Re: [STDS-802-16] Fwd: Re: [STDS-802-16] [PREAMBLE] - Technical discussion on preamble structure and sequence design
Dear Jeff,
Thank you for your reply.
As you mentioned on the coherent method, after picking the mid-point within a plateau, one of what we can do for refining timing synch is 'FFT-division by Preamble-IFFT'. I wonder if the above refining process can give the fine timing offset without fine freq. synchronization procedure.
I think that FFT without compensation of integer frequency offset will make the shifted preamble in freq. domain. The shfted preamble divided by the preamble does not make sense to provide the fine timing offset.
In summary, I think the whole steps on repetitive pattern are as follows:
1. coarse timing synch ( using correlation between two halves)
2. coarse freq. offset acquisition ( by phase rotation between two halves)
3. fine timing synch (by correlation between some points within plateau with time-domain version of preamble)
4. fine freq. offset acquisition
and the whole step of the coherent method on repetitive pattern I undetstand are as follows:
1. coarse timing synch ( using correlation between two halves)
2. coarse freq. offset acquisition ( by phase rotation between two halves)
3. fine timing synch (by the coherent method 'FFT-division by Preamble-IFFT)
In the coherent method, I wonder where is the 'fine freq. offset acquisition' inserted.
Best Regards,
Bin-Chul Ihm
----- Original Message -----
From: "Jeff Zhuang" <zhuang@labs.mot.com>
To: "bcihm" <bcihm@LGE.COM>
Cc: <STDS-802-16@listserv.ieee.org>
Sent: Friday, August 13, 2004 12:32 AM
Subject: Re: [STDS-802-16] Fwd: Re: [STDS-802-16] [PREAMBLE] - Technical discussion on preamble structure and sequence design
> Dear Bin-Chul, Jiho, and all,
>
> On using the conjugate symmetric pattern, I was wondering how good the
> pattern will be in a multipath channel. Meanwhile, the repetition
> pattern is kept in the received signal regardless of the channel.
> Initial frequency synchronization may be also easier to get with the
> repetition. The approach you mentioned, Bin-Chul, which uses the phase
> rotation between CP and the tail of symbol is not as good as the
> repetition structure, which actually uses the phase rotation between two
> halves.
>
> On using the repetition structure, you are right, Bin-Chul, that there
> are only "vague" peaks. Actually , you can only see a plateau with a
> theoretical width equal to the difference between CP and the channel
> length (the plateau may be defined by a threshold relative to the peak
> in practice). One can pick a point within the plateau as the coarse sync
> point (such as the mid-point). It is certainly coarse, but any sync
> point that can ensure the channel sitting within the CP is a "valid"
> sync point (for the example of one-tap channel, any point within the CP
> will not incur any inter-symbol or inter-carrier interference). But we
> should refine the timing offset so that we can have ISI protection over
> channels up to a full-CP length. This is what we can do: take the FFT
> over the window starting from the coarse sync point, divide in the
> frequency domain by the preamble, take an IFFT back to the time domain.
> Now we have the channel that may be shifted (delayed ) in time. The
> shift is your fine timing offset and you have also the channel. It is a
> coherent method. It is equivalent to a time domain sliding correlator,
> but more efficiently implemented in the frequency domain.
>
> On handover, of course, we need to use cell-specific preamble (not a
> global preamble). During handover or initial cell search, you can do the
> above frequency-domain division to as many preambles as you want and get
> the timing and channels to all the corresponding bases. I have used
> those channel information for interference suppression (when the
> subscriber is equipped with multiple antenna) in simulations for some
> low SINR (e.g., -10dB). You can also use them for soft handover.
>
> Best Regards,
>
> Jeff Zhuang
> Motorola
>
>
>
> bcihm wrote:
>
> > Dear Jiho Jang and all,
> >
> > I agree that the computational complexity required in the correlation
> > algorithm using symmetric pattern is higher than that of using
> > repetitive pattern. However, the peak value among correlation values
> > from the algorithm using the repetitive pattern is vague so that
> > multiple candidates for fine synchronization appear. As a result, a
> > lot of additional correlation process with time domain version of
> > preamble PN sequence are required to find fine timing. On the other
> > hand, the peak value from the algorithm using symmetric pattern is
> > easily detected even though Nfft/4 samples are used for correlation.
> > This results in few additional correlations for fine timing
> > synchronization. Therefore overall processing burdens of both
> > algorithms are almost same.
> >
> > Secondly, I also think that coarse frequency offset cannot be obtained
> > only with the symmetric pattern. The coarse frequency offset can be
> > acquired efficiently from phase rotation between CP and tail of symbol
> > in symmetric pattern. When the coarse freq. offset is computed after
> > the coarse time synchronization, the performance under the symmetric
> > pattern is better than that under the repetitive pattern.
> >
> > In addition, I wonder if common preamble is useful in handover. Since
> > preamble indices of neighbor cells are known to SS through the
> > messages related to handover, acquisition of synchronization with
> > neighbor cells can be achieved as fast as the common preamble does.
> > The common preamble right before the cell-specific preamble in this
> > case seems to be overhead. The common preamble every second frame
> > makes the time required for scanning longer. Since the main objective
> > of scanning is measurement of CINR of neighbor cells, SS has to find
> > the cell-specific preamble. Therefore, SS wastes a frame including
> > common preamble without obtaining CINR.
> >
> > This is my short comment on common preamble and the repetitive pattern.
> >
> > Best Regards,
> >
> > Bin-Chul Ihm
> >
> > ----- Original Message -----
> >
> > From: owner-stds-802-16@listserv.ieee.org
> > <mailto:owner-stds-802-16@listserv.ieee.org>
> > To: STDS-802-16@listserv.ieee.org
> > <mailto:STDS-802-16@listserv.ieee.org>
> > Sent: Thursday, August 12, 2004 9:54 AM
> > Subject: [STDS-802-16] Fwd: Re: [STDS-802-16] [PREAMBLE] -
> > Technical discussion on preamble structure and sequence design
> >
> > Hi, Bin-Chul:
> >
> > I think it is a good comment. The real valued sequences in
> > frequency domain yield conjugate symmetric patterns in time domain.
> >
> > But, the time synchronization using the symmetric structure has
> > some problems. In addition, the frequency offset estimation cannot
> > be done using the symmetric structure only.
> >
> > Firstly, the computational complexity required in the algorithm
> > using symmetric pattern is much higher than the algorithm using
> > repetitive pattern.
> >
> > The algorithm using the symmetric structure has to newly calculate
> > the correlation value by correlating all the time samples at every
> > sample timing. In the algorithm using the repetitive pattern, on
> > the other hand, the correlation value can be obtained by only
> > adding the newest one sample value and discarding the oldest one
> > sample value at every sample timing.
> >
> > Since the timing hypotheses should be collected at every sample
> > timing for a frame duration at least, the synchronization using
> > repetitive pattern can much lower the computational complexity
> > than the symmetric based algorithm.
> >
> > Secondly, the coarse frequency offset estimation cannot be
> > achieved from the symmetric structure only. Instead, you can
> > easily obtain the frequency offset estimation by using the
> > repetitive pattern, where you only need to compare the phase
> > rotations between a pair of samples.
> >
> > Regards,
> >
> > Jiho Jang
> >
> >
> >
> > ------------------------------------------------------------------------
> > Jiho Jang (Ph. D)
> >
> > NTP Development Team (System Lab. 2)
> >
> > Telecommunication System Division
> >
> > Telecommunication Network
> >
> > SAMSUNG ELECTRONICS CO.,LTD
> >
> > T. 82-31-279-3355
> >
> > M. 016-9233-8541
> >
> > jiho.jang@samsung.com <mailto:jiho.jang@samsung.com>
> >
> > ------------------------------------------------------------------------
> >
> >
> > ------- Original Message -------
> > Sender : bcihm<bcihm@LGE.COM>
> > Date : 2004-08-12 09:26
> > Title : Re: [STDS-802-16] [PREAMBLE] - Technical discussion on
> > preamble structure and sequence design
> > Dear Anand Dabak and all,
> >
> > I think that initial symbol timing acquisition can be done using
> > symmetric structure of preamble.
> >
> > Because the preamble is generally real-valued, its IFFT version in
> > time domain has symmetric structure inherently.
> >
> > For example, if Nfft=8 and preamble consists of only real, then
> > samples in time domain within a symbol duration are symmetric as
> > follows:
> >
> > [ a b c d e d* c* b* ]
> >
> > The samples around 'e' are conjugated symmetric and this property
> > can be used for acquisition of symbol timing.
> >
> > The symmetric structure is more usefull to finding correlation
> > peak than CP based method because the symmetric part has length of
> > (Nfft/2-1) longer than CP.
> >
> > And I think this inherent symmetric structure is more useful than
> > repetitive structure.
> >
> > For example, let's consider a repetitive structure as following:
> >
> > [ a b c d a b c d ]
> >
> > The correlation at sample 'a' is |a|^2 + |b|^2+ |c|^2+|d|^2 and
> > that at sample 'b' is |b|^2+ |c|^2+|d|^2 + ax*.
> >
> > This means that it is hard to discriminate the peak among
> > correlations over several positions.
> >
> > However, in the symmetric structure, the correlation at sample 'e'
> > is |b|^2+ |c|^2+|d|^2 and that at sample 'd*' is ec* + db* + cx.
> >
> > This means that it is easy to discriminate the peak among
> > correlations over several positions.
> >
> > In summary, it is enough for acquisition of symbol timing to use
> > the inherent symmetric structure obtained without any manipulation
> > to preamble structure in my opinion.
> >
> > Best Regards
> >
> > Bin-Chul Ihm
> >
> > LGE
> >
> > ----- Original Message -----
> > From: owner-stds-802-16@listserv.ieee.org
> > <mailto:owner-stds-802-16@listserv.ieee.org>
> > To: STDS-802-16@listserv.ieee.org
> > <mailto:STDS-802-16@listserv.ieee.org>
> > Sent: Thursday, August 12, 2004 5:29 AM
> >
> > Subject: [STDS-802-16] [PREAMBLE] - Technical discussion on
> > preamble structure amd sequence design
> >
> >
> > Dear All,
> >
> >
> > This mail is a followup to yesterday's conference call:
> >
> > In order to help us in evaulating the performance of the
> > existing preamble and compare it against any changes, we would
> > like to know the simulation scenarion that needs to be assumed
> > (eg. a 19 cell model) and the procedure for the initial
> > acquisition (after power on of mobile).
> >
> > Specifically, we want to know whether one can assume that
> > initial symbol timing acquisition is done using CP correlation
> > of the OFDM symbols and that the preamble is used only for
> > Frame timing. If this is a valid assumption, has any one run
> > any simulations using the CP technique to measure its
> > reliability? If so can they share their results with the group?
> >
> > In case the CP base symbol timing is not reliable, we have to
> > perform a time correlation against the preamble sequence
> > itself for initial symbol timing acquisition ? in this case,
> > we are concerned with the correlator complexity which could be
> > potentially large as we have to do these correlation
> > operations per ADC output sample. So, in this case we would
> > like to suggest that we look at alternative preamble
> > structures and sequences to reduce the complexity (and in turn
> > power consumption) of the mobile for initial symbol timing
> > acquisition.
> >
> >
> > Anand Dabak
> > Texas Instruments
> >
>
>
>