Well...

imagine we compare time on our wristwatches today.
Mine is five minutes late.

We do the same again tomorrow:
case a) Still five minutes - the difference hasn't changed
case b) The watches have drifted further apart - maybe six minutes
difference.

OFDM is quite sensitive to case b) because the subcarriers lose their
orthogonality... the "O" in OFDM :) 

Cheers

Markus

On Aug 20, 5:55 pm, "mnentwig" <mnent...@elisanet.fi> wrote:
Hi Markus,

> there are two things that might get mixed up:
> - sampling clock offset (in terms of absolute time delay)
> - sampling frequency offset

Did not understand the difference. Can you please explain.

> The first is trivial in OFDM, it simply appears as frequency dependent
> phase shift inside the FFT. But the number of samples before / after the
> offset model remain the same.
hmmm...need not be. If the packet is long, the presence of sampling
clock offset
can result in drifts more than the sampling period.

> The latter may do real damage to OFDM reception, but is harder to model
Am unable to understand the difference between the first and second.

Thanks,
Krishna
http://dsplog.blogspot.com

>Problem:
>the first value in resampled is NaN. 

Right... it falls out of the range.
A quick hack is to use

resampled=interp1(origSampleTime, testsig, nonidealSampleTime, 'extrap');

instead. Or offset the "nonidealSampleTime", so that it falls in range.

If you're building a simulator, it may be useful to generate two extra
symbols, one before and one after your simulation length.

-Markus

On Aug 20, 2:55 pm, "mnentwig" <mnent...@elisanet.fi> wrote:
> Hi,
>
> there are two things that might get mixed up:
> - sampling clock offset (in terms of absolute time delay)
> - sampling frequency offset
>
> The first is trivial in OFDM, it simply appears as frequency dependent
> phase shift inside the FFT. But the number of samples before / after the
> offset model remain the same.
>
> The latter may do real damage to OFDM reception, but is harder to model
> (because the model produces more or less samples than it consumes).
> You could use interp1, but it may introduce too much aliasing unless your
> signal is heavily oversampled.
>
> Here is some quick+dirty matlab program tested on octave:
> Note, this signal _is_ heavily oversampled, so you can't see the
> aliases...
> Higher order interpolation might do the job, though.
>
> The "exact" solution is band limited interpolation with sinc pulses around
> each sample (polyphase filtering).
>
> Cheers
>
> Markus
>
> origSampleTime=1:30;
> nonidealSampleTime=origSampleTime*0.99;
> testsig=sin(origSampleTime/30*2*pi);
> resampled=interp1(origSampleTime, testsig, nonidealSampleTime);
> figure(); hold on;
> plot(origSampleTime, testsig, 'r+-');
> plot(nonidealSampleTime, resampled, 'g+-');

I am exactly simulating the same thing as you wrote in the above
code ,
Problem:
the first value in resampled is NaN. So how can I take FFT ??
I tried to put some arbitrary value or a 0 in testsig before the first
value to get some value instead of NaN in the resampled.
Now I can take FFT but due to this arbitrary value or 0 , the result
at the output of FFT is not accurate and my algorithm of sampling
clock offset estimate is not giving correct result.

How will you address the problem ??

Thanks a lot for your reply.

Hi,

there are two things that might get mixed up:
- sampling clock offset (in terms of absolute time delay)
- sampling frequency offset

The first is trivial in OFDM, it simply appears as frequency dependent
phase shift inside the FFT. But the number of samples before / after the
offset model remain the same. 

The latter may do real damage to OFDM reception, but is harder to model
(because the model produces more or less samples than it consumes).
You could use interp1, but it may introduce too much aliasing unless your
signal is heavily oversampled.

Here is some quick+dirty matlab program tested on octave: 
Note, this signal _is_ heavily oversampled, so you can't see the
aliases...
Higher order interpolation might do the job, though.

The "exact" solution is band limited interpolation with sinc pulses around
each sample (polyphase filtering).

Cheers

Markus

origSampleTime=1:30;
nonidealSampleTime=origSampleTime*0.99;
testsig=sin(origSampleTime/30*2*pi);
resampled=interp1(origSampleTime, testsig, nonidealSampleTime);
figure(); hold on;
plot(origSampleTime, testsig, 'r+-');
plot(nonidealSampleTime, resampled, 'g+-');

On Aug 18, 11:46 am, nomee97 <nomi9...@hotmail.com> wrote:
> On Aug 17, 6:09 pm, Krishna <krishna.pil...@gmail.com> wrote:
>
> > > I mean I want to see my symbols after the FFT block and they should
> > > have sampling clock offset which should be zero for the first
> > > subcarrier and should accumulate as the sub carrier number increases.
>
> > Noman,
>
> > Corresponding to the sampling clock offset, one can see a linearly
> > increase phase
> > in frequency domain. Recall that delay in time domain corresponds to
> > phase growth
> > in frequency domain.
>
> > HTH,
> > Krishnahttp://dsplog.blogspot.com
>
> Thanks for the information,
> I am using the function resample as
> resample(data,3,2);
> in this was it will be resampled at the rate of 1.5 but in this was I
> am getting 64 + 32 points if I use the data with 64 points.
> Now I need to take the FFT which is 64 point FFT at the OFDM receiver
> which 64 points do I need to take from these 64+32 points ??
>
> Regards,
> Noman.

That's a ridiculously large offset. Try it with 100 ppm.

John

> I am using the function resample as
> resample(data,3,2);
> in this was it will be resampled at the rate of 1.5 but in this was I
> am getting 64 + 32 points if I use the data with 64 points.
> Now I need to take the FFT which is 64 point FFT at the OFDM receiver
> which 64 points do I need to take from these 64+32 points ??

I do not think that the scenario you tried is exact. Typically, the
clock offset between the
transmitter and receiver wont be at the rate 3/2. Maybe a 10MHz clock
at the Tx and a 10.0001MHz clock
at the Rx is reasonable.

Try looking at
M. Speth, S. A. Fechtel, G. Fock, H. Meyr, "Optimum Receiver Design
for Wireless Broad-Band Systems
Using OFDM-Part I," IEEE Transactions on Communications, Vol. 47, No.
11, November 1999, pp. 1668-
1677

Regards,
Krishna
http://dsplog.blogspot.com

On Aug 17, 6:09 pm, Krishna <krishna.pil...@gmail.com> wrote:
> > I mean I want to see my symbols after the FFT block and they should
> > have sampling clock offset which should be zero for the first
> > subcarrier and should accumulate as the sub carrier number increases.
>
> Noman,
>
> Corresponding to the sampling clock offset, one can see a linearly
> increase phase
> in frequency domain. Recall that delay in time domain corresponds to
> phase growth
> in frequency domain.
>
> HTH,
> Krishnahttp://dsplog.blogspot.com

Thanks for the information,
I am using the function resample as
resample(data,3,2);
in this was it will be resampled at the rate of 1.5 but in this was I
am getting 64 + 32 points if I use the data with 64 points.
Now I need to take the FFT which is 64 point FFT at the OFDM receiver
which 64 points do I need to take from these 64+32 points ??

Regards,
Noman.

> I mean I want to see my symbols after the FFT block and they should
> have sampling clock offset which should be zero for the first
> subcarrier and should accumulate as the sub carrier number increases.
>

Noman,

Corresponding to the sampling clock offset, one can see a linearly
increase phase
in frequency domain. Recall that delay in time domain corresponds to
phase growth
in frequency domain.

HTH,
Krishna
http://dsplog.blogspot.com

On Aug 16, 1:30 pm, John <sampson...@gmail.com> wrote:
> On Aug 16, 3:51 am, nomee97 <nomi9...@hotmail.com> wrote:
>
> > I am simulationg an OFDM system in Matlab for which I need to generate
> > a sampling clock offset in my transmitted signal so that I can write
> > an algorithm at the receiver to correct it.
>
> > Can any body tell me how to generate a sampling clock offset in the
> > Transmitted signal. I think I have to use interp1 function but I dont
> > have a clear idea. Sampling Clock offset should be a linear function
> > of frequency number.
>
> > Any help will be highly appreciated.
> > Thanks in advance.
>
> > Noman.
>
> You might consider using Matlab's 'resample' function.
>
> John

Ok fine I am using the matlab function resample to generate the
Sampling clock offset in time domain in the transmitted signal,
How can I see my symbols in frequency domain with the sampling clock
offset linearly increasing with frequency numbers.
I mean I want to see my symbols after the FFT block and they should
have sampling clock offset which should be zero for the first
subcarrier and should accumulate as the sub carrier number increases.

Thanks.