scott@scottincz.com wrote:>>If you want real numbers, you'll have to divulge the sample rate and the >>passband of the anti-alias filter. > > > > How about 48Khz 20Hz-20Khz > > >>JerryWhat is the actual bandwidth of your signal? (This is, as my father would say, like pulling teeth.) Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������

# most effecient hilbert transform method

Started by ●June 4, 2004

Reply by ●June 5, 20042004-06-05

Reply by ●June 6, 20042004-06-06

On Sat, 05 Jun 2004 17:19:22 -0400, Jerry Avins <jya@ieee.org> wrote:>scott@scottincz.com wrote: > >>>If you want real numbers, you'll have to divulge the sample rate and the >>>passband of the anti-alias filter. >> >> >> >> How about 48Khz 20Hz-20Khz >> >> >>>Jerry > >What is the actual bandwidth of your signal? (This is, as my father >would say, like pulling teeth.)20Hz-20Khz, already stated in the previous post.> >Jerry

Reply by ●June 6, 20042004-06-06

scott@scottincz.com wrote:> On Sat, 05 Jun 2004 17:19:22 -0400, Jerry Avins <jya@ieee.org> wrote: > > >>scott@scottincz.com wrote: >> >> >>>>If you want real numbers, you'll have to divulge the sample rate and the >>>>passband of the anti-alias filter. >>> >>> >>> >>>How about 48Khz 20Hz-20Khz >>> >>> >>> >>>>Jerry >> >>What is the actual bandwidth of your signal? (This is, as my father >>would say, like pulling teeth.) > > > 20Hz-20Khz, already stated in the previous post. > > > >>JerryYou stated that the range of interest is 20Hz-20KHz. Are you assured that there are no significant components above 20 KHz that can interfere with a design which assumes their absence? Maybe it doesn't matter. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������

Reply by ●June 6, 20042004-06-06

On Fri, 04 Jun 2004 21:54:14 +0100, scott@scottincz.com wrote:>What would be the most effecient method of hilbert transofrming audio? >my filter program genertates too many taps for the low frequency >performacnce i require. I don't want to use FFT due to memory >constraints in the DSP. >Thanks in advanceHi, Whew Scott! What you're trying to do is *VERY* computationally intensive. There are several ways to compute Hilbert transforms (to generate "analytic signals"), but as you're finding out, it may be impossible for you to compute an accurate Hilbert transform of a signal over the freq range of 30 Hz -to- 20 KHz when the sample rate is 48 KHz. It's possible to compute an accurate Hilbert transform of a signal over the narrow freq range of 30 Hz -to- 100 Hz when the sample rate is 48 KHz, using an "interpolated-coefficient filter" scheme, but to perform accurate Hilbert transforms over a 20 KHz range you'll need a fantastic amount of number-crunching power. I wonder if there's a way to implement your desired processing *without* using the Hilbert transform(?). Good Luck, [-Rick-]

Reply by ●June 7, 20042004-06-07

"Rick Lyons" <r.lyons@_BOGUS_ieee.org> wrote in message news:40c37201.78962218@news.sf.sbcglobal.net...> I wonder if there's a way to implement your > desired processing *without* using the > Hilbert transform(?).Well, you could use a complex IIR to isolate one sideband directly. I suggest using a complex filter on the audio instead of a real lowpass on a frequency-shifted version, because of the different widths of the upper and lower transition regions. I don't know of any tools to make designing such things a lot easier. I would try it like this: 1) Make a real analog Butterworth lowpass; 2) Shift all the poles and zeros up on the jw axis, until you have good attenuation at DC; 3) Convert to analog with the bilinear transform; and finally 4) swap the poles and zeros left to right (by negating the real component). The idea is that, by varying the parameters to the bilinear transform, you could use its frequency warping to produce a much sharper transition at Nyquist, when you would then move to DC, where it's needed. I haven't verified that it works, but if I had the OP's problem, I would give it a try. Perhaps someone with a lot of IIR experience can comment on the viability of this technique.