search

sfstoolbox / sfs-matlab

SFS Toolbox for Matlab/Octave
https://sfs-matlab.readthedocs.io
MIT License
98 stars 39 forks source link

Nfchoa sosfilter #65

Closed hagenw closed 8 years ago

hagenw commented 8 years ago

This is a subtask from #57, which discusses only the improvement of the SOS filter implementations.

hagenw commented 8 years ago

It is working under Octave, but the new filter implementation seems to be a lot more unstable. Try the following:

Under master branch:

>> conf = SFS_config;
>> conf.secondary_sources.number = 56;
>> conf.nfchoa.order = 7;
>> conf.N = 8;
>> x0 = secondary_source_positions(conf);
>> [~,dm] = driving_function_imp_nfchoa(x0,[0 -1 0],'pw',conf)
dm =

   1.00000   0.00000   0.00000   0.00000   0.00000   0.00000   0.00000   0.00000
  -0.99741   0.00516   0.00513   0.00511   0.00508   0.00505   0.00503   0.00500
   0.99226  -0.01540  -0.01524  -0.01508  -0.01492  -0.01476  -0.01461  -0.01446
  -0.98458   0.03055   0.03000   0.02945   0.02891   0.02838   0.02785   0.02733
   0.97444  -0.05039  -0.04896  -0.04755  -0.04617  -0.04481  -0.04349  -0.04219
  -0.96191   0.07462   0.07152   0.06852   0.06559   0.06275   0.06000   0.05732
   0.94708  -0.10285  -0.09699  -0.09135  -0.08593  -0.08072  -0.07571  -0.07091
  -0.93007   0.13467   0.12455   0.11494   0.10581   0.09717   0.08897   0.08122

Now do the same under the nfchoa_sosfilter branch:

>> conf = SFS_config;
>> conf.secondary_sources.number = 56;
>> conf.nfchoa.order = 7;
>> conf.N = 8;
>> x0 = secondary_source_positions(conf);
>> [~,dm] = driving_function_imp_nfchoa(x0,[0 -1 0],'pw',conf)
dm =

   1.0000e+00   0.0000e+00   0.0000e+00   0.0000e+00   0.0000e+00   0.0000e+00   0.0000e+00   0.0000e+00
  -2.0026e+00  -1.0357e-02  -4.0173e-05  -1.0304e-02  -9.3253e-05  -1.0251e-02  -1.4579e-04  -1.0199e-02
   9.9996e-01   8.0031e-05   1.9857e-05   3.1519e-04  -9.7690e-05   8.9220e-04  -9.0432e-04   2.8262e-03
  -2.0103e+00  -4.1390e-06  -1.5719e-04   4.7748e-01  -3.8074e-03   4.6288e+00  -1.4899e-01   4.0835e+01
   9.9974e-01   7.3942e-08   3.1905e-04  -3.8231e-02   9.1867e-04  -1.2901e+00  -1.0752e+00  -3.7306e+01
  -2.0182e+00  -6.8764e-09  -2.6361e-03  -1.7587e+01  -3.3668e-01  -2.3411e+03  -1.8376e+02  -2.3342e+05
   9.9911e-01   1.9117e-10   5.3132e-03   2.6142e+00   1.3499e+00   1.3028e+03  -7.8639e+02   4.5332e+05
  -2.0257e+00  -2.5651e-11  -4.3534e-02   5.6610e+02  -4.7270e+01   1.1766e+06  -1.9883e+05   1.4867e+09

If you do the same with larger conf.N values you will end up with very high number like e+250 and a lot of NaN.

hagenw commented 8 years ago

As there seems to be no easy solution to the problems we face under Octave, I would say we will not fix this in the near future.