I have no idea how this is justified. I'd say that if you know your client is 100% cool with EP then you could use EP s/n in your LOQ/LOD determinations because it is advantageous to do so, otherwise, if compatibility with USP is an issue then using the "normal" s/n noise is safer. Personally I don't like the EP approach because to me it makes absolutely no sense and I hope we don't all wind up harmonizing on it.

Stephen

[Slammy1]

It is possible to calc EP s/n in the truer sense using Empower (from the blank injection), we just didn't have much success in setting up the custom field. If you can request Waters technical document #TECN1852625 it has how to do it and gives better explanation of how the algorithm works.

EDIT: The only concern I have is from the relationship S/N = 1/RSD. It's not an exact expression for HPLC, but I do use it as justification of specs and results. I would not in any way advocate the USP calc, which is quite vague and open to interpretation (why RSD is the fundamental basis for accepting or rejecting in the validations we run here). The EP calc isn't really noise, but a range of noise as can be seen in the definition of h. I would be quite surprised if anyone has calculated noise for a USP product using 20*w(1/2) as their range (unless it's a VERY sharp peak). Usually, at least in data I've reviewed, people tend to ignore single excursions and attempt to focus on the bulk of the noise. In the last EP verification study I ran it was like 5 minutes of baseline to monitor. It would seem to overestimate true noise IMHO, but is it 2X? I have the report in front of me, and I had a s/n of 13 with a 13% RSD (s/b around 5% RSD). Based on the single point, it would seem that they really are doubling true noise, but it could have been a number of things (including carry over, which the Waters equipment is notorious for).