Chemometrics description and systematic investigation of measurement error in smartphone based spectrometer

Analytical chemistry is the science of chemical measurements, and inherent in any chemical measurement is the principle of measurement error [1,2]. Proper data analysis necessarily implies knowledge of the nature of error, with all the contributions that configure its correlated or uncorrelated nature. Once this is known, the whole error structure can be associated with physical factors and, eventually, be corrected or be taken into account in the data analysis through weighting schemes. In here we have investigated the underlying noise structures in a homemade smartphone spectrophotometer. Smart mobile phones with the digital camera associated as an attractive consumer electronic product and a small piece of digital versatile disks (DVDs) are capable of being used as portable analytic optical devices [3,4]. The images of transmitted light through cuvette were captured by smartphone rear camera and then analyzed by a home-built MATLAB program. The miniature spectrometer was used to acquisition of sampling replicates of transmittance spectrum from Orange G and Indigo Carmine solutions. The error structures of the transmittance spectra were investigated through for assessing error covariance and correlation matrices. PCA, MCR-ALS and PARAFAC were used to extract possible sources of noise. The effect of concentration change and cell positioning on error structure were investigated. To estimate the error covariance matrices, thirty five images were taken from each solution as replicates. These replicates were taken while the placed sample in the apparatus was not moved and replacing. These two conditions were done to check the effect of sample cell replacement on noise structures. In order to understand the error structure in absence of the signal of analyte, the same image acquisition procedure was done on the empty cell and the cell filled by deionized water (blank sample). This study revealed four main sources of noise in recorded signals: a) constant offset noise, b) colors border noise, c) heteroscedastic noise (uncorrelated independent error) which is proportional to transmittance spectrum and it is specific for different dyes, d) peak broadening noise which is observed when there is a broadening in of each RGB colors. This study can be used to improve instrumental setup and finding error among with it. This leads to use of measurement error information to enhance data analysis that make more accurate and precise inexpensive spectroscopic assays.