[ Log In | Sign Up | Lost password? ]

Cognitive filter for digital nuclear signals


Exchangeability of detectors

In radiation measurements, most instruments are tailored to a very individual detector. Even for one specific detector material, one experience quite notable differences in the measurement characteristics. An interesting question is therefore, whether we could find a digital filter, that automatically learns what type of detector is used.

Stability of spectral peak width

Ouput signals of nuclear scintillation detectors depend on the temperature. Not only the peak position is affected by this, but also the peak width, which is furthermore a mean to quantify the useable resolution. For nuclide identification devices, it is desirable to have quasi-static spectrum all the time, here referring to the mapping of position and resolution. As the identification compares against something "known", the positions and the resolution in the reference must match to the measurement as good as possible.

Cognitivity = adaptivity plus intrinsic physical quantity deduction

Stabilizing the peak width

We developed a special digital filter, a so-called cognitive filter, first with the simple aim to control the peak width. In modern detection equipment, one typically uses the signal shape to retrieve information about the temperature. Here, we use a deconvolution approach to unfold the exponential decays included in the digital signal and feed that knowledge back into a looped process. The filter "learns" the best configuration. With that, we have a neat way to control the resolution.

The picture shows convergence steps of the filter algorithm. With each iteration, the peak width gets smaller, which is actually what we want. The cognitive filter further stabilized the resolution variance in the regime between -12°C to +40°C.

Determining the detector material

Consequently, we have the numerical values of the signal decays for free, as part of our method. Therefore, we use them to set up a novel soft-sensor. This soft-sensor has the simple mission to compare the current decay values with a series of predefined material information. As the decay also strongly depends on the detector material and we know the current temperature, we just found a way to extract the detector material from our instrument. Our demo instrument learns the right filter configuration and tells us what material produced the pulses. The material is determined automatically.


Our results have been published in the Journal IEEE Transactions on Nuclear Science, 2014. A link to the document is coming soon.

Further reading

Kalman filter

Least-mean-squares filter