Following somebody's advice, I'm making a topic to expose my problem without talking about possible solutions. But first of all I'd like to restrict this topic to solutions on how to implement frequency shifting (that doesn't involve performing a DFT and shifting bins), I don't want to hear about performing a time-domain convolution with a kernel that changes through time, or doing some stuff I don't understand to get my signal in the DC and the noise in the AC (although I'm sure it would be a very good solution), my goal is not only to find a solution to my problem but also to know at last how to do that precise thing. So here's the problem : I got a signal of about 10,000,000 samples of a chirp going from frequencies 0 to 0.5 (1 being the sampling frequency), surrounded by lots of noise. I want to flatten that chirp at some frequency, so I can bandpass filter it to get rid of the noise. Right now I got a lil problem with understanding negative frequencies and the Hilbert Transform, so that would be cool if we could keep those two concepts out of this topic, if possible. Thanks.
How to shift frequencies in a non-flat manner
Started by ●January 21, 2006
Reply by ●January 21, 20062006-01-21
Michel Rouzic wrote:> Following somebody's advice, I'm making a topic to expose my problem > without talking about possible solutions. > > But first of all I'd like to restrict this topic to solutions on how to > implement frequency shifting (that doesn't involve performing a DFT and > shifting bins), I don't want to hear about performing a time-domain > convolution with a kernel that changes through time, or doing some > stuff I don't understand to get my signal in the DC and the noise in > the AC (although I'm sure it would be a very good solution), my goal is > not only to find a solution to my problem but also to know at last how > to do that precise thing. > > So here's the problem : I got a signal of about 10,000,000 samples of a > chirp going from frequencies 0 to 0.5 (1 being the sampling frequency), > surrounded by lots of noise. I want to flatten that chirp at some > frequency, so I can bandpass filter it to get rid of the noise. > > Right now I got a lil problem with understanding negative frequencies > and the Hilbert Transform, so that would be cool if we could keep those > two concepts out of this topic, if possible. Thanks.I saw an episode of 60 Minutes one time about fraudulent professional schools. One of their producers carried a hidden camera into an interview for a position at a nursing school. In the interview, the "candidate" confessed to a horrible fear of blood and needles. The school administrator smiled and said that would not be a problem! Back to the subject, though. Please describe what you mean by "flatten that chirp at some frequency", and then we might spare you the blood and needles, if we can. John
Reply by ●January 21, 20062006-01-21
Michel Rouzic wrote: ...> So here's the problem : I got a signal of about 10,000,000 samples of a > chirp going from frequencies 0 to 0.5 (1 being the sampling frequency), > surrounded by lots of noise. I want to flatten that chirp at some > frequency, so I can bandpass filter it to get rid of the noise.Sounds like an application for an adaptive filter in line enhancement mode. Another approach would be to dig out some literature about TDS (time delay spectrometry) - it's pretty much what you seem to describe. Regards, Andor
Reply by ●January 21, 20062006-01-21
"Andor" <andor.bariska@gmail.com> wrote in message news:1137849188.771937.270400@f14g2000cwb.googlegroups.com...> Michel Rouzic wrote: > ... >> So here's the problem : I got a signal of about 10,000,000 samples of a >> chirp going from frequencies 0 to 0.5 (1 being the sampling frequency), >> surrounded by lots of noise. I want to flatten that chirp at some >> frequency, so I can bandpass filter it to get rid of the noise. > > Sounds like an application for an adaptive filter in line enhancement > mode. Another approach would be to dig out some literature about TDS > (time delay spectrometry) - it's pretty much what you seem to describe. >If the frequency is proportional to time just mix it with another signal where the frequency is proportional to time and end up with a flat frequency. Why not? Best of Luck - Mike
Reply by ●January 21, 20062006-01-21
john wrote:> Michel Rouzic wrote: > > Following somebody's advice, I'm making a topic to expose my problem > > without talking about possible solutions. > > > > But first of all I'd like to restrict this topic to solutions on how to > > implement frequency shifting (that doesn't involve performing a DFT and > > shifting bins), I don't want to hear about performing a time-domain > > convolution with a kernel that changes through time, or doing some > > stuff I don't understand to get my signal in the DC and the noise in > > the AC (although I'm sure it would be a very good solution), my goal is > > not only to find a solution to my problem but also to know at last how > > to do that precise thing. > > > > So here's the problem : I got a signal of about 10,000,000 samples of a > > chirp going from frequencies 0 to 0.5 (1 being the sampling frequency), > > surrounded by lots of noise. I want to flatten that chirp at some > > frequency, so I can bandpass filter it to get rid of the noise. > > > > Right now I got a lil problem with understanding negative frequencies > > and the Hilbert Transform, so that would be cool if we could keep those > > two concepts out of this topic, if possible. Thanks. > > I saw an episode of 60 Minutes one time about fraudulent professional > schools. One of their producers carried a hidden camera into an > interview for a position at a nursing school. In the interview, the > "candidate" confessed to a horrible fear of blood and needles. The > school administrator smiled and said that would not be a problem! > > Back to the subject, though. Please describe what you mean by "flatten > that chirp at some frequency", and then we might spare you the blood > and needles, if we can. > > JohnI meant bring it to -for example- a flat 0.25 frequency sinusoid, or even at DC
Reply by ●January 21, 20062006-01-21
Andor wrote:> Michel Rouzic wrote: > ... > > So here's the problem : I got a signal of about 10,000,000 samples of a > > chirp going from frequencies 0 to 0.5 (1 being the sampling frequency), > > surrounded by lots of noise. I want to flatten that chirp at some > > frequency, so I can bandpass filter it to get rid of the noise. > > Sounds like an application for an adaptive filter in line enhancement > mode. Another approach would be to dig out some literature about TDS > (time delay spectrometry) - it's pretty much what you seem to describe.TDS? It's "a sophisticated method for obtaining anechoic results in echoic spaces", right? How is it gonna flatten my chirp?
Reply by ●January 21, 20062006-01-21
Mike Yarwood wrote:> "Andor" <andor.bariska@gmail.com> wrote in message > news:1137849188.771937.270400@f14g2000cwb.googlegroups.com... > >>Michel Rouzic wrote: >>... >> >>>So here's the problem : I got a signal of about 10,000,000 samples of a >>>chirp going from frequencies 0 to 0.5 (1 being the sampling frequency), >>>surrounded by lots of noise. I want to flatten that chirp at some >>>frequency, so I can bandpass filter it to get rid of the noise. >> >>Sounds like an application for an adaptive filter in line enhancement >>mode. Another approach would be to dig out some literature about TDS >>(time delay spectrometry) - it's pretty much what you seem to describe. >> > > If the frequency is proportional to time just mix it with another signal > where the frequency is proportional to time and end up with a flat > frequency. Why not? > > Best of Luck - Mike > >But that would involve time-domain convolution by a kernel that changes with time (i.e., multiply the original signal by a sine wave and filter). So it doesn't match the OP's criteria. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com
Reply by ●January 21, 20062006-01-21
Michel Rouzic wrote:> Following somebody's advice, I'm making a topic to expose my problem > without talking about possible solutions. > > But first of all I'd like to restrict this topic to solutions on how to > implement frequency shifting (that doesn't involve performing a DFT and > shifting bins), I don't want to hear about performing a time-domain > convolution with a kernel that changes through time, or doing some > stuff I don't understand to get my signal in the DC and the noise in > the AC (although I'm sure it would be a very good solution), my goal is > not only to find a solution to my problem but also to know at last how > to do that precise thing. > > So here's the problem : I got a signal of about 10,000,000 samples of a > chirp going from frequencies 0 to 0.5 (1 being the sampling frequency), > surrounded by lots of noise. I want to flatten that chirp at some > frequency, so I can bandpass filter it to get rid of the noise. > > Right now I got a lil problem with understanding negative frequencies > and the Hilbert Transform, so that would be cool if we could keep those > two concepts out of this topic, if possible. Thanks. >Well _I_ want to understand the Einstein's general theory of relativity, but I don't want to mess with any physics or calculus. If you can help me with that, within my constraints, then I'll help you with your problem within your constraints. Otherwise try to take away some of the constraints so we can actually suggest something that might work. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com
Reply by ●January 21, 20062006-01-21
"Tim Wescott" <tim@seemywebsite.com> wrote in message news:pq2dne9jHfFtD0_enZ2dnUVZ_smdnZ2d@web-ster.com...> Mike Yarwood wrote: >> "Andor" <andor.bariska@gmail.com> wrote in message >> news:1137849188.771937.270400@f14g2000cwb.googlegroups.com... >> >>>Michel Rouzic wrote: >>>... >>> >>>>So here's the problem : I got a signal of about 10,000,000 samples of a >>>>chirp going from frequencies 0 to 0.5 (1 being the sampling frequency), >>>>surrounded by lots of noise. I want to flatten that chirp at some >>>>frequency, so I can bandpass filter it to get rid of the noise. >>> >>>Sounds like an application for an adaptive filter in line enhancement >>>mode. Another approach would be to dig out some literature about TDS >>>(time delay spectrometry) - it's pretty much what you seem to describe. >>> >> >> If the frequency is proportional to time just mix it with another signal >> where the frequency is proportional to time and end up with a flat >> frequency. Why not? >> >> Best of Luck - Mike >> >> > But that would involve time-domain convolution by a kernel that changes > with time (i.e., multiply the original signal by a sine wave and filter). > > So it doesn't match the OP's criteria. >I think it does, he didn't want a solution involving DFTs and shifting bins but he just didn't want to hear about the other stuff so if we call it something else... Best of Luck - Mike
Reply by ●January 21, 20062006-01-21
Mike Yarwood wrote:> "Tim Wescott" <tim@seemywebsite.com> wrote in message > news:pq2dne9jHfFtD0_enZ2dnUVZ_smdnZ2d@web-ster.com......>>But that would involve time-domain convolution by a kernel that changes >>with time (i.e., multiply the original signal by a sine wave and filter). >> >>So it doesn't match the OP's criteria. >> > > I think it does, he didn't want a solution involving DFTs and shifting bins > but he just didn't want to hear about the other stuff so if we call it > something else...Michel knows how he wants to do it, he just doesn't know how to do it that way. He even knows by now that it's the hard way, but he thinks it will be easier than learning what he needs to know. Maybe he's right. How do you help someone who says he doesn't want to hear about chocks or jacks, just about using a wrench to change a tire? Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
