Forums

Question regarding pilot based channel estimation/frequency synchronization in DVB-T

Started by Unknown March 29, 2005
Hello

I am working on the implementation of a demodulator for the European
DVB-T standard.
I had a query that I hope will be answered by this group.

During the initial acquisition, how does the DVB-T receiver lock on to
the starting phase for the continual/scattered pilot carrier symbols.
The ETSI standard mentions that the pilot carriers are derived by
mapping the bits derived from a PN-11 sequence generator (that runs
continously) to QAM symbols. Now, at the receive side, how does the
receiver know what the pilot symbols are for the current transmission.

To put it more succintly, the receiver starts to tune in at a
particular point in time of the transmit OFDM waveform... How does it
generate the reference pilot carriers that were supposed to help it in
channel estimation in the first place, when it has no idea about the
start phase of the acquired signal relative to the beginning of the
transmission.

Does it try various combinations of the start seed for the PN sequence
generator till it finds a particular seed that helps it lock to the
acquired waveform??

Thank you
Regards
Vikram

See comments below.

cvikram@mac.com wrote in message news:<1112147820.629336.212140@o13g2000cwo.googlegroups.com>...
> Hello > > I am working on the implementation of a demodulator for the European > DVB-T standard. > I had a query that I hope will be answered by this group. > > During the initial acquisition, how does the DVB-T receiver lock on to > the starting phase for the continual/scattered pilot carrier symbols. > The ETSI standard mentions that the pilot carriers are derived by > mapping the bits derived from a PN-11 sequence generator (that runs > continously) to QAM symbols. Now, at the receive side, how does the > receiver know what the pilot symbols are for the current transmission. > > To put it more succintly, the receiver starts to tune in at a > particular point in time of the transmit OFDM waveform... How does it > generate the reference pilot carriers that were supposed to help it in > channel estimation in the first place, when it has no idea about the > start phase of the acquired signal relative to the beginning of the > transmission. >
To start out, I'm not overly familiar with the pattern used in this standard, does it have reserved subcarriers that are fixed in both time and freq? From my experience, the pilot sequence (because that's really all you're interested in) is periodic, with a period that is usually set as the frame rate. To ensure that you have landed on a frame correctly, then usually a frame header is used, whereby an OFDM symbol containing signalling stuff and also other pilots to help kick start your channel estimator is stuck at the beginning of the OFDM frame and off you go. If the DVB-T standard doesn't have a start OFDM symbol at the beginning of a frame (in fact it may not even have the concept of frames) then my guess would be that a cross-correlation is performed to tune into where you are in your pilot sequence.
> Does it try various combinations of the start seed for the PN sequence > generator till it finds a particular seed that helps it lock to the > acquired waveform?? >
I think that would be a messy solution, because they would have to generate the pilot sequence for each seed and then perform the cross-correlation. It makes more sense to me if they would use a periodic pilot sequence, but what the hell do I know about DVB-T?! col
Col:

Thanks a lot for your prompt reply!
The DVB-T uses continual and scattered pilots. However, these pilots
derive their symbols from a running PN-11 generator. In effect, pilots
are fixed only w.r.t position in the OFDM symbol (which is why they are
named continual carriers).

The standard says that the PRBS is initialized so that the first output
bit from the PRBS sequence coincides with the first active carrier. A
new value is generated by the PRBS on every used carrier (whether or
not it is a pilot).

In summary, as you said, a cross-correlation looks like the most
realistic possibility to determine the pilots corresponding to the
current transmission.

It is a bit ironic, that the pilots need to be demodulated to ascertain
their actual values, when they were supposed to help reception in the
first place.

Regards
Vikram


Col Brown wrote:
> See comments below. > > cvikram@mac.com wrote in message
news:<1112147820.629336.212140@o13g2000cwo.googlegroups.com>...
> > Hello > > > > I am working on the implementation of a demodulator for the
European
> > DVB-T standard. > > I had a query that I hope will be answered by this group. > > > > During the initial acquisition, how does the DVB-T receiver lock on
to
> > the starting phase for the continual/scattered pilot carrier
symbols.
> > The ETSI standard mentions that the pilot carriers are derived by > > mapping the bits derived from a PN-11 sequence generator (that runs > > continously) to QAM symbols. Now, at the receive side, how does the > > receiver know what the pilot symbols are for the current
transmission.
> > > > To put it more succintly, the receiver starts to tune in at a > > particular point in time of the transmit OFDM waveform... How does
it
> > generate the reference pilot carriers that were supposed to help it
in
> > channel estimation in the first place, when it has no idea about
the
> > start phase of the acquired signal relative to the beginning of the > > transmission. > > > > To start out, I'm not overly familiar with the pattern used in this > standard, does it have reserved subcarriers that are fixed in both > time and freq? > > From my experience, the pilot sequence (because that's really all > you're interested in) is periodic, with a period that is usually set > as the frame rate. To ensure that you have landed on a frame > correctly, then usually a frame header is used, whereby an OFDM
symbol
> containing signalling stuff and also other pilots to help kick start > your channel estimator is stuck at the beginning of the OFDM frame
and
> off you go. > > If the DVB-T standard doesn't have a start OFDM symbol at the > beginning of a frame (in fact it may not even have the concept of > frames) then my guess would be that a cross-correlation is performed > to tune into where you are in your pilot sequence. > > > > Does it try various combinations of the start seed for the PN
sequence
> > generator till it finds a particular seed that helps it lock to the > > acquired waveform?? > > > > I think that would be a messy solution, because they would have to > generate the pilot sequence for each seed and then perform the > cross-correlation. It makes more sense to me if they would use a > periodic pilot sequence, but what the hell do I know about DVB-T?! > > col
Hello vikram,

I once wrote a basic DVBT demodulator, so I can give you some steps to
follow : (BTW if some expert have comment on this, inputs are welcome)

step 1) temporal synchronisation : use the guard interval . some
correlation technique will do the job (I call it "estimation par le
maximum de vraisemblance" in french )

step 2) FTT . easy

step 3) basic  egalisation : the amplitude of the continual pilots
should be matched to E(cp)=3D4/3 . use linear interpolation to derive the
egalisation for the carriers between the continual pilots

step 4) "Synchronisation du peigne des pilotes dispers=E9s " -
synchronisation of the comb of the scattered pilots . find the position
(4 possibilities) that maximise the probability of a comb of pilots
.you have to use the fact that  E(pilots) =3D 4/3 whereas mean(E(data
carriers))=3D 1 .
Once its done, you have a larger set of pilots : continual pilots +
scaterred pilots

step 5) " Estimation de l'erreur de synchro "  estimation of the
synhronisation error.
that is : any drift in the temporal synchronisation (dt error) will
lead to the constellation being rotated continuously (dt in temporal
domain =3D> dphi in frequency domain)
estimate dphi ;  there you need the Wk PRBS sequence; depending on this
sequence,
the phase of the pilots is 0 or PI. Once dphi parameter  is found,
compensate for it.

step 6) the constellation looks better , but you still have a phase
error (=3Drotation of constellation) . same solution ;: estimate then
compensation

step 7) the constellation should be ok now, I let you do the following
steps :
symbol decision , bit mapping , ...

Good luck,

Julien

Julien

Excellent pointers. Just half an hour ago, the light in my mind started
to glow.
I realized that the PRBS 11 sequence is periodic with period 2048 which
implies that w(n+2048)=3Dw(n).

This implies that once basic synchronization with the FFT start window
is achieved, we know that the continual pilots start with the seed of
the PN11 sequence irrespective of whether a 2k/4k/8k transmitter is
used.

As you mentioned in your email, estimation of the scattered pilot
position is more difficult since there are four possibilities per DVB-T
symbol. So we need to estimate the actual location using temporal
correlation.

Incidentally, I had another question, is it possible to estimate
frequency offsets greater that 1/(2*Tfft)
for an OFDM system. The algorithm that I use now uses the periodicity
of the CP to obtain a jonit estimation of the symbol timing and
frequency offset (referre paper by Meyr et al). It is stable upto
estimating offsets of 0.5*caarrier spacing.....


Regards
Vikram
Julien wrote:
> Hello vikram, > > I once wrote a basic DVBT demodulator, so I can give you some steps
to
> follow : (BTW if some expert have comment on this, inputs are
welcome)
> > step 1) temporal synchronisation : use the guard interval . some > correlation technique will do the job (I call it "estimation par le > maximum de vraisemblance" in french ) > > step 2) FTT . easy > > step 3) basic egalisation : the amplitude of the continual pilots > should be matched to E(cp)=3D4/3 . use linear interpolation to derive
the
> egalisation for the carriers between the continual pilots > > step 4) "Synchronisation du peigne des pilotes dispers=E9s " - > synchronisation of the comb of the scattered pilots . find the
position
> (4 possibilities) that maximise the probability of a comb of pilots > .you have to use the fact that E(pilots) =3D 4/3 whereas mean(E(data > carriers))=3D 1 . > Once its done, you have a larger set of pilots : continual pilots + > scaterred pilots > > step 5) " Estimation de l'erreur de synchro " estimation of the > synhronisation error. > that is : any drift in the temporal synchronisation (dt error) will > lead to the constellation being rotated continuously (dt in temporal > domain =3D> dphi in frequency domain) > estimate dphi ; there you need the Wk PRBS sequence; depending on
this
> sequence, > the phase of the pilots is 0 or PI. Once dphi parameter is found, > compensate for it. > > step 6) the constellation looks better , but you still have a phase > error (=3Drotation of constellation) . same solution ;: estimate then > compensation > > step 7) the constellation should be ok now, I let you do the
following
> steps : > symbol decision , bit mapping , ... >=20 > Good luck, >=20 > Julien