Hi Guys; We are now designing a digital commnication system. Signal's sample rate is 2.56MHz, and signal bandwidth is 120kHz. I would like to run low pass filtering and decimation by 8 (320kHz) continiously. In this case, can I use frequnecy sampling filter(FSF) which is described in Richard Lyon's book, "Undestanding Digital Signal Processing". This book says that FSF is effective when the passband and the transition are narrow. Is FSF still effective for our case ? As FSF has recursive architecture, I think we have to complete the FSF's calculation per 2.56MHz(400ns). On the other hands, we can use 3us (1/320kHz) for traditional FIR filter's calculation. Am I correct and which is better way ? By the way, this book describes not only theory, but also many practical tips. In addtion to this, English is very plain for the Japanese. Mikio
Frequency Sampling Filter
Started by ●March 23, 2006
Reply by ●March 25, 20062006-03-25
On Thu, 23 Mar 2006 07:13:12 -0600, "mikio" <mikio.hagiwara@kcf.biglobe.ne.jp> wrote:>Hi Guys;Hello Mikio,>We are now designing a digital commnication system. Signal's sample rate >is 2.56MHz, and signal bandwidth is 120kHz. I would like to run low pass >filtering and decimation by 8 (320kHz) continiously. In this case, can I >use frequnecy sampling filter(FSF) which is described in Richard Lyon's >book, "Undestanding Digital Signal Processing". This book says that FSF >is effective when the passband and the transition are narrow. Is FSF >still effective for our case ?You did not say what your lowpass filter's passband width or transition region width were. If you know those widths, my Figures 7-29 and 7-30 should help you decide if an FSF or a standard Parks-McClellan-designed lowpass filter will be more computationally efficient for your application. If you want to perform decimation, perhaps the interpolated FIR (IFIR) filters in Section 7.2 of my book would be more sensible. Although the book does not discuss IFIR filter in relation to decimation, I wrote a conference paper that does discuss decimation with IFIR filters. I presented that paper at the 2005 Comp.dsp Conference and at the Gspx 2004 Conference. That paper is available at: http://www.techonline.com/pdf/pavillions/gspx/2004/859.pdf>As FSF has recursive architecture, I think we have to complete the FSF's >calculation per 2.56MHz(400ns). On the other hands, we can use 3us >(1/320kHz) for traditional FIR filter's calculation. >Am I correct and which is better way ?Humm, I'm not sure exactly what you mean here. Sorry.> >By the way, this book describes not only theory, but also many practical >tips. In addtion to this, English is very plain for the Japanese.I am not sure what your words "very plain" mean, but I hope they mean "understandable". Good Luck Mikio, and let me know if I can help in some way. [-Rick-]
Reply by ●March 25, 20062006-03-25
Dear Mr.Lyons;>If you want to perform decimation, perhaps the interpolated >FIR (IFIR) filters in Section 7.2 of my book would be >more sensible. Although the book does not discuss >IFIR filter in relation to decimation, I wrote a conference >paper that does discuss decimation with IFIR filters. >I presented that paper at the 2005 Comp.dsp Conference and >at the Gspx 2004 Conference. That paper is >available at: > >http://www.techonline.com/pdf/pavillions/gspx/2004/859.pdf >Thank you for the comment. I'll read this section carefully and your paper.>>As FSF has recursive architecture, I think we have to complete theFSF's>>calculation per 2.56MHz(400ns). On the other hands, we can use 3us >>(1/320kHz) for traditional FIR filter's calculation. >>Am I correct and which is better way ? > >Humm, I'm not sure exactly what you mean >here. Sorry. >>Our system's master clock is 61.44MHz, input sample rate and final sample rate is 2.56MHz, and 320kHz respevtively. pass-band is 120kHz. I thought case1) and case 2) case 1) If the output rate is one eighth, 320kHz, I can use 384 taps FIR filter by using falded filter because ue can have 184 muliplication time per sample period. case 2) If the input sample rate and output sample rate is same and 2.56MHz, I can execute 24 multiplication calculation per sample period. Altought final sample rate is 320kHz, I thought I couldn't use 184 multiplication because FSF architecture seemed to be recursive. On the other hand, the ratio of fs vs pass-band is narrower than 0.05, FSF seemed to be better than Parks-McClellan filter from your book. So I run FSF first, and then decimate data simply. I didn't know which way was better. From your suggestion, I'll check third way, IFIR filter.>>By the way, this book describes not only theory, but also manypractical>>tips. In addtion to this, English is very plain for the Japanese. > >I am not sure what your words "very plain" mean, >but I hope they mean "understandable". >Good Luck Mikio, and let me know if I can help >in some way.Yes, I mean "understandable" sentenses and many figures.
Reply by ●March 28, 20062006-03-28
On Sat, 25 Mar 2006 11:49:18 -0600, "mikio" <mikio.hagiwara@kcf.biglobe.ne.jp> wrote:> >Dear Mr.Lyons;Hi, please call me Rick. My friends do. (Actually, my enemies also call me Rick.) (snipped)>On the other hand, the ratio of fs vs pass-band is narrower than 0.05, FSF >seemed to be better than Parks-McClellan filter from your book. >So I run FSF first, and then decimate data simply.If the ratio (passband)/(sample rate) is 0.05, then I think an FSF will require fewer computations per output sample than a Parks-McClellan-designed (and folded) tapped-delay line FIR filter. You didn't say what the desired transition width was, but I believe an FSF will be more efficient than a P-M-designed FIR. Remember now, your desired transition width is what will determine the value of "N" in an FSF.>>I am not sure what your words "very plain" mean, >>but I hope they mean "understandable". >>Good Luck Mikio, and let me know if I can help >>in some way. > >Yes, I mean "understandable" sentenses and many figures.Ah, great. Glad to hear that. Thanks!! Good Luck, [-Rick-]
