![]() Such a combination of noise contains the sum of harmonic functions, which is completely predictable, therefore causing a decrease of the entropy. The noise was generated as a sum of harmonic functions, both with 50 Hz and harmonics as frequencies and with a number of 300 frequencies randomly generated over the expected frequency range of the frequency spectrum. Reference describes a procedure that can be used to add both a white-type of noise x noise and a 50 Hz component and the harmonics, x h, as the noise of the power grid is ubiquitarian. In order to simulate the DLS TS in a realistic manner, an amount of noise was added. With these in mind, the ACR was described using Equation (7), with the experimental setup in such a manner that β = 1 (average speckle size equal with the size of the detector) and 1 subtracted to have an exponential decay of the values computed from the experimental TS. While we can chose start values close to the real ones for a known sample, such a choice can be difficult for an unknown sample and can lead to a local minimum, therefore, to values that are relatively far from the accurate values. Such a fit, with three free parameters, is not fully reproducible and the result might depend on the start values chosen for the parameters, a choice that can be more or less educated. Secondly, a combination of two exponentials has three parameters to be determined by a minimization procedure, the two diffusion coefficients at the exponent, and one multiplication coefficient. Using the fit of a combination of two or more exponentials to describe the normalized ACR is not within the remit of this work. Firstly, the work presented here is on the line of simplifying the experimental setup and the data processing procedure for DLS. This alternative was not used though, for two reasons. An improvement in the use of ANNs is presented in, where the range of the particle size to be measured was extended up to 6000 nm.Īn alternative to Equation (7) for describing the normalized ACR decay would consist of using a combination of two exponentials, under the assumption that the sample consists of two types of mono-sized particles. Both ANNs reported in proved to be several thousand times faster than fitting either the Lorentzian line to the frequency spectrum or the autocorrelation, with small relative errors. The output was the average diameter of the particles, as well, but the range of the particle size was increased to 1200 nm. A continuation of the work in is reported in and uses the autocorrelation of the DLS TS as input to an ANN. ![]() ![]() The size range was up to 350 nm, very small though, but the work was a proof of concept for using ANNs to process DLS TSs. The ANN had three layers and had the average diameter of the suspended particles as output. An averaged scattered light intensity frequency spectrum was used as the input in the work reported in. There are other procedures that have been used in processing experimentally recorded data, based on Artificial Neural Networks (ANN), DLS TS being included. ![]()
0 Comments
Leave a Reply. |