As we said, what we need besides an amplification is a signal elaboration in the frequency domain  to compensate the lack of the auditory line of a deaf subject, in comparison with norm .

For this reason we resort to an internal equalization of the nominal telephone band providing the final user with the possibility to have a fequency response of the system that is opposite, in a certain way, to his own distorced auditory line. What we have exactly achieved is a uniform subdivision of the band [0, 4kHz], in which the height of every "channel" created is adjustable. In every case this choice let us obtain a not uniform subdivision of the whole band; a succession of bands with an increasing width according to a 2ⁿ factor (where n correspond to the position of the band) gives the reason for the decrease of the discrimination capability of human ear with the increasing of frequencies. But we can't forget that the present project addresses to particular users, who has different requirements from normal hearing people. In every case it put itself as an alternative and as a term of comparison with a previous analogical prototype that was characterized by having 5 not uniform "channels".

What has been achieved consists in the polyphase implementation of a uniform Discrete Fourier Transform filter bank; that allows to use one only prototype low-pass filter whose shifted versions allow to cover the whole band of interest. A schematic representation is given here:

  Figure 1:        Uniform DFT filter bank polyphase implementation.

The W_pk shown in the figure are the polyphase components of the basic prototype filter, that is a FIR filter: This choice assures the phase linearity of the whole system. More characteristics of the final product are the monotonicity of the transitions between adjacent subbands adjusted with differents heights and the absence in the same transition points of bumps or subsidences in the case of "channels" with equal attenuation. The DFT shown in fig. 1 is a 16 point and imply the creation of 16 subbands, 8 of them are real in the limits of the nominal telephone band. The choice of the order of the DFT depends, in particular on considerations about SNR: it has to be kept high because of the characteristics of the final user and for the opportunity to use a radix-4 FFT algorithm to calculate the Fourier transformation. This kind of algorithm gives advantages in terms of computation in comparison with a more common radix-2. In the above mentioned figure some P_k factors has also been included: they represent the adjustable attenuations to define the equalization line. All this has finally to be put at the user disposal. Finally we give a note about the delay introduced by the system that is, with the order fixed for the filter, 11ms, that's equivalent to (L-1)/2 samples (being L the lenght of the impulsive response of the prototype filter).

[  Index | Previous page | Next page ]