On 09/02, Dario Sanfilippo wrote:
>
> Niemitalo's isn't great either, depending on the frequency range you're
> analysing; here is a better one, courtesy of Martin Vicanek, who provided
> the coefficients for a 16th-order one in the paper:
> https://vicanek.de/articles/ReverseIIR.pdf.
>
> combAP(coeff, x) = f ~ _ with {
> f(s) = x'' - coeff * x + coeff * s';
> };
> coeffsHilbertVicanek = waveform {
> 0.0406273391966415,
> 0.2984386654059753,
> 0.5938455547890998,
> 0.7953345677003365,
> 0.9040699927853059,
> 0.9568366727621767,
> 0.9815966237057977,
> 0.9938718801312583,
> 0.1500685240941415,
> 0.4538477444783975,
> 0.7081016258869689,
> 0.8589957406397113,
> 0.9353623391637175,
> 0.9715130669899118,
> 0.9886689766148302,
> 0.9980623781456869
> };
> coeffHilbertVicanek(i) = coeffsHilbertVicanek , int(i) : rdtable;
> hilbertVicanek(x) = (x : seq(i, 8, combAP(coeffHilbertVicanek(i)))) , (x :
> seq(i, 8, combAP(coeffHilbertVicanek(i + 8))) : mem);

Confused... the freq responce of

        process = 1-1' : hilbertVicanek;

doesn't look as a hilbert transform at all...

So I tried to simplify and fix your code:

        hilbertVicanek = H(a1), H(a2)' with {
                a1 = 0.0406273391966415, 0.2984386654059753, 
0.5938455547890998, 0.7953345677003365,
                     0.9040699927853059, 0.9568366727621767, 
0.9815966237057977, 0.9938718801312583;
                a2 = 0.1500685240941415, 0.4538477444783975, 
0.7081016258869689, 0.8589957406397113,
                     0.9353623391637175, 0.9715130669899118, 
0.9886689766148302, 0.9980623781456869;
                H_sect(coeff, x) = f ~ _ with { f(s) = x'' - coeff * x + coeff 
* s'; };
                H(as) = seq(i, outputs(as), H_sect(ba.take(i+1, as)));
        };

seems to work... And now we can see that allpass/pipeline is the same as in
hilbert() from https://sourceforge.net/p/faudiostream/mailman/message/36710453/

        hilbert = _ <: H(a2), H(a1)' with {
                a1 = 0.6923878, 0.9360654322959, 0.9882295226860 , 
0.9987488452737;
                a2 = 0.4021921162426, 0.8561710882420, 0.9722909545651, 
0.9952884791278;
                H_sect(a) = f ~ _ with { f(y, x) = a^2 * (x + y') - x''; };
                H(as) = seq(i, outputs(as), H_sect(ba.take(i+1, as)));
        };

just the order and coeffs differ.

Now. IIRC this filter is based on hiir library

        https://ldesoras.fr/prod.html
        https://ldesoras.fr/src/hiir-1.50.zip

which allows to make the filter of any order/transition. Let me show my
very old implementation written in fpp which uses this library, it
computes the coeffs at init time.

        hiir_hilber(N, T) = hilbert with {
                coef = FPP() eval {
                MAKE:   -I /home/oleg/SRC/DSP/
                FILE:   #include <hiir/PolyphaseIir2Designer.cpp>
                decl:   double $c[2*$(N)];
                init:   
hiir::PolyphaseIir2Designer::compute_coefs_spec_order_tbw($c, 2*$(N), $(T));
                exec:   $c
                };

                getc = FPP(i, n) { $(coef)[$i + 2*$n] };

                hilbert = _ <: H(0), H(1)' with {
                        ap(a) = f ~ _ with { f(y, x) = a * (x + y') - x''; };
                        H(i) = seq(n, N, ap(getc(i,n)));
                };
        };

A "random" test-case to compare it with your (fixed) hilbertVicanek:

        process = 1-1' <: hiir_hilber(8, .001), hilbertVicanek;

freq responces are very close, and I guess could be even closer with
another T=.001 argument.

Oleg.



_______________________________________________
Faudiostream-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/faudiostream-users

Reply via email to