I am trying to accurately recover the carrier component of a sampled modulated RF signal. When I transform a sample window I see my carrier and it's information side bands. The side bands appear to be at different amplitudes (non symmetrical) around the carrier and the carriers frequency seems to be distorted a few Hz. This frequency distortion of the carrier and what appears to be single side band suppression is depended on what random data happens to be in the sample window at the time I sample. It appears that I might be able to recover the true frequency of the carrier by taking the average frequency of the power spectrum. I just don't know how to do this yet ....pivot table maybe? Any thoughts? Thomas
Frequency average of power spectrum
Started by ●August 20, 2004
Reply by ●August 21, 20042004-08-21
On Fri, 20 Aug 2004 17:05:42 GMT, "Thomas Magma" <somewhere@overtherainbow.com> wrote:>I am trying to accurately recover the carrier component of a sampled >modulated RF signal. When I transform a sample window I see my carrier and >it's information side bands. The side bands appear to be at different >amplitudes (non symmetrical) around the carrier and the carriers frequency >seems to be distorted a few Hz. This frequency distortion of the carrier and >what appears to be single side band suppression is depended on what random >data happens to be in the sample window at the time I sample. It appears >that I might be able to recover the true frequency of the carrier by taking >the average frequency of the power spectrum. I just don't know how to do >this yet ....pivot table maybe? > >Any thoughts? > >Thomas >Hi, if your signal's sidebands are close to zero Hz (DC), there will be spectral wraparound (overlap) at zero Hz that might make your sidelobes look asymmetrical when they're really not! Thomas, I'll bet someone here can help you but you'll have to clearly explain what "frequency distortion of the carrier" means. Also you'll have to explain what you mean by "by taking the average frequency of the power spectrum". Do you mean computing the average of a series of carrier frequency measurements obtained from a series of FFTs? [-Rick-]
Reply by ●August 23, 20042004-08-23
"Thomas Magma" <somewhere@overtherainbow.com> wrote in message news:<GbqVc.164131$gE.150048@pd7tw3no>...> I am trying to accurately recover the carrier component of a sampled > modulated RF signal. When I transform a sample window I see my carrier and > it's information side bands. The side bands appear to be at different > amplitudes (non symmetrical) around the carrier and the carriers frequency > seems to be distorted a few Hz. This frequency distortion of the carrier and > what appears to be single side band suppression is depended on what random > data happens to be in the sample window at the time I sample. It appears > that I might be able to recover the true frequency of the carrier by taking > the average frequency of the power spectrum.There are several important details that are missing here: - What's the sample rate? - What's the data frame size? - What's the (nominal) signal bandwidth? - What's the (nominal) carrier frequency? - What's the SNR? A naive attempt to explain the offset of the carrier frequency as well as the asymmetry of the sidebands, would be based on that they span just a few frequency bins in the spectrum. If that's the case, the exact (mis)alignment of the carrier with the frequency bins might explain an "unaccurate" frequency estimate as well as the asymmetry between side bands. This theory would also be consistent with your observation that the exact shapes of the spctrum varies in a random way as function of which data frame one analyzes. One way of improving the frequency resolution wold be to increase the data frame length and compute the periodogram. If you have a clear, nice carrier, you might be able to see it in the periodograms. Unfortunately, the periodogram is very sensitive to noise, so you would need to be a bit careful in your analysis. Another way to improve your impression of the spectrum would be to average several spectra. This way the "random" dependency on input data would be removed, and you might get a more coherent picture. The downside here is that consistent spectral averages come at the expence of detail and time dependency. I suggest you check out a text on spectrum estimation, e.g. Kay: "Modern Spectral Estimation - Theory and Application" Prentice-Hall, 1988. Rune
Reply by ●August 23, 20042004-08-23
Thomas Magma wrote:> I am trying to accurately recover the carrier component of a sampled > modulated RF signal. When I transform a sample window I see my carrier and > it's information side bands. The side bands appear to be at different > amplitudes (non symmetrical) around the carrier and the carriers frequency > seems to be distorted a few Hz. This frequency distortion of the carrier and > what appears to be single side band suppression is depended on what random > data happens to be in the sample window at the time I sample. It appears > that I might be able to recover the true frequency of the carrier by taking > the average frequency of the power spectrum. I just don't know how to do > this yet ....pivot table maybe? > > Any thoughts? > > ThomasIs the receiver passband symmetric? Is the signal in the center of it? Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●August 24, 20042004-08-24
I'm currently working with sampled data generated by a computer. I did this to remove any unseen hardware variables. I will then re-introduce the hardware later when I'm happy with my algorithm. "Jerry Avins" <jya@ieee.org> wrote in message news:4129ff07$0$21757$61fed72c@news.rcn.com...> Thomas Magma wrote: > > > I am trying to accurately recover the carrier component of a sampled > > modulated RF signal. When I transform a sample window I see my carrierand> > it's information side bands. The side bands appear to be at different > > amplitudes (non symmetrical) around the carrier and the carriersfrequency> > seems to be distorted a few Hz. This frequency distortion of the carrierand> > what appears to be single side band suppression is depended on whatrandom> > data happens to be in the sample window at the time I sample. It appears > > that I might be able to recover the true frequency of the carrier bytaking> > the average frequency of the power spectrum. I just don't know how to do > > this yet ....pivot table maybe? > > > > Any thoughts? > > > > Thomas > > Is the receiver passband symmetric? Is the signal in the center of it? > > Jerry > -- > Engineering is the art of making what you want from things you can get. > ����������������������������������������������������������������������� >
Reply by ●August 24, 20042004-08-24
"frequency distortion of the carrier" I meant that the carrier's frequency appears to be offset from where I know it should be, by up to a Hz or so, depending on what information data is present. I'm currently working with sampled data generated by a computer. I did this to remove any unseen hardware variables and so I knew exactly what frequency the carrier should appear at. I will then re-introduce the hardware later when I'm happy with my algorithm. It turns out that I was right. You can reconstruct the carriers true frequency by taking the average frequency of the power spectrum centered around the offset carrier. Here is the formula how: SUM(f_i * mag(i)/SUM(mag(i)) The method works great but takes too much processing power for my application, so I can't really use it. So back to the drawing board. Thomas "Rick Lyons" <r.lyons@_BOGUS_ieee.org> wrote in message news:41272fb0.1387517671@news.sf.sbcglobal.net...> On Fri, 20 Aug 2004 17:05:42 GMT, "Thomas Magma" > <somewhere@overtherainbow.com> wrote: > > >I am trying to accurately recover the carrier component of a sampled > >modulated RF signal. When I transform a sample window I see my carrierand> >it's information side bands. The side bands appear to be at different > >amplitudes (non symmetrical) around the carrier and the carriersfrequency> >seems to be distorted a few Hz. This frequency distortion of the carrierand> >what appears to be single side band suppression is depended on whatrandom> >data happens to be in the sample window at the time I sample. It appears > >that I might be able to recover the true frequency of the carrier bytaking> >the average frequency of the power spectrum. I just don't know how to do > >this yet ....pivot table maybe? > > > >Any thoughts? > > > >Thomas > > > > Hi, > if your signal's sidebands are close to > zero Hz (DC), there will be spectral > wraparound (overlap) at zero Hz that might > make your sidelobes look asymmetrical when they're > really not! > > Thomas, I'll bet someone here can help you but you'll > have to clearly explain what "frequency distortion of > the carrier" means. Also you'll have to explain what > you mean by "by taking the average frequency of the > power spectrum". Do you mean computing the > average of a series of carrier frequency measurements > obtained from a series of FFTs? > > [-Rick-] > > > >
Reply by ●August 24, 20042004-08-24
Thomas Magma wrote: How do you read one Hz? What is your bin spacing? Are you accounting for how a signal not exactly at a bin center "bleeds" into adjacent (and more distant) bins? Do you shape the window to minimize that? I haven't been following closely enough to remember if those issues have already been addressed. If they were, please pardon me. Jerry -- Engineering is the art of making what you want from things you can get. ����������������������������������������������������������������������� Thomas Magma wrote:> "frequency distortion of the carrier" I meant that the carrier's frequency > appears to be offset from where I know it should be, by up to a Hz or so, > depending on what information data is present. I'm currently working with > sampled data generated by a computer. I did this to remove any unseen > hardware variables and so I knew exactly what frequency the carrier should > appear at. I will then re-introduce the hardware later when I'm happy with > my algorithm. > > It turns out that I was right. You can reconstruct the carriers true > frequency by taking the average frequency of the power spectrum centered > around the offset carrier. Here is the formula how: SUM(f_i * > mag(i)/SUM(mag(i)) > > The method works great but takes too much processing power for my > application, so I can't really use it. > > So back to the drawing board. > > Thomas > > > > > > "Rick Lyons" <r.lyons@_BOGUS_ieee.org> wrote in message > news:41272fb0.1387517671@news.sf.sbcglobal.net... > >>On Fri, 20 Aug 2004 17:05:42 GMT, "Thomas Magma" >><somewhere@overtherainbow.com> wrote: >> >> >>>I am trying to accurately recover the carrier component of a sampled >>>modulated RF signal. When I transform a sample window I see my carrier > > and > >>>it's information side bands. The side bands appear to be at different >>>amplitudes (non symmetrical) around the carrier and the carriers > > frequency > >>>seems to be distorted a few Hz. This frequency distortion of the carrier > > and > >>>what appears to be single side band suppression is depended on what > > random > >>>data happens to be in the sample window at the time I sample. It appears >>>that I might be able to recover the true frequency of the carrier by > > taking > >>>the average frequency of the power spectrum. I just don't know how to do >>>this yet ....pivot table maybe? >>> >>>Any thoughts? >>> >>>Thomas >>> >> >>Hi, >> if your signal's sidebands are close to >>zero Hz (DC), there will be spectral >>wraparound (overlap) at zero Hz that might >>make your sidelobes look asymmetrical when they're >>really not! >> >>Thomas, I'll bet someone here can help you but you'll >>have to clearly explain what "frequency distortion of >>the carrier" means. Also you'll have to explain what >>you mean by "by taking the average frequency of the >>power spectrum". Do you mean computing the >>average of a series of carrier frequency measurements >>obtained from a series of FFTs? >> >>[-Rick-]
Reply by ●August 24, 20042004-08-24
Reading sub-Hz is easy if you don't use the FFT or the Fourier series. Just discretely transform your samples. I have no problem accurately measuring the frequency of a CW at a resolution of .0001 Hz using an 8bit ADC and a 100mS sample period. It's the modulation that is screwing me up. "Jerry Avins" <jya@ieee.org> wrote in message news:412b9650$0$21748$61fed72c@news.rcn.com...> Thomas Magma wrote: > > How do you read one Hz? What is your bin spacing? Are you accounting for > how a signal not exactly at a bin center "bleeds" into adjacent (and > more distant) bins? Do you shape the window to minimize that? > > I haven't been following closely enough to remember if those issues have > already been addressed. If they were, please pardon me. > > Jerry > -- > Engineering is the art of making what you want from things you can get. > ����������������������������������������������������������������������� > Thomas Magma wrote: > > > "frequency distortion of the carrier" I meant that the carrier'sfrequency> > appears to be offset from where I know it should be, by up to a Hz orso,> > depending on what information data is present. I'm currently workingwith> > sampled data generated by a computer. I did this to remove any unseen > > hardware variables and so I knew exactly what frequency the carriershould> > appear at. I will then re-introduce the hardware later when I'm happywith> > my algorithm. > > > > It turns out that I was right. You can reconstruct the carriers true > > frequency by taking the average frequency of the power spectrum centered > > around the offset carrier. Here is the formula how: SUM(f_i * > > mag(i)/SUM(mag(i)) > > > > The method works great but takes too much processing power for my > > application, so I can't really use it. > > > > So back to the drawing board. > > > > Thomas > > > > > > > > > > > > "Rick Lyons" <r.lyons@_BOGUS_ieee.org> wrote in message > > news:41272fb0.1387517671@news.sf.sbcglobal.net... > > > >>On Fri, 20 Aug 2004 17:05:42 GMT, "Thomas Magma" > >><somewhere@overtherainbow.com> wrote: > >> > >> > >>>I am trying to accurately recover the carrier component of a sampled > >>>modulated RF signal. When I transform a sample window I see my carrier > > > > and > > > >>>it's information side bands. The side bands appear to be at different > >>>amplitudes (non symmetrical) around the carrier and the carriers > > > > frequency > > > >>>seems to be distorted a few Hz. This frequency distortion of thecarrier> > > > and > > > >>>what appears to be single side band suppression is depended on what > > > > random > > > >>>data happens to be in the sample window at the time I sample. Itappears> >>>that I might be able to recover the true frequency of the carrier by > > > > taking > > > >>>the average frequency of the power spectrum. I just don't know how todo> >>>this yet ....pivot table maybe? > >>> > >>>Any thoughts? > >>> > >>>Thomas > >>> > >> > >>Hi, > >> if your signal's sidebands are close to > >>zero Hz (DC), there will be spectral > >>wraparound (overlap) at zero Hz that might > >>make your sidelobes look asymmetrical when they're > >>really not! > >> > >>Thomas, I'll bet someone here can help you but you'll > >>have to clearly explain what "frequency distortion of > >>the carrier" means. Also you'll have to explain what > >>you mean by "by taking the average frequency of the > >>power spectrum". Do you mean computing the > >>average of a series of carrier frequency measurements > >>obtained from a series of FFTs? > >> > >>[-Rick-] >
Reply by ●August 24, 20042004-08-24
Thomas Magma wrote:> Reading sub-Hz is easy if you don't use the FFT or the Fourier series. Just > discretely transform your samples. I have no problem accurately measuring > the frequency of a CW at a resolution of .0001 Hz using an 8bit ADC and a > 100mS sample period. It's the modulation that is screwing me up.That figures. The carrier alone lets you take advantage of knowing that you're dealing with a single frequency in rather straightforward ways. Modulation adds sidebands, which among other effects, can move the zero crossings around. When that happens single-frequency methods fail. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●August 24, 20042004-08-24
That should read "100mS sample window", not period. I am sampling at around 1 Mhz or so. "Thomas Magma" <somewhere@overtherainbow.com> wrote in message news:IDNWc.209441$gE.169449@pd7tw3no...> Reading sub-Hz is easy if you don't use the FFT or the Fourier series.Just> discretely transform your samples. I have no problem accurately measuring > the frequency of a CW at a resolution of .0001 Hz using an 8bit ADC and a > 100mS sample period. It's the modulation that is screwing me up. > > "Jerry Avins" <jya@ieee.org> wrote in message > news:412b9650$0$21748$61fed72c@news.rcn.com... > > Thomas Magma wrote: > > > > How do you read one Hz? What is your bin spacing? Are you accounting for > > how a signal not exactly at a bin center "bleeds" into adjacent (and > > more distant) bins? Do you shape the window to minimize that? > > > > I haven't been following closely enough to remember if those issues have > > already been addressed. If they were, please pardon me. > > > > Jerry > > -- > > Engineering is the art of making what you want from things you can get. > > ����������������������������������������������������������������������� > > Thomas Magma wrote: > > > > > "frequency distortion of the carrier" I meant that the carrier's > frequency > > > appears to be offset from where I know it should be, by up to a Hz or > so, > > > depending on what information data is present. I'm currently working > with > > > sampled data generated by a computer. I did this to remove any unseen > > > hardware variables and so I knew exactly what frequency the carrier > should > > > appear at. I will then re-introduce the hardware later when I'm happy > with > > > my algorithm. > > > > > > It turns out that I was right. You can reconstruct the carriers true > > > frequency by taking the average frequency of the power spectrumcentered> > > around the offset carrier. Here is the formula how: SUM(f_i * > > > mag(i)/SUM(mag(i)) > > > > > > The method works great but takes too much processing power for my > > > application, so I can't really use it. > > > > > > So back to the drawing board. > > > > > > Thomas > > > > > > > > > > > > > > > > > > "Rick Lyons" <r.lyons@_BOGUS_ieee.org> wrote in message > > > news:41272fb0.1387517671@news.sf.sbcglobal.net... > > > > > >>On Fri, 20 Aug 2004 17:05:42 GMT, "Thomas Magma" > > >><somewhere@overtherainbow.com> wrote: > > >> > > >> > > >>>I am trying to accurately recover the carrier component of a sampled > > >>>modulated RF signal. When I transform a sample window I see mycarrier> > > > > > and > > > > > >>>it's information side bands. The side bands appear to be at different > > >>>amplitudes (non symmetrical) around the carrier and the carriers > > > > > > frequency > > > > > >>>seems to be distorted a few Hz. This frequency distortion of the > carrier > > > > > > and > > > > > >>>what appears to be single side band suppression is depended on what > > > > > > random > > > > > >>>data happens to be in the sample window at the time I sample. It > appears > > >>>that I might be able to recover the true frequency of the carrier by > > > > > > taking > > > > > >>>the average frequency of the power spectrum. I just don't know how to > do > > >>>this yet ....pivot table maybe? > > >>> > > >>>Any thoughts? > > >>> > > >>>Thomas > > >>> > > >> > > >>Hi, > > >> if your signal's sidebands are close to > > >>zero Hz (DC), there will be spectral > > >>wraparound (overlap) at zero Hz that might > > >>make your sidelobes look asymmetrical when they're > > >>really not! > > >> > > >>Thomas, I'll bet someone here can help you but you'll > > >>have to clearly explain what "frequency distortion of > > >>the carrier" means. Also you'll have to explain what > > >>you mean by "by taking the average frequency of the > > >>power spectrum". Do you mean computing the > > >>average of a series of carrier frequency measurements > > >>obtained from a series of FFTs? > > >> > > >>[-Rick-] > > > >
