OFDM demodulation-sampling time offset

Hello

On trying to demodulate an OFDM signal based on the DVB-T standard, I
immediately encountered difficulties in doing so. The problem lies in:

1. Frequency offset estimation.
2. Symbol timing estimation (symbol timing ambiguity occurs during the
sampling process at the receiver A/D converter.
3. Channel coefficient estimation.

Problem 1 is addressed in a straightforward manner using the
periodicity of the cyclic prefix in the time domain. However, 2 and 3
seem to be more difficult to resolve.

Note that we are trying to estimate h(k), and m in the expression
R(k)=H(k)*X(k)*exp(-j*2*pi*mk/N) where m is the symbol timing
ambiguity, H(k) is that channel coefficient in subchannel k,X(k) is the
information symbol, and R(k) is the FFT output at the receiver.

Is there a good way to de-couple the problem of symbol timing
estimation from channel coefficient estimation? It seems to be that
there is no straightforward way to find out where the start of the FFT
window lies, without estimating the channel first....
Next, the value "m" may not be an integer, it may infact be a real
number since the sampling may start at an arbitrary location in the
OFDM waveform...

Alternatively, are there any blind algorithms to find out the sampling
time offset of an OFDM signal....

Thank you
Vikram

>Hello
>
>On trying to demodulate an OFDM signal based on the DVB-T standard, I
>immediately encountered difficulties in doing so. The problem lies in:
>
>1. Frequency offset estimation.
>2. Symbol timing estimation (symbol timing ambiguity occurs during the
>sampling process at the receiver A/D converter.
>3. Channel coefficient estimation.
>
>Problem 1 is addressed in a straightforward manner using the
>periodicity of the cyclic prefix in the time domain. However, 2 and 3
>seem to be more difficult to resolve.
>
>Note that we are trying to estimate h(k), and m in the expression
>R(k)=H(k)*X(k)*exp(-j*2*pi*mk/N) where m is the symbol timing
>ambiguity, H(k) is that channel coefficient in subchannel k,X(k) is the
>information symbol, and R(k) is the FFT output at the receiver.
>
>Is there a good way to de-couple the problem of symbol timing
>estimation from channel coefficient estimation? It seems to be that
>there is no straightforward way to find out where the start of the FFT
>window lies, without estimating the channel first....
>Next, the value "m" may not be an integer, it may infact be a real
>number since the sampling may start at an arbitrary location in the
>OFDM waveform...
>
>Alternatively, are there any blind algorithms to find out the sampling
>time offset of an OFDM signal....
>
>Thank you
>Vikram
>
>

In my mind, sampling time offset is not a big problem in OFDM system.
Because channel estimation in Frequency-domain can compensate it very
well.
But if symbol-bounded is shift too far, it will leads to interference.

About symbol timing estimation, a typical way is to use
Cycle-information.
Detail information please : Optimum receiver design for OFDM-based
Broadband transmission-Part II (Michael speth)

hi,every body. Now, I am doing sampling time offset estimation, i just do
it in frequency direction(after FFT), and recived sampling time offset : 
x=(Ti-Ts)/Ts, x is used to via loop filter and NCO, and Mu is output of
output, it is used to adjust integrator , yes , i am useing gardner
algorithm, reference is :Interpolation in digital modems I
:foundmentals,and Interpolation in digital modems II.

but i want to ask a question:
x above can be adjust loop filter directly? and x is different from timing
error which is computated using gardner algorithm?

3x!