A novel approach for the efficient digital decoding of DTMF signals used in
telecommunications is based on an artificial neural network (ANN) implemen
ted on a digital signal processor (DSP). After the appropriate training pha
se the ANN is able to associate a digital output corresponding to the recei
ved DTMF signal. The proposed approach may allow the possibility of impleme
nting a real-time non-intrusive decoder. This decoder is fully described an
d particular attention paid to the solutions adopted for optimising the AWN
's computation on the DSP. A comparison is made of the performance obtained
by the proposed approach, and that, recommended by the CCITT standard. Fin
ally, experimental results are given which highlight the performance of the
apparatus in different real conditions in DTMF signal decoding.