Page 2 of 2

Re: hilbert transform in green array (for statistics)

Posted: Wed Jun 01, 2016 7:47 pm
by martinvicanek
I haven't read the paper so I cannot comment on your formulas. Anyway, here is a "green" Hilbert transformer, optimized for low latency.

Re: hilbert transform in green array (for statistics)

Posted: Wed Jun 01, 2016 9:50 pm
by tester
The schematic randomly crashes when I open it or navigate into submodules. FS3081 skylark, winxppro.

Re: hilbert transform in green array (for statistics)

Posted: Wed Jun 01, 2016 10:17 pm
by martinvicanek
tester wrote:The schematic randomly crashes when I open it or navigate into submodules. FS3081 skylark, winxppro.
Hmm, it does not crash here. FS 3.0.8.1 Skylark, Win7 64bit.

Re: hilbert transform in green array (for statistics)

Posted: Wed Jun 01, 2016 11:24 pm
by tester
Maybe it's something winxp related. it seems to work on win8. and probably it will work in exported exe.

Now, some clarification where this hilbert part is used. Some quotes and my questions.

"we used zero-phase filters, with Fourier amplitude shaped as the positive portion of a cosine function."

Is this what kind of filter? butterworth with Q=2 and abs on output - or something else?

"We used a bank of 30 such filters, with center frequencies equally spaced on an equivalent rectangular bandwidth (ERB)N scale, spanning 52–8844 Hz. Their (3 dB) bandwidths were..."

While there is info on this ERB scale, I still don't know how it defines split points.

Then there is info, that envelopes were extracted, compressed by a power x^0.3 and downsampled (following low-
pass filtering) to 400Hz (sample rate), which gives 200Hz bandwidths. I'm not sure if just sharp lowpass filters are enough here?

Having these 30 subbands, 200Hz each - various configurations of modulation filters are used. For the part with hilbert transform - there are 7 modulation filters used, with octave-spaced modulation frequencies, ranging from 1.5625 to 100Hz. And computations are made between various pairs of filters.

So this hilbert approximation must be rather relatively good for low and very low frequencies.