Statistical techniques for digital modulation recognition
Automatic modulation recognition is an important part of communications electronic
monitoring and surveillance systems where it is used for signal sorting and
receiver switching. '
This thesis introduces a novel application of multivariate statistical techniques
to the problem of automatic modulation classification.
The classification technique uses modulation features derived from time-domain
parameters of instantaneous signal envelope, frequency and phase. Principal component
analysis (PCA) is employed for data reduction and multivariate analysis of
variance (MANOVA) is used to investigate the data and to construct a discriminant
function to enable the classification of modulation type. MANOVA is shown to offer
advantages over the techniques already used for modulation recognition, even when
simple features are used.
The technique is used to construct a universal discriminator which is independent
of the unknown signal to noise ratio (SNR) of the received signal. The universal
discriminator is shown to extend the range of signal-to-noise ratios (SNRs) over
which discrimination is possible, being effective over an SNR range of 0-4OdB. Development
of discriminant functions using MANOVA is shown to be an extensible
technique, capable of application to more complex problems.