Subject: Re: Calibration
From: ptillma@gwdu19.gwdg.de (Peter Tillmann )
Date: 25 Nov 1996 09:04:18 GMT
gsettembre@vnet.ibm.com wrote:
:When calibrating (standardization) an instrument, for 2 orders of magnitude
:(ie: 0.1ppb - 10ppb), the following is observed: in order to obtain "accurate"
:recoveries from the low end and upper end of the calibration different types of
:regression is needed. In both regressions a good r2 value is obtained. Does this
:indicate that a nonlinear regression is needed? Any advice for low level
:calibration is appreciated.
Instrument response seems to be nonlinear to analyte concentration.
Three approaches are suitable:
1) local regressions: one for low concentrations, one for higher concentrations
2) data transformation of the response data to get a linear relationship
3) neural networks to allow for a nonlinear relationship between instrument
response and analyte
My choice is in order with the numbering of the alternatives.
1) is solely based on linear regression and the modell is clearly defined.
2) is also based on linear regression, with all its advantages, but data
transformation is kind of arbitrary.
3) is very popular and may help.
Nonlinear regression methods are in my opinion not useful, because of the as
far as I know unknown properties of the estimators, the computational
inefficiency, ...
(But my education in this subject is now 7-10 years outdated.)
Peter Tillmann
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Institute of Agronomy and Plant Breeding Peter Tillmann
Von-Siebold-Str. 8 Tel.: ++49-551-39 43 60
D-37075 Goettingen Fax: ++49-551-39 46 01
Germany
e-mail: ptillma@gwdg.de
fuer Freunde der Nahinfrarotspektroskopie: http://www.gwdg.de/~ptillma/
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Subject: Re: Inexpensive Graphing/Plotting Software?
From: ptillma@gwdu19.gwdg.de (Peter Tillmann )
Date: 25 Nov 1996 08:45:57 GMT
Forrest Weesner (fweesner) wrote:
: Can anyone suggest a DOS/windows package for scientific data
: presentation? I find typical spreadsheet programs very inadequate. I
: have used SigmaPlot and it's great, but I can't afford it right now.
Take a look at GNUPLOT, available on most FTP servers. It does ploting
of funktions and data in 2D and 3D. Output is available for many devices
as for postscript and TeX. It runs on most platforms (Unix, Windows, DOS, ...).
Peter Tillmann
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Institute of Agronomy and Plant Breeding Peter Tillmann
Von-Siebold-Str. 8 Tel.: ++49-551-39 43 60
D-37075 Goettingen Fax: ++49-551-39 46 01
Germany
e-mail: ptillma@gwdg.de
fuer Freunde der Nahinfrarotspektroskopie: http://www.gwdg.de/~ptillma/
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Subject: PC board Fiber Optics Spectrometer
From: Victor B. Loschenov
Date: Mon, 25 Nov 1996 11:49:19 +0300 (WSU)
THE NEW LOW COST PC-PLUG IN FIBER OPTIC SPECTROMETER
developed by Biospec (Russia) turns your PS into sophisticated
spectrometer system.
The systems includes fiber optics SMA input spectrograph
mounted on 12 bit ADC converter PC-plug in board, MS Windows
software for spectrometer control and data analysis, MS Windows DLL
drivers for custom software development, bifurcated fiber probes
for reflection, absorption and fluorescent measurements. The system
can be completed also with laser and light sources with SMA output.
The silicon CCD array with 2200 active elements is used as a
detector. The dynamic range of the detector is 2000. The NA of the
spectrometer is matched to that of silica fiber to increase
sensitivity and diminish stray light. The fibers at the detector
end of the probe are formed into the row (slit) to increase the
optical resolution and sensitivity. The cylinder lens installed
before detector and big detector element height allows one to
concentrate all light from the input slit into the detector what
increases the spectrometer sensitivity. The spectrometer has been
successfully applied in various biomedical and clinical
applications including fluorescent tumor diagnostics, blood oxygen
saturation evaluation, photodynamic therapy control. The special
fiber probe design allows one to make optical biopsy during
endoscopic observations.
Contact: Dr. A.A. Stratonnikov, Laser Biospectroscopy
Laboratory (Biospec), General Physics Institute, 38, Vavilov street
Moscow, 117942, Russia.
Fax , 7-095-3246363, 7-095-1350270, Tel 7-095-3246363, E mail:
biospec@glas.apc.org
General Specification
Optical parameters
Spectral range * 400 - 800 nm
Optical resolution (100 mm fiber & 600 gr/mm) 2 nm (FWHM)
Numerical aperture f/2.7
Collimating focal length 51.5 mm
Camera focal length 40 mm
Dispersion (600 gr/mm grating) (600 nm) 22 nm/mm
Stray Light 5 *10-4
Input slit height ** 1.8mm
Input fiber connector SMA-905
Detector & Electronics
Detector type CCD linear array
Number of active elements 2200
Detector element size 200 mm ' 8 mm
Exposition time 100 ms - 100 sec
Dynamic range*** 2000
ADC converter resolution 12 bit
Dimensions 25*10 cm
Power consumption 0.5 A at 5 V
* Spectral range may be enhanced from 250 nm to 1100 nm by using
another grating
** Spectrometer has the cylinder lens so that all light from the
input slit (input fibers arranged into the row) is focused into
detector increasing the sensitivity.
*** The Dynamic Range is defined as I(saturation)/I(noise,RMS)
Subject: Re: Calibration
From: Martin@nezumi.demon.co.uk (Martin Tom Brown)
Date: Mon, 25 Nov 96 15:30:46 GMT
In article <57auj7$1806@mdnews.btv.ibm.com> gsettembre@vnet.ibm.com writes:
> When calibrating (standardization) an instrument, for 2 orders of magnitude
> (ie: 0.1ppb - 10ppb), the following is observed: in order to obtain "accurate"
> recoveries from the low end and upper end of the calibration different types of
> regression is needed. In both regressions a good r2 value is obtained. Does this> indicate that a nonlinear regression is needed? Any advice for low level
> calibration is appreciated.
Without knowing which technique you are using it is hard to say.
There can be problems with the least squares fitting if you use
calibration points at log intervals like 0.1, 1, 10 if you post
example of your calibration points it might help decide.
ICP-MS can be linearly calibrated over 4 or 5 orders of magnitude
using a pulse counting detector (whereas OES or AES is much less).
Unfortunately, the regression coefficient is a poor indicator of fit.
Regards,
--
Martin Brown __ CIS: 71651,470
Scientific Software Consultancy /^,,)__/
