Logo of the Physikalisch-Technische Bundesanstalt
symbolic picture: "magazines"

New correction procedure for light source measurement

Especially interesting for
  • illuminant calibration
  • spectrometry

The distribution curves observed in the spectral measurements of light sources or heat radiators typically exhibit spectral distortions and a spectral resolution that is insufficient for accurate characterization. A mathematical procedure which was developed at PTB improves both these parameters.

Light source measurement (left) using a camera (right) Light source measurement (left) using a camera (right)

When measuring spectral distribution curves of optical and thermal radiation sources, one of the challenges resides in the treatment of spectral distortions in the distribution curve which are caused by the measuring instrument used and are difficult to prevent. In most applications, the distribution curve observed can be mathematically modelled as the convolution of the actual spectral distribution curve of the source with the line spread function of the measuring instrument used. Determining the spectral distribution curve of the investigated source from the measured data then requires a deconvolution. The deconvolution of measured data by means of a transfer function of the measuring instrument is frequently needed in image and signal processing, hence, numerous methods are described in the literature.

Simulated example of a spectral correction made with a standard method and with the new procedure.

PTB, together with the TechnoTeam Bildverarbeitung company, has made the so-called “Richardson-Lucy” method (RLM) – which originates in digital image processing – applicable to spectrometry. RLM is an iterative procedure. In each iteration, a new estimation of the distribution curve is obtained from the previous estimation. A challenge which has often been discussed in the literature when using RLM is the determination of the number of iterations. At PTB, a criterion for the selection of the number of iterations has been developed which does not require any parameters in order to determine the optimal number. For this purpose, the change in the estimation results in the course of the iterations is mathematically assessed in order to detect specifically that iteration from which the estimated distribution curve no longer improves and noise-related distortions start.


This new procedure was tested within the scope of comprehensive simulations and of investigations on measurement data. It turned out to be extremely robust with regard to the noise and to the resolution of the measured values of the distribution function. This makes this new procedure very well suited for practical application. Newly developed software simplifies its utilization even further.

This procedure can be applied in numerous fields of radiometry and photometry, for example, to improve measurements of both broadband spectral disitribution curves (e.g. of heat radiators) and of narrowband spectral distribution curves (e.g. of LEDs) by means of array spectrometers or monochromators.

Scientific publication

S. Eichstädt, F. Schmähling, G. Wübbeler, K. Anhalt, L. Bünger, U. Krüger, C. Elster: Comparison of the Richardson-Lucy method and a classical approach for spectrometer bandpass correction. Metrologia 50, 107–118 (2013)