Method of analysing a spectral peak using a neural network

Spectral emissions of different wavelengths overlap to form an interfered peak at a position on a detector. For one or more overlapping components, a neural net of a spectrometer controller generates and outputs an associated curve. The neural net is trained to output data characterising the shape of the associated curve based on the detector position. The neural net may output an encoded representation of the associated curve shape. Generating the associated curve may include decoding the encoded representation. Training may be based on multiple training peaks, each being a single spectral emission at a different detector position, perhaps associated with a single-element transition metal solution. Training may be repeated by identifying a calibration sample. A sample peak may be identified as an interfered peak, perhaps by counting the number of zero crossings in the first derivative of the peak. The spectrometer may be an optical emission spectrometer. It may comprise an echelle grating that diffracts light onto a two-dimensional array detector.