Hi all, As we know the burst size is defined in GSM/EDGE mobile system as 156.25 symbols which takes 577 micro secs. This helps us to update our "Equalizer" burst by burst. The symbol synchronization is done per burst basis-meaning we calculate the time of arrival of a burst correlating known training sequence with received samples.After calculating the actual time offset we adjust start of a burst and then subsequent other processes are carried forward. As 802.11a specifies that burst size for a OFDM based PHY burst is of variable size. Say, It starts with 10 short OFDM preambles,2 long OFDM preambles,1 Signal OFDM symbol and payload OFDM data symbols of variable length .We use both short symbols and long symbols for robust double correlation based timing sychronization. Now I am wondering if we donot know the burst size in advance(since it is variable) how to get burst by burst samples in the routine to carry out time sychronization. A solution could be to use a buffer size of (what length?). Now how can we search(using known short preamble) through the buffer to identify for peaks ? If the buffer is too big the complexity will be very high. So what should be the optimum buffer size? or any other way to carry out time synchronization from continuous stream(I thought it could be difficult!!). I googled without luck. I know there are guys who can answer it instantly- that's why it is my fabourite group. Regards, Santosh
OFDM burst Size!!
Started by ●July 24, 2003
Reply by ●July 24, 20032003-07-24
You don't have to buffer the whole burst and search for the proper synchronization. Use a conventional early-late algorithm and the PLL. Vladimir Vassilevsky, Ph.D. DSP and Mixed Signal Design Consultant http://www.abvolt.com santosh nath wrote:> > Hi all, > > As we know the burst size is defined in GSM/EDGE mobile system as > 156.25 symbols > which takes 577 micro secs. This helps us to update our "Equalizer" > burst by burst. The symbol synchronization is done per burst > basis-meaning we calculate the time of arrival of a burst correlating > known training sequence with received samples.After calculating the > actual time offset we adjust start of a burst and then subsequent > other processes are carried forward. > > As 802.11a specifies that burst size for a OFDM based PHY burst is of > variable > size. Say, It starts with 10 short OFDM preambles,2 long OFDM > preambles,1 Signal OFDM symbol and payload OFDM data symbols of > variable length .We use both short symbols and long symbols for robust > double correlation based timing sychronization. > > Now I am wondering if we donot know the burst size in advance(since it > is variable) how to get burst by burst samples in the routine to > carry out time sychronization. A solution could be to use a buffer > size of (what length?). > > Now how can we search(using known short preamble) through the buffer > to identify for peaks ? If the buffer is too big the complexity will > be very high. > > So what should be the optimum buffer size? or any other way to carry > out time synchronization from continuous stream(I thought it could be > difficult!!). > > I googled without luck. I know there are guys who can answer it > instantly- that's why it is my fabourite group. > > Regards, > Santosh
Reply by ●July 25, 20032003-07-25
Vladimir Vassilevsky <vlv@abvolt.com> wrote in message news:<3F209773.B4327435@abvolt.com>...> You don't have to buffer the whole burst and search for the proper > synchronization. Use a conventional early-late algorithm and the PLL.Are you also talking about programable synthesized PLL with the following components: R divder,N diverder with A & B programable pre-scalar,Phase Detector,Loop filter and VCO. We are aiming at total software solution, probably PLL requires extra HW ckt. Could you see how it could be done with correlator. All popular papers in IEEE uses the correlation with ML estimation. - unfortunately I found them to mention total number of samples like 1000000 as a simulation parameter but nobody mention about buffering or detect burst by burst which could have been much easier to deal the problem. Can anybody ref. a paper? santosh> > Vladimir Vassilevsky, Ph.D. > > DSP and Mixed Signal Design Consultant > > http://www.abvolt.com > > > santosh nath wrote: > > > > Hi all, > > > > As we know the burst size is defined in GSM/EDGE mobile system as > > 156.25 symbols > > which takes 577 micro secs. This helps us to update our "Equalizer" > > burst by burst. The symbol synchronization is done per burst > > basis-meaning we calculate the time of arrival of a burst correlating > > known training sequence with received samples.After calculating the > > actual time offset we adjust start of a burst and then subsequent > > other processes are carried forward. > > > > As 802.11a specifies that burst size for a OFDM based PHY burst is of > > variable > > size. Say, It starts with 10 short OFDM preambles,2 long OFDM > > preambles,1 Signal OFDM symbol and payload OFDM data symbols of > > variable length .We use both short symbols and long symbols for robust > > double correlation based timing sychronization. > > > > Now I am wondering if we donot know the burst size in advance(since it > > is variable) how to get burst by burst samples in the routine to > > carry out time sychronization. A solution could be to use a buffer > > size of (what length?). > > > > Now how can we search(using known short preamble) through the buffer > > to identify for peaks ? If the buffer is too big the complexity will > > be very high. > > > > So what should be the optimum buffer size? or any other way to carry > > out time synchronization from continuous stream(I thought it could be > > difficult!!). > > > > I googled without luck. I know there are guys who can answer it > > instantly- that's why it is my fabourite group. > > > > Regards, > > Santosh
Reply by ●July 26, 20032003-07-26
Santosh, You're aiming at a total software solution for 802.11a? Is your intent to actually implement a real-time usable system this way? I think you're going to have trouble meeting the SIFS timing requirement for turnaround to generate an ACK on a received packet. I don't think anybody makes processors fast enough to do that in software. A lot of people have trouble meeting timing with hardware solutions. Cheers, Eric On 25 Jul 2003 04:39:39 -0700, santosh.nath@ntlworld.com (santosh nath) wrote:>Vladimir Vassilevsky <vlv@abvolt.com> wrote in message news:<3F209773.B4327435@abvolt.com>... >> You don't have to buffer the whole burst and search for the proper >> synchronization. Use a conventional early-late algorithm and the PLL. > >Are you also talking about programable synthesized PLL with the >following components: > >R divder,N diverder with A & B programable pre-scalar,Phase >Detector,Loop filter and VCO. > >We are aiming at total software solution, probably PLL requires extra >HW ckt. Could you see how it could be done with correlator. All >popular papers in IEEE uses the correlation with ML estimation. - >unfortunately I found them to mention total number of samples like >1000000 as a simulation parameter but nobody mention about buffering >or detect burst by burst which could have been much easier to deal the >problem. Can anybody ref. a paper? >santosh > >> >> Vladimir Vassilevsky, Ph.D. >> >> DSP and Mixed Signal Design Consultant >> >> http://www.abvolt.com >> >> >> santosh nath wrote: >> > >> > Hi all, >> > >> > As we know the burst size is defined in GSM/EDGE mobile system as >> > 156.25 symbols >> > which takes 577 micro secs. This helps us to update our "Equalizer" >> > burst by burst. The symbol synchronization is done per burst >> > basis-meaning we calculate the time of arrival of a burst correlating >> > known training sequence with received samples.After calculating the >> > actual time offset we adjust start of a burst and then subsequent >> > other processes are carried forward. >> > >> > As 802.11a specifies that burst size for a OFDM based PHY burst is of >> > variable >> > size. Say, It starts with 10 short OFDM preambles,2 long OFDM >> > preambles,1 Signal OFDM symbol and payload OFDM data symbols of >> > variable length .We use both short symbols and long symbols for robust >> > double correlation based timing sychronization. >> > >> > Now I am wondering if we donot know the burst size in advance(since it >> > is variable) how to get burst by burst samples in the routine to >> > carry out time sychronization. A solution could be to use a buffer >> > size of (what length?). >> > >> > Now how can we search(using known short preamble) through the buffer >> > to identify for peaks ? If the buffer is too big the complexity will >> > be very high. >> > >> > So what should be the optimum buffer size? or any other way to carry >> > out time synchronization from continuous stream(I thought it could be >> > difficult!!). >> > >> > I googled without luck. I know there are guys who can answer it >> > instantly- that's why it is my fabourite group. >> > >> > Regards, >> > SantoshEric Jacobsen Minister of Algorithms, Intel Corp. My opinions may not be Intel's opinions. http://www.ericjacobsen.org
Reply by ●July 26, 20032003-07-26
santosh.nath@ntlworld.com (santosh nath) wrote in message news:<6afd943a.0307250339.1e805aa2@posting.google.com>...> Vladimir Vassilevsky <vlv@abvolt.com> wrote in message news:<3F209773.B4327435@abvolt.com>... > > You don't have to buffer the whole burst and search for the proper > > synchronization. Use a conventional early-late algorithm and the PLL. > > Are you also talking about programable synthesized PLL with the > following components: > > R divder,N diverder with A & B programable pre-scalar,Phase > Detector,Loop filter and VCO. > > We are aiming at total software solution, probably PLL requires extra > HW ckt. Could you see how it could be done with correlator. All > popular papers in IEEE uses the correlation with ML estimation. - > unfortunately I found them to mention total number of samples like > 1000000 as a simulation parameter but nobody mention about buffering > or detect burst by burst which could have been much easier to deal the > problem. Can anybody ref. a paper? > santosh > > > > > Vladimir Vassilevsky, Ph.D. > > > > DSP and Mixed Signal Design Consultant > > > > http://www.abvolt.com > > > > > > santosh nath wrote: > > > > > > Hi all, > > > > > > As we know the burst size is defined in GSM/EDGE mobile system as > > > 156.25 symbols > > > which takes 577 micro secs. This helps us to update our "Equalizer" > > > burst by burst. The symbol synchronization is done per burst > > > basis-meaning we calculate the time of arrival of a burst correlating > > > known training sequence with received samples.After calculating the > > > actual time offset we adjust start of a burst and then subsequent > > > other processes are carried forward. > > > > > > As 802.11a specifies that burst size for a OFDM based PHY burst is of > > > variable > > > size. Say, It starts with 10 short OFDM preambles,2 long OFDM > > > preambles,1 Signal OFDM symbol and payload OFDM data symbols of > > > variable length .We use both short symbols and long symbols for robust > > > double correlation based timing sychronization. > > > > > > Now I am wondering if we donot know the burst size in advance(since it > > > is variable) how to get burst by burst samples in the routine to > > > carry out time sychronization. A solution could be to use a buffer > > > size of (what length?). > > > > > > Now how can we search(using known short preamble) through the buffer > > > to identify for peaks ? If the buffer is too big the complexity will > > > be very high. > > > > > > So what should be the optimum buffer size? or any other way to carry > > > out time synchronization from continuous stream(I thought it could be > > > difficult!!). > > > > > > I googled without luck. I know there are guys who can answer it > > > instantly- that's why it is my fabourite group. > > > > > > Regards, > > > SantoshI am re-sending the post - thinking somebody has missed it! I guess "minister of algorithm" has something to comment on this. I have enjoyed his thoughtful posts on communication-couple of being in OFDM. J Kusuma writes also in this area - My question is bit implementation specific to real hardware and real application. If somebody comes up with a solution I will appriciate it very much. Regards, Santosh
Reply by ●July 26, 20032003-07-26
eric.jacobsen@ieee.org (Eric Jacobsen) wrote in message news:<3f22bced.47734556@news.earthlink.net>...> Santosh, > > You're aiming at a total software solution for 802.11a? Is your > intent to actually implement a real-time usable system this way? > > I think you're going to have trouble meeting the SIFS timing > requirement for turnaround to generate an ACK on a received packet. I > don't think anybody makes processors fast enough to do that in > software. A lot of people have trouble meeting timing with hardware > solutions. > > Cheers, > > EricSorry Eric, I did not clarify Vladimir's post. I meant total software solution for time synchronization algorithm only. Since he has given a good solution with PLL , unfortunately I can not use that since it needs extra HW Ckt. As you know, Cox et al gave nice solution on time synchonization with ML correlator which is indeed a total software solution.What I am missing in all these paper is a selection of a proper buffer size/ or burst size(Mind it it is variable) without which frame boundary detection would be difficult. At least my experience with GSM/EDGE equalization say so.I thought I could have mentioned that we will indeed go for a total hardware solution of the system at the very end - a flow like simulink( with MATLAB or C S functions)->VHDL->Optimization->FPGA->ASIC tap out. I guess I understand processor load etc and have some knowledge on the specification of the processors around us. I have posted an earlier mail specific to get attention of you. Hope I have clarified it better to get help from you. Regards, Santosh> > > On 25 Jul 2003 04:39:39 -0700, santosh.nath@ntlworld.com (santosh > nath) wrote: > > >Vladimir Vassilevsky <vlv@abvolt.com> wrote in message news:<3F209773.B4327435@abvolt.com>... > >> You don't have to buffer the whole burst and search for the proper > >> synchronization. Use a conventional early-late algorithm and the PLL. > > > >Are you also talking about programable synthesized PLL with the > >following components: > > > >R divder,N diverder with A & B programable pre-scalar,Phase > >Detector,Loop filter and VCO. > > > >We are aiming at total software solution, probably PLL requires extra > >HW ckt. Could you see how it could be done with correlator. All > >popular papers in IEEE uses the correlation with ML estimation. - > >unfortunately I found them to mention total number of samples like > >1000000 as a simulation parameter but nobody mention about buffering > >or detect burst by burst which could have been much easier to deal the > >problem. Can anybody ref. a paper? > >santosh > > > >> > >> Vladimir Vassilevsky, Ph.D. > >> > >> DSP and Mixed Signal Design Consultant > >> > >> http://www.abvolt.com > >> > >> > >> santosh nath wrote: > >> > > >> > Hi all, > >> > > >> > As we know the burst size is defined in GSM/EDGE mobile system as > >> > 156.25 symbols > >> > which takes 577 micro secs. This helps us to update our "Equalizer" > >> > burst by burst. The symbol synchronization is done per burst > >> > basis-meaning we calculate the time of arrival of a burst correlating > >> > known training sequence with received samples.After calculating the > >> > actual time offset we adjust start of a burst and then subsequent > >> > other processes are carried forward. > >> > > >> > As 802.11a specifies that burst size for a OFDM based PHY burst is of > >> > variable > >> > size. Say, It starts with 10 short OFDM preambles,2 long OFDM > >> > preambles,1 Signal OFDM symbol and payload OFDM data symbols of > >> > variable length .We use both short symbols and long symbols for robust > >> > double correlation based timing sychronization. > >> > > >> > Now I am wondering if we donot know the burst size in advance(since it > >> > is variable) how to get burst by burst samples in the routine to > >> > carry out time sychronization. A solution could be to use a buffer > >> > size of (what length?). > >> > > >> > Now how can we search(using known short preamble) through the buffer > >> > to identify for peaks ? If the buffer is too big the complexity will > >> > be very high. > >> > > >> > So what should be the optimum buffer size? or any other way to carry > >> > out time synchronization from continuous stream(I thought it could be > >> > difficult!!). > >> > > >> > I googled without luck. I know there are guys who can answer it > >> > instantly- that's why it is my fabourite group. > >> > > >> > Regards, > >> > Santosh > > Eric Jacobsen > Minister of Algorithms, Intel Corp. > My opinions may not be Intel's opinions. > http://www.ericjacobsen.org
Reply by ●July 27, 20032003-07-27
santosh nath wrote:> > > ... I meant total software > solution for time synchronization algorithm only. Since he has given a > good solution with PLL , unfortunately I can not use that since it > needs extra HW Ckt. ...Not necessarily. There are software PLLs. Google for "Costas loop". e.g. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������