Pattern Classification Techniques for Lung Cancer Diagnosis by an Electronic Nose - Computational Intelligence in Healthcare: Advanced Methodologies - page 388

5.2 Signal Conditioning

The electrical signals generated by sensor interface circuits are often not adequate
for acquisition into a computer and must be further processed by a number
of analog signal conditioning circuits. The main objective of these operations
consists in:

- Buffering , required to isolate different electronic stages and avoid impedance
loading errors
- Amplifying , in order to bring the signal of the interface circuits to a level
suitable for the dynamic range of the subsequent analog-to-digital converter
and filtering
- Filtering , to remove unwanted frequency components from the sensor signals

For more details on signal conditioning refer to [4].

5.3 Signal Preprocessing

The first computational stage after the sensor array data have been acquired is
signal preprocessing, which aims mainly at extracting the relevant information
from the sensor responses and to prepare data for further multivariate pattern
analysis minimizing those effect (e.g., baseline drift) that can affect the quality
of the signal. During this phase, three main operations are executed:

1. Baseline manipulation
2. Feature extraction
3. Normalization

Baseline manipulation. Baseline manipulation is performed for the purposes
of drift compensation, contrast enhancement and scaling. The three most com-
mon techniques, all based on a manipulation of the sensor response with respect
to its baseline are:

- Differential . The baseline x s (0) is subtracted from the sensor response x s ( t ):

y s ( t )=( x s ( t )+ δ a ) − ( x s (0) + δ a )= x s ( t ) − x s (0) (1)

This allows to compensate any additive noise or drift δ A that may be present
in the sensor signal.
- Relative . The sensors response x s ( t ) is divided by the baseline x s (0):

y s ( t )= x s ( t )(1 + δ M )

x s ( t )
x s (0) x s (0)(1 + δ M ) =

(2)

This allows to compensate any multiplicative noise or drift δ M that may be
present in the sensor signal. It provides dimensionless response.