Hi, I want to recover the carrier frequency offset in a QPSK receiver for burst transmissions, but the normalized CFO with respect the symbol time is high (>0.15). Therefore, I need a coarse estimation of the CFO before the timing recovery is done (using a preamble). I read about algorithms based on FFT, but is there any other method for a coarse estimation? Fine tuning can be done with a PLL. Thanks in advance.
Frequency acquisition without timing information
Started by ●January 4, 2012
Reply by ●January 4, 20122012-01-04
On Wed, 04 Jan 2012 16:48:48 -0600, "nahemoth" <nahemoth@n_o_s_p_a_m.gmail.com> wrote:>Hi, > > I want to recover the carrier frequency offset in a QPSK receiver for >burst transmissions, but the normalized CFO with respect the symbol time is >high (>0.15). Therefore, I need a coarse estimation of the CFO before the >timing recovery is done (using a preamble). I read about algorithms based >on FFT, but is there any other method for a coarse estimation? Fine tuning >can be done with a PLL. > >Thanks in advance.In order to remove frequency offset without timing information you need the portion of the preamble used for correcting frequency to be insensitive to timing errors. A CW works for this, or, if you want to use some modulated signal, a short repeating sequence of signal chunks that have good autocorrelation properties. A CW allows the use of an FFT or PLL directly. The use of repeating sequences requires a couple of correlators. Which will suit the application best depends on a lot of things not mentioned. Eric Jacobsen Anchor Hill Communications www.anchorhill.com
Reply by ●January 5, 20122012-01-05
>On Wed, 04 Jan 2012 16:48:48 -0600, "nahemoth" ><nahemoth@n_o_s_p_a_m.gmail.com> wrote: > >>Hi, >> >> I want to recover the carrier frequency offset in a QPSK receiver for >>burst transmissions, but the normalized CFO with respect the symbol timeis>>high (>0.15). Therefore, I need a coarse estimation of the CFO beforethe>>timing recovery is done (using a preamble). I read about algorithmsbased>>on FFT, but is there any other method for a coarse estimation? Finetuning>>can be done with a PLL. >> >>Thanks in advance. > >In order to remove frequency offset without timing information you >need the portion of the preamble used for correcting frequency to be >insensitive to timing errors. A CW works for this, or, if you want >to use some modulated signal, a short repeating sequence of signal >chunks that have good autocorrelation properties. > >A CW allows the use of an FFT or PLL directly. The use of repeating >sequences requires a couple of correlators. Which will suit the >application best depends on a lot of things not mentioned. > > > > >Eric Jacobsen >Anchor Hill Communications >www.anchorhill.com >Hi Eric I do not need a precise frequency offset estimation, because I calculate it again after timing synchronization jointly with the frame start (UW) using differential modulation. Thanks
Reply by ●January 5, 20122012-01-05
>>On Wed, 04 Jan 2012 16:48:48 -0600, "nahemoth" >><nahemoth@n_o_s_p_a_m.gmail.com> wrote: >> >>>Hi, >>> >>> I want to recover the carrier frequency offset in a QPSK receiver for >>>burst transmissions, but the normalized CFO with respect the symboltime>is >>>high (>0.15). Therefore, I need a coarse estimation of the CFO before >the >>>timing recovery is done (using a preamble). I read about algorithms >based >>>on FFT, but is there any other method for a coarse estimation? Fine >tuning >>>can be done with a PLL. >>> >>>Thanks in advance. >> >>In order to remove frequency offset without timing information you >>need the portion of the preamble used for correcting frequency to be >>insensitive to timing errors. A CW works for this, or, if you want >>to use some modulated signal, a short repeating sequence of signal >>chunks that have good autocorrelation properties. >> >>A CW allows the use of an FFT or PLL directly. The use of repeating >>sequences requires a couple of correlators. Which will suit the >>application best depends on a lot of things not mentioned. >> >> >> >> >>Eric Jacobsen >>Anchor Hill Communications >>www.anchorhill.com >> > >Hi Eric > >I do not need a precise frequency offset estimation, because I calculateit>again after timing synchronization jointly with the frame start (UW)using>differential modulation. > >Thanks > >I have another question... just after the unknown parameters (timing, frequency and phase) have been estimated, and supposing that the channel suffers from multipath, is this the correct place to start the training of the equalizer? Or are there other approaches? Thanks in advance
Reply by ●January 5, 20122012-01-05
On Thu, 05 Jan 2012 11:31:26 -0600, "nahemoth" <nahemoth@n_o_s_p_a_m.gmail.com> wrote:>>>On Wed, 04 Jan 2012 16:48:48 -0600, "nahemoth" >>><nahemoth@n_o_s_p_a_m.gmail.com> wrote: >>> >>>>Hi, >>>> >>>> I want to recover the carrier frequency offset in a QPSK receiver for >>>>burst transmissions, but the normalized CFO with respect the symbol >time >>is >>>>high (>0.15). Therefore, I need a coarse estimation of the CFO before >>the >>>>timing recovery is done (using a preamble). I read about algorithms >>based >>>>on FFT, but is there any other method for a coarse estimation? Fine >>tuning >>>>can be done with a PLL. >>>> >>>>Thanks in advance. >>> >>>In order to remove frequency offset without timing information you >>>need the portion of the preamble used for correcting frequency to be >>>insensitive to timing errors. A CW works for this, or, if you want >>>to use some modulated signal, a short repeating sequence of signal >>>chunks that have good autocorrelation properties. >>> >>>A CW allows the use of an FFT or PLL directly. The use of repeating >>>sequences requires a couple of correlators. Which will suit the >>>application best depends on a lot of things not mentioned. >>> >>> >>> >>> >>>Eric Jacobsen >>>Anchor Hill Communications >>>www.anchorhill.com >>> >> >>Hi Eric >> >>I do not need a precise frequency offset estimation, because I calculate >it >>again after timing synchronization jointly with the frame start (UW) >using >>differential modulation. >> >>Thanks >> >> > >I have another question... just after the unknown parameters (timing, >frequency and phase) have been estimated, and supposing that the channel >suffers from multipath, is this the correct place to start the training of >the equalizer? Or are there other approaches? > >Thanks in advanceMost channel estimation and equalization strategies assume that the signal is at least reasonably well synchronized. If you can come up with or find an equalizer or channel estimator that will work without the signal being synchronized, then you'll have the option to start it prior to synchronization. It's pretty rare to find a system that doesn't synchronize prior to equalization. For single carrier systems many equalizers will try to use up some degrees of freedom by correcting syncrhonization errors, which then plays havoc with the systems that are trying to do that job. So often the EQ is built without that capability, so that syncrhonization is done only by the synchronization circuits and the EQ can focus on equalizing the signal. Eric Jacobsen Anchor Hill Communications www.anchorhill.com
Reply by ●January 7, 20122012-01-07
>On Thu, 05 Jan 2012 11:31:26 -0600, "nahemoth" ><nahemoth@n_o_s_p_a_m.gmail.com> wrote: > >>>>On Wed, 04 Jan 2012 16:48:48 -0600, "nahemoth" >>>><nahemoth@n_o_s_p_a_m.gmail.com> wrote: >>>> >>>>>Hi, >>>>> >>>>> I want to recover the carrier frequency offset in a QPSK receiverfor>>>>>burst transmissions, but the normalized CFO with respect the symbol >>time >>>is >>>>>high (>0.15). Therefore, I need a coarse estimation of the CFO before >>>the >>>>>timing recovery is done (using a preamble). I read about algorithms >>>based >>>>>on FFT, but is there any other method for a coarse estimation? Fine >>>tuning >>>>>can be done with a PLL. >>>>> >>>>>Thanks in advance. >>>> >>>>In order to remove frequency offset without timing information you >>>>need the portion of the preamble used for correcting frequency to be >>>>insensitive to timing errors. A CW works for this, or, if you want >>>>to use some modulated signal, a short repeating sequence of signal >>>>chunks that have good autocorrelation properties. >>>> >>>>A CW allows the use of an FFT or PLL directly. The use of repeating >>>>sequences requires a couple of correlators. Which will suit the >>>>application best depends on a lot of things not mentioned. >>>> >>>> >>>> >>>> >>>>Eric Jacobsen >>>>Anchor Hill Communications >>>>www.anchorhill.com >>>> >>> >>>Hi Eric >>> >>>I do not need a precise frequency offset estimation, because Icalculate>>it >>>again after timing synchronization jointly with the frame start (UW) >>using >>>differential modulation. >>> >>>Thanks >>> >>> >> >>I have another question... just after the unknown parameters (timing, >>frequency and phase) have been estimated, and supposing that the channel >>suffers from multipath, is this the correct place to start the trainingof>>the equalizer? Or are there other approaches? >> >>Thanks in advance > >Most channel estimation and equalization strategies assume that the >signal is at least reasonably well synchronized. If you can come up >with or find an equalizer or channel estimator that will work without >the signal being synchronized, then you'll have the option to start it >prior to synchronization. > >It's pretty rare to find a system that doesn't synchronize prior to >equalization. For single carrier systems many equalizers will try to >use up some degrees of freedom by correcting syncrhonization errors, >which then plays havoc with the systems that are trying to do that >job. So often the EQ is built without that capability, so that >syncrhonization is done only by the synchronization circuits and the >EQ can focus on equalizing the signal. > > > > >Eric Jacobsen >Anchor Hill Communications >www.anchorhill.com >I assume that the equalization should be done to an oversampled signal, where sampling time is lower than the symbol time, that is, an FSE. Am I correct? At the moment my signal is oversampled twice, and the frequency and phase feedforward algorithms work at the symbol rate. Another question... I have placed the receiving matched filter (root raised cosine) just in the output of the timing recovery interpolator. Is it the correct place? Thanks!
Reply by ●January 7, 20122012-01-07
> > Another question... I have placed the receiving matched filter (root raised > cosine) just in the output of the timing recovery interpolator. Is it the > correct place? > > Thanks!and I have a more basic question... if you want to recover the carrier of a QPSK signal, why can't you simply square the signal twice... doesn't that (nominally) remove the modulation (so no timing recovery needed) and leave you with 4x? the carrier? then divide by 4??? thanks Mark
Reply by ●January 7, 20122012-01-07
On Sat, 07 Jan 2012 03:09:20 -0600, "nahemoth" <nahemoth@n_o_s_p_a_m.gmail.com> wrote:>>On Thu, 05 Jan 2012 11:31:26 -0600, "nahemoth" >><nahemoth@n_o_s_p_a_m.gmail.com> wrote: >> >>>>>On Wed, 04 Jan 2012 16:48:48 -0600, "nahemoth" >>>>><nahemoth@n_o_s_p_a_m.gmail.com> wrote: >>>>> >>>>>>Hi, >>>>>> >>>>>> I want to recover the carrier frequency offset in a QPSK receiver >for >>>>>>burst transmissions, but the normalized CFO with respect the symbol >>>time >>>>is >>>>>>high (>0.15). Therefore, I need a coarse estimation of the CFO before >>>>the >>>>>>timing recovery is done (using a preamble). I read about algorithms >>>>based >>>>>>on FFT, but is there any other method for a coarse estimation? Fine >>>>tuning >>>>>>can be done with a PLL. >>>>>> >>>>>>Thanks in advance. >>>>> >>>>>In order to remove frequency offset without timing information you >>>>>need the portion of the preamble used for correcting frequency to be >>>>>insensitive to timing errors. A CW works for this, or, if you want >>>>>to use some modulated signal, a short repeating sequence of signal >>>>>chunks that have good autocorrelation properties. >>>>> >>>>>A CW allows the use of an FFT or PLL directly. The use of repeating >>>>>sequences requires a couple of correlators. Which will suit the >>>>>application best depends on a lot of things not mentioned. >>>>> >>>>> >>>>> >>>>> >>>>>Eric Jacobsen >>>>>Anchor Hill Communications >>>>>www.anchorhill.com >>>>> >>>> >>>>Hi Eric >>>> >>>>I do not need a precise frequency offset estimation, because I >calculate >>>it >>>>again after timing synchronization jointly with the frame start (UW) >>>using >>>>differential modulation. >>>> >>>>Thanks >>>> >>>> >>> >>>I have another question... just after the unknown parameters (timing, >>>frequency and phase) have been estimated, and supposing that the channel >>>suffers from multipath, is this the correct place to start the training >of >>>the equalizer? Or are there other approaches? >>> >>>Thanks in advance >> >>Most channel estimation and equalization strategies assume that the >>signal is at least reasonably well synchronized. If you can come up >>with or find an equalizer or channel estimator that will work without >>the signal being synchronized, then you'll have the option to start it >>prior to synchronization. >> >>It's pretty rare to find a system that doesn't synchronize prior to >>equalization. For single carrier systems many equalizers will try to >>use up some degrees of freedom by correcting syncrhonization errors, >>which then plays havoc with the systems that are trying to do that >>job. So often the EQ is built without that capability, so that >>syncrhonization is done only by the synchronization circuits and the >>EQ can focus on equalizing the signal. >> >> >> >> >>Eric Jacobsen >>Anchor Hill Communications >>www.anchorhill.com >> > >I assume that the equalization should be done to an oversampled signal, >where sampling time is lower than the symbol time, that is, an FSE. Am I >correct? At the moment my signal is oversampled twice, and the frequency >and phase feedforward algorithms work at the symbol rate.Depending on the channel and the system an FSE often isn't necessary. Whether it is or not in your case I don't know.>Another question... I have placed the receiving matched filter (root raised >cosine) just in the output of the timing recovery interpolator. Is it the >correct place?It'll work there. Remember that linear functions can be done in any order, so as long as you're not doing any non-linear operations in between the linear ones their order can be changed. Eric Jacobsen Anchor Hill Communications www.anchorhill.com
Reply by ●January 8, 20122012-01-08
REPOST (some people don't see Google Groups) and I have a more basic question... if you want to recover the carrier of a QPSK signal, why can't you simply square the signal twice... doesn't that (nominally) remove the modulation (so no timing recovery needed) and leave you with 4x? the carrier? then divide by 4??? thanks Mark
Reply by ●January 8, 20122012-01-08
On Sun, 8 Jan 2012 12:46:11 -0500, "MarkK" <makolber@yahoo.com> wrote:>REPOST (some people don't see Google Groups) > >and I have a more basic question... if you want to recover the carrier >of a QPSK signal, why can't you simply square the signal twice... > >doesn't that (nominally) remove the modulation (so no timing recovery >needed) and leave you with 4x? the carrier? > >then divide by 4??? > >thanks > >MarkThere's a paper commonly referred to as Viterbi and Viterbi (by Andrew and his daughter), that shows how to do that. It works, but it doesn't work as well as a PLL, and works best with differentially-encoded modulation. Remember that raising the symbol to an exponent amplifies the noise more than the signal, so the derived reference generated by the nonlinearity can get pretty noisy. Nevertheless, it does have its applications. The performance lands about halfway between coherent demodulation and differential detection, so it is useful when needed. Let me know if you need help with this, I've used it a lot. Andrew.J.Viterbi and Audrey.J.Viterbi, "Nonlinear Estimation of PSK-Modulated Carrier phase with Application to Burst Digital Transmission", IEEE. Trans. Information Theory,VOL,IT-29,No.4, July 1983. Eric Jacobsen Anchor Hill Communications www.anchorhill.com
