Equalization is necessary in OFDMA-PHY?

I have a little Matlab simulation for a OFDMA-PHY tranceiver based on the
802.16e standard. The BER results over a simple AWGN channel using a QPSK
constellation coincide with those of a single carrier QPSK. With this I
assume my code is correct.

The problem is that once I add a rayleigh fading channel to the simulation
the BER values stay high (0.5) for all SNR values. This of course does not
coincide with the theory.

I use a 1/8 Cyclic prefix and still get high BER values and I am thinking
maybe I need to compensate the channel (equalize). But in all papers/books
I read they always mention that as long the CP time is larger than the
largest path delay then OFDM should not be affected by the dispersive
channel....

Am I missing somthing here?? do I need to compensate the channel anyway?
if so how to perfectly compensate? 

In my simulation the guard time (CP) is 11.256e-6 sec while the path
delays I use are 100e-9 and 200e-9 in a 2-path model. I use the
rayleighchan function to create the channel and the filter function to
pass the signal.

any tips are greatly appreciated.

Fs = 1424000;       % Sampling frequency
Tb = 8.9888e-05;    % Useful OFDM symbol time
Tg = 1.1236e-05;    % Guard time
channel = rayleighchan(1/Fs, 0, [100e-9 200e-9],[0 -2.5])

On Sat, 25 Nov 2006 20:03:52 -0000, ryujin_ssdt <hsanson@gmail.com> wrote:

> I have a little Matlab simulation for a OFDMA-PHY tranceiver based on the
> 802.16e standard. The BER results over a simple AWGN channel using a QPSK
> constellation coincide with those of a single carrier QPSK. With this I
> assume my code is correct.
>
> The problem is that once I add a rayleigh fading channel to the  
> simulation
> the BER values stay high (0.5) for all SNR values. This of course does  
> not
> coincide with the theory.
>
> I use a 1/8 Cyclic prefix and still get high BER values and I am thinking
> maybe I need to compensate the channel (equalize). But in all  
> papers/books
> I read they always mention that as long the CP time is larger than the
> largest path delay then OFDM should not be affected by the dispersive
> channel....

In an OFDM system, a dispersive channel introduces three effects;  
intersymbol interference (ISI), intercarrier interference (ICI), and  
frequency selectivity (phase/frequency distortion of the individual  
sub-carriers).

Use of a time-domain guard interval (longer than the delay spread)  
eliminates the first effect (ISI), as successive symbols can no longer  
affect each other.  If a cyclic prefix is used as the guard interval, the  
second effect (ICI) is also eliminated, due to the circular convolution  
theorem.

However, the third effect is not compensated for by the cyclic prefix, and  
so in general, equalisation is required.


> Am I missing somthing here?? do I need to compensate the channel anyway?
> if so how to perfectly compensate?

In general, the channel response must be estimated, and then the  
sub-carriers equalised.  Most methods are performed in the frequency  
domain (due to the structure of OFDM), and utilise pilot symbols.  For a  
simple example, see [1].


[1] R. Negi and J. Cioffi, "Pilot tone selection for channel estimation in  
a mobile OFDM system," IEEE Trans. Consumer Electronics, pp. 1122-1128,  
Aug. 1998.


-- 
Oli

Thanks a lot... now I understand better the situation.

Since my interest is not OFDM equalization (I am working on sync) I would
like to have the signal equalized in the easiest manner. For example
having perfect knowledge of the CIR and using simple zero forcing C=Y/X.

I am using the rayleighchan function to create the channel but I am unable
to find (even in matlab's help files) how to obtain the CIR from the
rayleigh object to equalize the signal. Any tips would be greately
apreciated. A simple example with actual code (Using the rayleighchan
fuction) would be extremely useful.

thanks.

>On Sat, 25 Nov 2006 20:03:52 -0000, ryujin_ssdt <hsanson@gmail.com>
wrote:
>
>> I have a little Matlab simulation for a OFDMA-PHY tranceiver based on
the
>> 802.16e standard. The BER results over a simple AWGN channel using a
QPSK
>> constellation coincide with those of a single carrier QPSK. With this
I
>> assume my code is correct.
>>
>> The problem is that once I add a rayleigh fading channel to the  
>> simulation
>> the BER values stay high (0.5) for all SNR values. This of course does 

>> not
>> coincide with the theory.
>>
>> I use a 1/8 Cyclic prefix and still get high BER values and I am
thinking
>> maybe I need to compensate the channel (equalize). But in all  
>> papers/books
>> I read they always mention that as long the CP time is larger than the
>> largest path delay then OFDM should not be affected by the dispersive
>> channel....
>
>In an OFDM system, a dispersive channel introduces three effects;  
>intersymbol interference (ISI), intercarrier interference (ICI), and  
>frequency selectivity (phase/frequency distortion of the individual  
>sub-carriers).
>
>Use of a time-domain guard interval (longer than the delay spread)  
>eliminates the first effect (ISI), as successive symbols can no longer  
>affect each other.  If a cyclic prefix is used as the guard interval, the
 
>second effect (ICI) is also eliminated, due to the circular convolution 

>theorem.
>
>However, the third effect is not compensated for by the cyclic prefix,
and  
>so in general, equalisation is required.
>
>
>> Am I missing somthing here?? do I need to compensate the channel
anyway?
>> if so how to perfectly compensate?
>
>In general, the channel response must be estimated, and then the  
>sub-carriers equalised.  Most methods are performed in the frequency  
>domain (due to the structure of OFDM), and utilise pilot symbols.  For a 

>simple example, see [1].
>
>
>[1] R. Negi and J. Cioffi, "Pilot tone selection for channel estimation
in  
>a mobile OFDM system," IEEE Trans. Consumer Electronics, pp. 1122-1128, 

>Aug. 1998.
>
>
>-- 
>Oli
>