of multiple analytes in mixtures by sensors is faced by several problems. For
the determination of the regression parameters of the MLR (expression
(7)) and of all other multivariate calibration methods
as many sensor signals as analytes to be quantified are needed since otherwise
the set of equations would be statistically underdetermined .
In real world measurements, even more signals are needed as noise, drifts, interfering
unknown analytes and other non-ideal influences "consume" additional
degrees of freedom. Typically, the sensors used are not selective for a single
analyte also known as crossreactivity. The resulting covariance of the sensor
signals causes additional problems for the calibration resulting in the need
of as many sensors as possible, which show different sensitivity patterns for
the analytes of interest. Thus, the common approach to quantify several analytes
simultaneously is the combination of many sensors on a sensor array with the
sensors showing different sensitivity patterns for the different analytes and
a subsequent multivariate data analysis -.
An advantage of this unspecific multisensor approach is the possibility of using
the same array setup for many different analytes without the need of finding
specific sensor materials .
Usually only one single feature of the sensor response (value of the sensor
signal) is extracted, such as the height of the sensor response at equilibrium
or the slope of the sensor response resulting in the need of more sensors in
the sensor array than the number of analytes to be quantified.