Waqas-

> I am working on an OFDM system using Matlab. The bits received
> at the receiver have a constant error ( i.e., avg 200
> bits out of 1000 are always wrong).. when I decreased no. of
> bits in a symbol to let say 49 and FFT length to 64,
> there was no error.
> It looks like that there is some error in the computation of
> IFFT/FFT bolcks by Matlab (maybe some rounding of the
> data by Matlab). I am unsure how to correct this error. any
> help in this regard will be appreciated..

If you are the person who, after 25 years, can find an error in MATLAB FFT and iFFT functions, then you would be
famous, and Toyota might hire you to prove/disprove their car acceleration issue :-)

My suggestion would be to give a more detailed problem description: what is symbol and FFT size in non-working case?
Which bits are wrong? How are you calling FFT function (window, zero fill, etc) ?

-Jeff
Waqas-

> I went through my code and found that the cause of error was not FFT/IFFT
> but the filters. The FIR filter was introducing a delay which was causing a
> shift in the data. that shift was responsible for the fixed number of
> errors. I use convolution to find the delay and now my code is working all
> right.
> thanks once again

Hey great that results make sense now. Good work.

> I wish to be someone who find an error in matlab..:P

Ya me 2, hehe.

-Jeff

> On Fri, Mar 19, 2010 at 11:45 AM, Jeff Brower wrote:
>
>> Waqas-
>>
>> > I am working on an OFDM system using Matlab. The bits received
>> > at the receiver have a constant error ( i.e., avg 200
>> > bits out of 1000 are always wrong).. when I decreased no. of
>> > bits in a symbol to let say 49 and FFT length to 64,
>> > there was no error.
>> > It looks like that there is some error in the computation of
>> > IFFT/FFT bolcks by Matlab (maybe some rounding of the
>> > data by Matlab). I am unsure how to correct this error. any
>> > help in this regard will be appreciated..
>>
>> If you are the person who, after 25 years, can find an error in MATLAB FFT
>> and iFFT functions, then you would be
>> famous, and Toyota might hire you to prove/disprove their car acceleration
>> issue :-)
>>
>> My suggestion would be to give a more detailed problem description: what
>> is symbol and FFT size in non-working case?
>> Which bits are wrong? How are you calling FFT function (window, zero fill,
>> etc) ?
>>
>> -Jeff
> --
> Waqas Abdullah,
>
> "you will never know what u can achieve unless you try..."
>