Hi, I am developing a QPSK receiver (as an exercise), and after reading a lots of chapters of coarse synchronization algorithms for frequency and timing offset adquisition, I do not know what is done first, frequency or timing. I have choosen pilot aided techniques which use preambles for fast adquisition. I found pilot aided algoritms for timing sync that suppose 0 Hz frequency offset, and the same for frequency sync algorithms. So, what is done in the first stage of the received samples? Any advice? Thanks in advance.
Frequency and timing adquisition
Started by ●November 6, 2011
Reply by ●November 6, 20112011-11-06
>Hi, > > I am developing a QPSK receiver (as an exercise), and after reading alots>of chapters of coarse synchronization algorithms for frequency and timing >offset adquisition, I do not know what is done first, frequency ortiming.>I have choosen pilot aided techniques which use preambles for fast >adquisition. I found pilot aided algoritms for timing sync that suppose 0 >Hz frequency offset, and the same for frequency sync algorithms. So, what >is done in the first stage of the received samples? Any advice? > >Thanks in advance.Symbol timing recovery is usually the topic handled most poorly in books on communication. There are simple symbol timing recovery algorithms, like Gardner's, which perform well for QPSK, and work before carrier recovery. For more complex modulation schemes other pre-carrier recovery algorithms, like Godard's, exist. Once you have locked to the symbol timing, you need to recover the carrier. Most preambles begin with a period of alternations between two symbol states. If you have the symbol timing locked, you should be able to track the phase rotations of successive recovered symbols in order to correct that rotation and lock down the phase. Now, as long as you have something distinctive happen at the end of the preamble's initial phase alternations, so you can switch to the next phase of signal recovery at exactly the right symbol, you should be OK. The next phase of recovery is typically equaliser training. Steve
Reply by ●November 7, 20112011-11-07
On Sun, 06 Nov 2011 18:51:11 -0600, "nahemoth" <nahemoth@n_o_s_p_a_m.gmail.com> wrote:>Hi, > > I am developing a QPSK receiver (as an exercise), and after reading a lots >of chapters of coarse synchronization algorithms for frequency and timing >offset adquisition, I do not know what is done first, frequency or timing. >I have choosen pilot aided techniques which use preambles for fast >adquisition. I found pilot aided algoritms for timing sync that suppose 0 >Hz frequency offset, and the same for frequency sync algorithms. So, what >is done in the first stage of the received samples? Any advice? > >Thanks in advance.There are lots of ways to do it depending on the signal type, the symbol rate, the expected channel conditions, etc., etc. A well-designed preamble accommodates the needs of the system. I've worked with preambles that are designed to allow signal detection and frequency correction first, before timing correction, and others that facilitate timing correction first before frequency correction. In the first case the expected frequency error was large compared to the signal bandwidth and needed to be removed first, but that's unusual. Usually timing is done first, but, it does depend on the system. The 802.11a preamble is well documented and is a good example of a well-designed preamble. The first part of the preamble allows timing estimation as well as coarse frequency estimation. The idea is to get things close enough that the remaining (long symbol) part of the preamble can be used for fine timing adjustment, fine frequency adjustment, and channel estimation. You might do a search on 802.11 preamble topics as there's been a lot of methodology published around those. Eric Jacobsen Anchor Hill Communications www.anchorhill.com
Reply by ●November 7, 20112011-11-07
nahemoth wrote:> Hi, > > I am developing a QPSK receiver (as an exercise), and after reading a lots > of chapters of coarse synchronization algorithms for frequency and timing > offset adquisition, I do not know what is done first, frequency or timing.For exercise purpose, use Costas and Gardner algorithms for carrier and symbol sync. Those are simple and could be applied in either sequence. Usually, carrier sync goes first as it has to deal with large offsets.> I have choosen pilot aided techniques which use preambles for fast > adquisition. I found pilot aided algoritms for timing sync that suppose 0 > Hz frequency offset, and the same for frequency sync algorithms. So, what > is done in the first stage of the received samples? Any advice?Too much of knowledge is harmful to fools. VLV
Reply by ●November 7, 20112011-11-07
Thank you very much Steve and Eric, it has been very useful your help. I was thinking in a QPSK receiver for 1 Mbaud/s with a preamble on each transmitted data packet of finite duration (not continuous broadcasting). I read that Gardner timing recovery algorithm needs around 500 symbols to converge, which I think it is too much. In this situation what might I do? Thanks
Reply by ●November 7, 20112011-11-07
On Mon, 07 Nov 2011 14:11:44 -0600, "nahemoth" <nahemoth@n_o_s_p_a_m.gmail.com> wrote:>Thank you very much Steve and Eric, it has been very useful your help. > >I was thinking in a QPSK receiver for 1 Mbaud/s with a preamble on each >transmitted data packet of finite duration (not continuous broadcasting). I >read that Gardner timing recovery algorithm needs around 500 symbols to >converge, which I think it is too much. In this situation what might I do? > >ThanksFor burst acquisition the preamble needs to provide adequate timing registration since, as you mention, there isn't time to wait for a loop to converge. There are a lot of things that can affect how you might approach this, and a lot of different knobs that can be turned that will affect complexity and performance. Depending on how bad the channel is (and other factors), the preamble might be anything from a long training sequence to just a Unique Word. For the cases where just a UW might be used, the usual acquisition system is just a bank of N correlators that look for the UW with N times oversampling on the symbols. The one that triggers with the largest output magnitude will be most likely sample closest to the proper timing instance. If the channel is especially noisy or has a large delay spread or has other impairments, the UW correlator alone may not work or may get too complex to get to work properly. In that case making the preamble longer to try to sort things out may help. Adjusting the pulse shaping to make the "eye opening" of the symbol as broad as possible will also help mitigate the effects of getting the timing a little bit wrong, but there's a tradeoff in the occupied spectrum of the signal. Eric Jacobsen Anchor Hill Communications www.anchorhill.com
Reply by ●November 7, 20112011-11-07
>On Mon, 07 Nov 2011 14:11:44 -0600, "nahemoth" ><nahemoth@n_o_s_p_a_m.gmail.com> wrote: > >>Thank you very much Steve and Eric, it has been very useful your help. >> >>I was thinking in a QPSK receiver for 1 Mbaud/s with a preamble on each >>transmitted data packet of finite duration (not continuous broadcasting).I>>read that Gardner timing recovery algorithm needs around 500 symbols to >>converge, which I think it is too much. In this situation what might Ido?>> >>Thanks > >For burst acquisition the preamble needs to provide adequate timing >registration since, as you mention, there isn't time to wait for a >loop to converge. There are a lot of things that can affect how you >might approach this, and a lot of different knobs that can be turned >that will affect complexity and performance. > >Depending on how bad the channel is (and other factors), the preamble >might be anything from a long training sequence to just a Unique Word. >For the cases where just a UW might be used, the usual acquisition >system is just a bank of N correlators that look for the UW with N >times oversampling on the symbols. The one that triggers with the >largest output magnitude will be most likely sample closest to the >proper timing instance. > >If the channel is especially noisy or has a large delay spread or has >other impairments, the UW correlator alone may not work or may get too >complex to get to work properly. In that case making the preamble >longer to try to sort things out may help. > >Adjusting the pulse shaping to make the "eye opening" of the symbol as >broad as possible will also help mitigate the effects of getting the >timing a little bit wrong, but there's a tradeoff in the occupied >spectrum of the signal. > > >Eric Jacobsen >Anchor Hill Communications >www.anchorhill.com >I found this paper which deals with UWB signal, but I suppose that it can be useful any PSK signal: Liuqing Yang; Giannakis, G.B.; , "Low-complexity training for rapid timing acquisition in ultra wideband communications," Global Telecommunications Conference, 2003. GLOBECOM '03. IEEE , vol.2, no., pp. 769- 773 Vol.2, 1-5 Dec. 2003 Do the TDMA systems use this kind of burst timing acquisition?
Reply by ●November 8, 20112011-11-08
>>On Mon, 07 Nov 2011 14:11:44 -0600, "nahemoth" >><nahemoth@n_o_s_p_a_m.gmail.com> wrote: >> >>>Thank you very much Steve and Eric, it has been very useful your help. >>> >>>I was thinking in a QPSK receiver for 1 Mbaud/s with a preamble on each >>>transmitted data packet of finite duration (not continuousbroadcasting).>I >>>read that Gardner timing recovery algorithm needs around 500 symbols to >>>converge, which I think it is too much. In this situation what might I >do? >>> >>>Thanks >> >>For burst acquisition the preamble needs to provide adequate timing >>registration since, as you mention, there isn't time to wait for a >>loop to converge. There are a lot of things that can affect how you >>might approach this, and a lot of different knobs that can be turned >>that will affect complexity and performance. >> >>Depending on how bad the channel is (and other factors), the preamble >>might be anything from a long training sequence to just a Unique Word. >>For the cases where just a UW might be used, the usual acquisition >>system is just a bank of N correlators that look for the UW with N >>times oversampling on the symbols. The one that triggers with the >>largest output magnitude will be most likely sample closest to the >>proper timing instance. >> >>If the channel is especially noisy or has a large delay spread or has >>other impairments, the UW correlator alone may not work or may get too >>complex to get to work properly. In that case making the preamble >>longer to try to sort things out may help. >> >>Adjusting the pulse shaping to make the "eye opening" of the symbol as >>broad as possible will also help mitigate the effects of getting the >>timing a little bit wrong, but there's a tradeoff in the occupied >>spectrum of the signal. >> >> >>Eric Jacobsen >>Anchor Hill Communications >>www.anchorhill.com >> > >I found this paper which deals with UWB signal, but I suppose that it can >be useful any PSK signal: > >Liuqing Yang; Giannakis, G.B.; , "Low-complexity training for rapidtiming>acquisition in ultra wideband communications," Global Telecommunications >Conference, 2003. GLOBECOM '03. IEEE , vol.2, no., pp. 769- 773 Vol.2,1-5>Dec. 2003 > >Do the TDMA systems use this kind of burst timing acquisition? > > >Hi, any sugestion for preamble selection? Eric said to investigate with 802.11a preamble, but it is for OFDM signals. Barker sequences are a good choice? Thanks.
Reply by ●November 8, 20112011-11-08
On Tue, 08 Nov 2011 14:00:27 -0600, "nahemoth" <nahemoth@n_o_s_p_a_m.gmail.com> wrote:>>>On Mon, 07 Nov 2011 14:11:44 -0600, "nahemoth" >>><nahemoth@n_o_s_p_a_m.gmail.com> wrote: >>> >>>>Thank you very much Steve and Eric, it has been very useful your help. >>>> >>>>I was thinking in a QPSK receiver for 1 Mbaud/s with a preamble on each >>>>transmitted data packet of finite duration (not continuous >broadcasting). >>I >>>>read that Gardner timing recovery algorithm needs around 500 symbols to >>>>converge, which I think it is too much. In this situation what might I >>do? >>>> >>>>Thanks >>> >>>For burst acquisition the preamble needs to provide adequate timing >>>registration since, as you mention, there isn't time to wait for a >>>loop to converge. There are a lot of things that can affect how you >>>might approach this, and a lot of different knobs that can be turned >>>that will affect complexity and performance. >>> >>>Depending on how bad the channel is (and other factors), the preamble >>>might be anything from a long training sequence to just a Unique Word. >>>For the cases where just a UW might be used, the usual acquisition >>>system is just a bank of N correlators that look for the UW with N >>>times oversampling on the symbols. The one that triggers with the >>>largest output magnitude will be most likely sample closest to the >>>proper timing instance. >>> >>>If the channel is especially noisy or has a large delay spread or has >>>other impairments, the UW correlator alone may not work or may get too >>>complex to get to work properly. In that case making the preamble >>>longer to try to sort things out may help. >>> >>>Adjusting the pulse shaping to make the "eye opening" of the symbol as >>>broad as possible will also help mitigate the effects of getting the >>>timing a little bit wrong, but there's a tradeoff in the occupied >>>spectrum of the signal. >>> >>> >>>Eric Jacobsen >>>Anchor Hill Communications >>>www.anchorhill.com >>> >> >>I found this paper which deals with UWB signal, but I suppose that it can >>be useful any PSK signal: >> >>Liuqing Yang; Giannakis, G.B.; , "Low-complexity training for rapid >timing >>acquisition in ultra wideband communications," Global Telecommunications >>Conference, 2003. GLOBECOM '03. IEEE , vol.2, no., pp. 769- 773 Vol.2, >1-5 >>Dec. 2003 >> >>Do the TDMA systems use this kind of burst timing acquisition? >> >> >> > >Hi, > > any sugestion for preamble selection? Eric said to investigate with >802.11a preamble, but it is for OFDM signals. Barker sequences are a good >choice? > >Thanks.The 802.11a preamble is an example of how constructing the preamble helps acquisition. A preamble for a single-carrier system will be different, but the basic idea of accommodating timing and frequency synchronization, plus channel estimation, using features of the preamble is similar. There are many types of TDMA systems that have different characteristics and constraints. A satellite system will likely have a different preamble configuration than a mobile terrestrial cellular system, for example. You might start from your requirements and then see what similar systems use. Eric Jacobsen Anchor Hill Communications www.anchorhill.com
Reply by ●November 9, 20112011-11-09
>On Tue, 08 Nov 2011 14:00:27 -0600, "nahemoth" ><nahemoth@n_o_s_p_a_m.gmail.com> wrote: > >>>>On Mon, 07 Nov 2011 14:11:44 -0600, "nahemoth" >>>><nahemoth@n_o_s_p_a_m.gmail.com> wrote: >>>> >>>>>Thank you very much Steve and Eric, it has been very useful your help.>>>>> >>>>>I was thinking in a QPSK receiver for 1 Mbaud/s with a preamble oneach>>>>>transmitted data packet of finite duration (not continuous >>broadcasting). >>>I >>>>>read that Gardner timing recovery algorithm needs around 500 symbolsto>>>>>converge, which I think it is too much. In this situation what mightI>>>do? >>>>> >>>>>Thanks >>>> >>>>For burst acquisition the preamble needs to provide adequate timing >>>>registration since, as you mention, there isn't time to wait for a >>>>loop to converge. There are a lot of things that can affect how you >>>>might approach this, and a lot of different knobs that can be turned >>>>that will affect complexity and performance. >>>> >>>>Depending on how bad the channel is (and other factors), the preamble >>>>might be anything from a long training sequence to just a Unique Word. >>>>For the cases where just a UW might be used, the usual acquisition >>>>system is just a bank of N correlators that look for the UW with N >>>>times oversampling on the symbols. The one that triggers with the >>>>largest output magnitude will be most likely sample closest to the >>>>proper timing instance. >>>> >>>>If the channel is especially noisy or has a large delay spread or has >>>>other impairments, the UW correlator alone may not work or may get too >>>>complex to get to work properly. In that case making the preamble >>>>longer to try to sort things out may help. >>>> >>>>Adjusting the pulse shaping to make the "eye opening" of the symbol as >>>>broad as possible will also help mitigate the effects of getting the >>>>timing a little bit wrong, but there's a tradeoff in the occupied >>>>spectrum of the signal. >>>> >>>> >>>>Eric Jacobsen >>>>Anchor Hill Communications >>>>www.anchorhill.com >>>> >>> >>>I found this paper which deals with UWB signal, but I suppose that itcan>>>be useful any PSK signal: >>> >>>Liuqing Yang; Giannakis, G.B.; , "Low-complexity training for rapid >>timing >>>acquisition in ultra wideband communications," GlobalTelecommunications>>>Conference, 2003. GLOBECOM '03. IEEE , vol.2, no., pp. 769- 773 Vol.2, >>1-5 >>>Dec. 2003 >>> >>>Do the TDMA systems use this kind of burst timing acquisition? >>> >>> >>> >> >>Hi, >> >> any sugestion for preamble selection? Eric said to investigate with >>802.11a preamble, but it is for OFDM signals. Barker sequences are agood>>choice? >> >>Thanks. > >The 802.11a preamble is an example of how constructing the preamble >helps acquisition. A preamble for a single-carrier system will be >different, but the basic idea of accommodating timing and frequency >synchronization, plus channel estimation, using features of the >preamble is similar. > >There are many types of TDMA systems that have different >characteristics and constraints. A satellite system will likely have >a different preamble configuration than a mobile terrestrial cellular >system, for example. You might start from your requirements and then >see what similar systems use. > > >Eric Jacobsen >Anchor Hill Communications >www.anchorhill.com >I found this interesting paper (http://cegt201.bradley.edu/projects/proj2006/sradio/archives/Timing_sync_802_wlan_multipath.pdf) for timing synchronization using the 802.11a short preambles, but I think that it can be reused for single-carrier systems as well. I will try to simulate it.
