Reply by Hany ●June 17, 20042004-06-17
CB,

Yes, that's exactly what i'm doing, but my ifft transmitter process is
a matrix of coefficents, with 64 samples input, and 320 samples
output. (i did that for the oversampling needed to pass the signal to
the channels with a min tap delay of 10ns). the fft receiver is the
transpose matrix of the previous one ..
anyway..
when i turn completely the noise off, i get a ZER BER. I detect the
symbols exactly with no errors.!

i really don't know where the problem comes from, when i introduce the
noise with the multi-path channel model.


 

cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406090950.7bed10d3@posting.google.com>...

> Hany,
> 
> The bin-by-bin multiplication is performed after you take the fft of
> your received signal.  Say the size of the fft=32.  Then you perform
> the multiplication of the inverse 32-point fft of the chnl impulse
> response and the fft of the received signal.  I assume that this is
> exactly what you're doing?  Is it not?  The result should be exactly
> what you transmitted.  Where are you getting the problems?  What
> happens when you turn the noise off completely?
> 
> As an example in Matlab code, the equ looks something like:
> 
> chnl = [1 0.5 0.1];
> equ = 1./fft(chnl,fft_size); % form zero forcing equ
> data = fft(rx,fft_size).* equ; % perform bin-by-bin multiplication
> 
> where rx is the received time domain signal after the prefix removal.
> 
> CB
>   

<snipped the rest of the message>
Reply by Col Brown ●June 9, 20042004-06-09
Hany,

The bin-by-bin multiplication is performed after you take the fft of
your received signal.  Say the size of the fft=32.  Then you perform
the multiplication of the inverse 32-point fft of the chnl impulse
response and the fft of the received signal.  I assume that this is
exactly what you're doing?  Is it not?  The result should be exactly
what you transmitted.  Where are you getting the problems?  What
happens when you turn the noise off completely?

As an example in Matlab code, the equ looks something like:

chnl = [1 0.5 0.1];
equ = 1./fft(chnl,fft_size); % form zero forcing equ
data = fft(rx,fft_size).* equ; % perform bin-by-bin multiplication

where rx is the received time domain signal after the prefix removal.

CB
  


hany.kamal@ifrance.com (Hany) wrote in message news:<415b1c6f.0406090002.391f33af@posting.google.com>...

> Dear CB;
> 
>  u r right... i got a curve which is identical to the normal QPSK-SNR
> curve,AWGN case, when i simulated the system with an impulse channel
> (1,0,0,...).
> Also if i use any other channel, other than an impulse one, i get a
> curve with higher BER values than the QPSK standard one..
> 
> if the problem is occuring when taking the fft and equalizing as u r
> saying. how could i do this bin-by-bin multiplication of the channel
> and the equaliser knowing that the channel i have in the form of
> tapped-delay model, which i convert to an impulse, or frequency
> response.
> my equalizer = 1/(channel frequency response).
> 
> THANKS.
> 
> Hany
> 
> 
> 
> 
> cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406070846.3cdd42ec@posting.google.com>...
> > Hany,  
> > 
> > There are a number of things to consider here.  First off, yes, you're
> > correct, the result should be an exact cancellation of the channel
> > multipath.
> > 
> > From what I can infer from your description, the problem in the BER
> > plot occurs when you compare the performance after introducing a
> > multipath channel.  This means that you have the correct relationship
> > between the energy per bit in the frequency domain (or SNR if you
> > prefer) and the fact that you add the noise to your signal in the time
> > domain.  This can be a tricky one to understand if you're new to OFDM.
> > 
> > So, you only see a problem with the BER curve when you add the channel
> > and equaliser?  Does this mean that you see a problem when the channel
> > is an impulse? h = {1,0,0}?  If this is okay, then you're obviously
> > getting the problem when you take the fft and perform the bin-by-bin
> > multiplication of the channel and equaliser.  Can you confirm that the
> > bin-by-bin multiplication of the channel and the equaliser (forget the
> > signal and the noise for the moment) gives you a constant K in each
> > bin?
> > 
> > CB
> > 
> > 
> > hany.kamal@ifrance.com (Hany) wrote in message news:<415b1c6f.0406052229.421a1ce@posting.google.com>...
> > > CB,
> > > 
> > > I just want to know if I'm having the right results ..
> > >  as i've said before , to ma knowlage , when i equalize the multi-path
> > > channel with the exact known one (not an estimated one with preamble
> > > or whatever), i should have the same curve as if i don't have a
> > > channel at all (i'd have only the AWGN).
> > > 
> > > first : is what i'm thinking is right ??
> > > if so , the curves i have doesn't show that .. in the case of
> > > equalization , I have a BER which is higher than that I have for
> > > standard QPSK one..
> > > 
> > > Thanks 
> > > 
> > > 
> > > 
> > > cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406030722.40f9a415@posting.google.com>...
> > > > Hany,
> > > > 
> > > > Okay, so what is it that you're trying to model/show then?  Is there a
> > > > reason why you add the AWGN (with emphasis on the word White) prior to
> > > > the multipath channel?  If you want to match your system to the
> > > > theoretical BER for QPSK, then have you thought about what the BER is
> > > > actually showing?
> > > > 
> > > > CB
> > > > 
> > > > 
> > > > hany.kamal@ifrance.com (Hany) wrote in message news:<415b1c6f.0406030104.3554f1d6@posting.google.com>...
> > > > > CB,
> > > > > 
> > > > > what i do is,
> > > > > 
> > > > > QPSK  mapper --> inserting pilots --> IFFT --> cyclic perfix --> AWGN
> > > > > --> multi-path channel (tapped-delay model) --> cyclic perfix
> > > > > extraction --> FFT ---> Equalizing (ZERo forcing equalizer) --> pilots
> > > > > extraction --> de-mapping.
> > > > > 
> > > > > 
> > > > > 
> > > > > 
> > > > > cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406020533.775b340b@posting.google.com>...
> > > > > > Hany,
> > > > > > 
> > > > > > Can you explain what this bit means then, you said previously that you
> > > > > > did this:
> > > > > > 
> > > > > > > > > Thanks Laki,
> > > > > > > > > 
> > > > > > > > > ur answer could be right if I apply the effect of the channel to the
> > > > > > > > > signal then i add the noise, but what i do is adding the noise efect
> > > > > > > > > first then applying the channel effect to the sum (signal+noise), to
> > > > > > > > > get
> > > > > > > > > 
> > > > > > > > > channel*(signal+noise), so after equalization i'd have ---> 
> > > > > > > > > (channel*(signal+noise))/channel -- > signal+noise.
> > > > > > > > > 
> > > > > > > > > 
> > > > > > 
> > > > > > which looks like you're adding the noise at the output of the
> > > > > > transmitter BEFORE the mulitpath channel.  However, in your last post
> > > > > > you said
> > > > > > 
> > > > > > <snipped>
> > > > > > > I add this noise model to the signal yes, before the receiver to get a
> > > > > > > noisy signal that should be detected using a certain channel
> > > > > > > estimator.
> > > > > > 
> > > > > > so which is it? where do you add the effects of the channel and the
> > > > > > noise?
> > > > > > 
> > > > > > As others have already hinted, you seem to be implementing a zero
> > > > > > forcing equaliser, however, you should really only see deleterious
> > > > > > effects when there are nulls in the channel response.  What kind of
> > > > > > channel are you equalising?  If you make the channel a unit pulse, and
> > > > > > equalise that, what do you see in terms of the BER curve?
> > > > > > 
> > > > > > <snipped the rest of the message>
> > > > > > 
> > > > > > CB
Reply by Hany ●June 9, 20042004-06-09
Dear CB;

 u r right... i got a curve which is identical to the normal QPSK-SNR
curve,AWGN case, when i simulated the system with an impulse channel
(1,0,0,...).
Also if i use any other channel, other than an impulse one, i get a
curve with higher BER values than the QPSK standard one..

if the problem is occuring when taking the fft and equalizing as u r
saying. how could i do this bin-by-bin multiplication of the channel
and the equaliser knowing that the channel i have in the form of
tapped-delay model, which i convert to an impulse, or frequency
response.
my equalizer = 1/(channel frequency response).

THANKS.

Hany




cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406070846.3cdd42ec@posting.google.com>...

> Hany,  
> 
> There are a number of things to consider here.  First off, yes, you're
> correct, the result should be an exact cancellation of the channel
> multipath.
> 
> From what I can infer from your description, the problem in the BER
> plot occurs when you compare the performance after introducing a
> multipath channel.  This means that you have the correct relationship
> between the energy per bit in the frequency domain (or SNR if you
> prefer) and the fact that you add the noise to your signal in the time
> domain.  This can be a tricky one to understand if you're new to OFDM.
> 
> So, you only see a problem with the BER curve when you add the channel
> and equaliser?  Does this mean that you see a problem when the channel
> is an impulse? h = {1,0,0}?  If this is okay, then you're obviously
> getting the problem when you take the fft and perform the bin-by-bin
> multiplication of the channel and equaliser.  Can you confirm that the
> bin-by-bin multiplication of the channel and the equaliser (forget the
> signal and the noise for the moment) gives you a constant K in each
> bin?
> 
> CB
> 
> 
> hany.kamal@ifrance.com (Hany) wrote in message news:<415b1c6f.0406052229.421a1ce@posting.google.com>...
> > CB,
> > 
> > I just want to know if I'm having the right results ..
> >  as i've said before , to ma knowlage , when i equalize the multi-path
> > channel with the exact known one (not an estimated one with preamble
> > or whatever), i should have the same curve as if i don't have a
> > channel at all (i'd have only the AWGN).
> > 
> > first : is what i'm thinking is right ??
> > if so , the curves i have doesn't show that .. in the case of
> > equalization , I have a BER which is higher than that I have for
> > standard QPSK one..
> > 
> > Thanks 
> > 
> > 
> > 
> > cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406030722.40f9a415@posting.google.com>...
> > > Hany,
> > > 
> > > Okay, so what is it that you're trying to model/show then?  Is there a
> > > reason why you add the AWGN (with emphasis on the word White) prior to
> > > the multipath channel?  If you want to match your system to the
> > > theoretical BER for QPSK, then have you thought about what the BER is
> > > actually showing?
> > > 
> > > CB
> > > 
> > > 
> > > hany.kamal@ifrance.com (Hany) wrote in message news:<415b1c6f.0406030104.3554f1d6@posting.google.com>...
> > > > CB,
> > > > 
> > > > what i do is,
> > > > 
> > > > QPSK  mapper --> inserting pilots --> IFFT --> cyclic perfix --> AWGN
> > > > --> multi-path channel (tapped-delay model) --> cyclic perfix
> > > > extraction --> FFT ---> Equalizing (ZERo forcing equalizer) --> pilots
> > > > extraction --> de-mapping.
> > > > 
> > > > 
> > > > 
> > > > 
> > > > cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406020533.775b340b@posting.google.com>...
> > > > > Hany,
> > > > > 
> > > > > Can you explain what this bit means then, you said previously that you
> > > > > did this:
> > > > > 
> > > > > > > > Thanks Laki,
> > > > > > > > 
> > > > > > > > ur answer could be right if I apply the effect of the channel to the
> > > > > > > > signal then i add the noise, but what i do is adding the noise efect
> > > > > > > > first then applying the channel effect to the sum (signal+noise), to
> > > > > > > > get
> > > > > > > > 
> > > > > > > > channel*(signal+noise), so after equalization i'd have ---> 
> > > > > > > > (channel*(signal+noise))/channel -- > signal+noise.
> > > > > > > > 
> > > > > > > > 
> > > > > 
> > > > > which looks like you're adding the noise at the output of the
> > > > > transmitter BEFORE the mulitpath channel.  However, in your last post
> > > > > you said
> > > > > 
> > > > > <snipped>
> > > > > > I add this noise model to the signal yes, before the receiver to get a
> > > > > > noisy signal that should be detected using a certain channel
> > > > > > estimator.
> > > > > 
> > > > > so which is it? where do you add the effects of the channel and the
> > > > > noise?
> > > > > 
> > > > > As others have already hinted, you seem to be implementing a zero
> > > > > forcing equaliser, however, you should really only see deleterious
> > > > > effects when there are nulls in the channel response.  What kind of
> > > > > channel are you equalising?  If you make the channel a unit pulse, and
> > > > > equalise that, what do you see in terms of the BER curve?
> > > > > 
> > > > > <snipped the rest of the message>
> > > > > 
> > > > > CB
Reply by Col Brown ●June 7, 20042004-06-07
Hany,  

There are a number of things to consider here.  First off, yes, you're
correct, the result should be an exact cancellation of the channel
multipath.

From what I can infer from your description, the problem in the BER
plot occurs when you compare the performance after introducing a
multipath channel.  This means that you have the correct relationship
between the energy per bit in the frequency domain (or SNR if you
prefer) and the fact that you add the noise to your signal in the time
domain.  This can be a tricky one to understand if you're new to OFDM.

So, you only see a problem with the BER curve when you add the channel
and equaliser?  Does this mean that you see a problem when the channel
is an impulse? h = {1,0,0}?  If this is okay, then you're obviously
getting the problem when you take the fft and perform the bin-by-bin
multiplication of the channel and equaliser.  Can you confirm that the
bin-by-bin multiplication of the channel and the equaliser (forget the
signal and the noise for the moment) gives you a constant K in each
bin?

CB


hany.kamal@ifrance.com (Hany) wrote in message news:<415b1c6f.0406052229.421a1ce@posting.google.com>...

> CB,
> 
> I just want to know if I'm having the right results ..
>  as i've said before , to ma knowlage , when i equalize the multi-path
> channel with the exact known one (not an estimated one with preamble
> or whatever), i should have the same curve as if i don't have a
> channel at all (i'd have only the AWGN).
> 
> first : is what i'm thinking is right ??
> if so , the curves i have doesn't show that .. in the case of
> equalization , I have a BER which is higher than that I have for
> standard QPSK one..
> 
> Thanks 
> 
> 
> 
> cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406030722.40f9a415@posting.google.com>...
> > Hany,
> > 
> > Okay, so what is it that you're trying to model/show then?  Is there a
> > reason why you add the AWGN (with emphasis on the word White) prior to
> > the multipath channel?  If you want to match your system to the
> > theoretical BER for QPSK, then have you thought about what the BER is
> > actually showing?
> > 
> > CB
> > 
> > 
> > hany.kamal@ifrance.com (Hany) wrote in message news:<415b1c6f.0406030104.3554f1d6@posting.google.com>...
> > > CB,
> > > 
> > > what i do is,
> > > 
> > > QPSK  mapper --> inserting pilots --> IFFT --> cyclic perfix --> AWGN
> > > --> multi-path channel (tapped-delay model) --> cyclic perfix
> > > extraction --> FFT ---> Equalizing (ZERo forcing equalizer) --> pilots
> > > extraction --> de-mapping.
> > > 
> > > 
> > > 
> > > 
> > > cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406020533.775b340b@posting.google.com>...
> > > > Hany,
> > > > 
> > > > Can you explain what this bit means then, you said previously that you
> > > > did this:
> > > > 
> > > > > > > Thanks Laki,
> > > > > > > 
> > > > > > > ur answer could be right if I apply the effect of the channel to the
> > > > > > > signal then i add the noise, but what i do is adding the noise efect
> > > > > > > first then applying the channel effect to the sum (signal+noise), to
> > > > > > > get
> > > > > > > 
> > > > > > > channel*(signal+noise), so after equalization i'd have ---> 
> > > > > > > (channel*(signal+noise))/channel -- > signal+noise.
> > > > > > > 
> > > > > > > 
> > > > 
> > > > which looks like you're adding the noise at the output of the
> > > > transmitter BEFORE the mulitpath channel.  However, in your last post
> > > > you said
> > > > 
> > > > <snipped>
> > > > > I add this noise model to the signal yes, before the receiver to get a
> > > > > noisy signal that should be detected using a certain channel
> > > > > estimator.
> > > > 
> > > > so which is it? where do you add the effects of the channel and the
> > > > noise?
> > > > 
> > > > As others have already hinted, you seem to be implementing a zero
> > > > forcing equaliser, however, you should really only see deleterious
> > > > effects when there are nulls in the channel response.  What kind of
> > > > channel are you equalising?  If you make the channel a unit pulse, and
> > > > equalise that, what do you see in terms of the BER curve?
> > > > 
> > > > <snipped the rest of the message>
> > > > 
> > > > CB
Reply by Hany ●June 6, 20042004-06-06
CB,

I just want to know if I'm having the right results ..
 as i've said before , to ma knowlage , when i equalize the multi-path
channel with the exact known one (not an estimated one with preamble
or whatever), i should have the same curve as if i don't have a
channel at all (i'd have only the AWGN).

first : is what i'm thinking is right ??
if so , the curves i have doesn't show that .. in the case of
equalization , I have a BER which is higher than that I have for
standard QPSK one..

Thanks 



cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406030722.40f9a415@posting.google.com>...

> Hany,
> 
> Okay, so what is it that you're trying to model/show then?  Is there a
> reason why you add the AWGN (with emphasis on the word White) prior to
> the multipath channel?  If you want to match your system to the
> theoretical BER for QPSK, then have you thought about what the BER is
> actually showing?
> 
> CB
> 
> 
> hany.kamal@ifrance.com (Hany) wrote in message news:<415b1c6f.0406030104.3554f1d6@posting.google.com>...
> > CB,
> > 
> > what i do is,
> > 
> > QPSK  mapper --> inserting pilots --> IFFT --> cyclic perfix --> AWGN
> > --> multi-path channel (tapped-delay model) --> cyclic perfix
> > extraction --> FFT ---> Equalizing (ZERo forcing equalizer) --> pilots
> > extraction --> de-mapping.
> > 
> > 
> > 
> > 
> > cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406020533.775b340b@posting.google.com>...
> > > Hany,
> > > 
> > > Can you explain what this bit means then, you said previously that you
> > > did this:
> > > 
> > > > > > Thanks Laki,
> > > > > > 
> > > > > > ur answer could be right if I apply the effect of the channel to the
> > > > > > signal then i add the noise, but what i do is adding the noise efect
> > > > > > first then applying the channel effect to the sum (signal+noise), to
> > > > > > get
> > > > > > 
> > > > > > channel*(signal+noise), so after equalization i'd have ---> 
> > > > > > (channel*(signal+noise))/channel -- > signal+noise.
> > > > > > 
> > > > > > 
> > > 
> > > which looks like you're adding the noise at the output of the
> > > transmitter BEFORE the mulitpath channel.  However, in your last post
> > > you said
> > > 
> > > <snipped>
> > > > I add this noise model to the signal yes, before the receiver to get a
> > > > noisy signal that should be detected using a certain channel
> > > > estimator.
> > > 
> > > so which is it? where do you add the effects of the channel and the
> > > noise?
> > > 
> > > As others have already hinted, you seem to be implementing a zero
> > > forcing equaliser, however, you should really only see deleterious
> > > effects when there are nulls in the channel response.  What kind of
> > > channel are you equalising?  If you make the channel a unit pulse, and
> > > equalise that, what do you see in terms of the BER curve?
> > > 
> > > <snipped the rest of the message>
> > > 
> > > CB
Reply by Col Brown ●June 3, 20042004-06-03
Hany,

Okay, so what is it that you're trying to model/show then?  Is there a
reason why you add the AWGN (with emphasis on the word White) prior to
the multipath channel?  If you want to match your system to the
theoretical BER for QPSK, then have you thought about what the BER is
actually showing?

CB


hany.kamal@ifrance.com (Hany) wrote in message news:<415b1c6f.0406030104.3554f1d6@posting.google.com>...

> CB,
> 
> what i do is,
> 
> QPSK  mapper --> inserting pilots --> IFFT --> cyclic perfix --> AWGN
> --> multi-path channel (tapped-delay model) --> cyclic perfix
> extraction --> FFT ---> Equalizing (ZERo forcing equalizer) --> pilots
> extraction --> de-mapping.
> 
> 
> 
> 
> cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406020533.775b340b@posting.google.com>...
> > Hany,
> > 
> > Can you explain what this bit means then, you said previously that you
> > did this:
> > 
> > > > > Thanks Laki,
> > > > > 
> > > > > ur answer could be right if I apply the effect of the channel to the
> > > > > signal then i add the noise, but what i do is adding the noise efect
> > > > > first then applying the channel effect to the sum (signal+noise), to
> > > > > get
> > > > > 
> > > > > channel*(signal+noise), so after equalization i'd have ---> 
> > > > > (channel*(signal+noise))/channel -- > signal+noise.
> > > > > 
> > > > > 
> > 
> > which looks like you're adding the noise at the output of the
> > transmitter BEFORE the mulitpath channel.  However, in your last post
> > you said
> > 
> > <snipped>
> > > I add this noise model to the signal yes, before the receiver to get a
> > > noisy signal that should be detected using a certain channel
> > > estimator.
> > 
> > so which is it? where do you add the effects of the channel and the
> > noise?
> > 
> > As others have already hinted, you seem to be implementing a zero
> > forcing equaliser, however, you should really only see deleterious
> > effects when there are nulls in the channel response.  What kind of
> > channel are you equalising?  If you make the channel a unit pulse, and
> > equalise that, what do you see in terms of the BER curve?
> > 
> > <snipped the rest of the message>
> > 
> > CB
Reply by Hany ●June 3, 20042004-06-03
CB,

what i do is,

QPSK  mapper --> inserting pilots --> IFFT --> cyclic perfix --> AWGN
--> multi-path channel (tapped-delay model) --> cyclic perfix
extraction --> FFT ---> Equalizing (ZERo forcing equalizer) --> pilots
extraction --> de-mapping.




cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406020533.775b340b@posting.google.com>...

> Hany,
> 
> Can you explain what this bit means then, you said previously that you
> did this:
> 
> > > > Thanks Laki,
> > > > 
> > > > ur answer could be right if I apply the effect of the channel to the
> > > > signal then i add the noise, but what i do is adding the noise efect
> > > > first then applying the channel effect to the sum (signal+noise), to
> > > > get
> > > > 
> > > > channel*(signal+noise), so after equalization i'd have ---> 
> > > > (channel*(signal+noise))/channel -- > signal+noise.
> > > > 
> > > > 
> 
> which looks like you're adding the noise at the output of the
> transmitter BEFORE the mulitpath channel.  However, in your last post
> you said
> 
> <snipped>
> > I add this noise model to the signal yes, before the receiver to get a
> > noisy signal that should be detected using a certain channel
> > estimator.
> 
> so which is it? where do you add the effects of the channel and the
> noise?
> 
> As others have already hinted, you seem to be implementing a zero
> forcing equaliser, however, you should really only see deleterious
> effects when there are nulls in the channel response.  What kind of
> channel are you equalising?  If you make the channel a unit pulse, and
> equalise that, what do you see in terms of the BER curve?
> 
> <snipped the rest of the message>
> 
> CB
Reply by Col Brown ●June 2, 20042004-06-02
Hany,

Can you explain what this bit means then, you said previously that you
did this:


> > > Thanks Laki,
> > > 
> > > ur answer could be right if I apply the effect of the channel to the
> > > signal then i add the noise, but what i do is adding the noise efect
> > > first then applying the channel effect to the sum (signal+noise), to
> > > get
> > > 
> > > channel*(signal+noise), so after equalization i'd have ---> 
> > > (channel*(signal+noise))/channel -- > signal+noise.
> > > 
> > > 


which looks like you're adding the noise at the output of the
transmitter BEFORE the mulitpath channel.  However, in your last post
you said

<snipped>

> I add this noise model to the signal yes, before the receiver to get a
> noisy signal that should be detected using a certain channel
> estimator.


so which is it? where do you add the effects of the channel and the
noise?

As others have already hinted, you seem to be implementing a zero
forcing equaliser, however, you should really only see deleterious
effects when there are nulls in the channel response.  What kind of
channel are you equalising?  If you make the channel a unit pulse, and
equalise that, what do you see in terms of the BER curve?

<snipped the rest of the message>

CB
Reply by Hany ●June 2, 20042004-06-02
CB

what i'm trying to do is to examin the system with the AWGN.
so i model the noise to get Eb/No virsus BER curve.
where Eb = energy/Bit .

I add this noise model to the signal yes, before the receiver to get a
noisy signal that should be detected using a certain channel
estimator.

What i do is that i add the multi-path channel effect in the time
domain (after IFFT process, according to HIPERLAN/2 standard) and i do
the equalization in the frequency domain (after FFT process in the
receiver).

The exact channel frequency response that i use to equalize the system
is the one i catculated from the taped delay model I affected the
system with.
so i guess convolving the channel and the equalizer would mean that i
use the calculated CFR and convolve it with itself which would have no
meaning.
 i really didn't understand that you ment by ""substitute the signal S

> for an impulse function (no noise) and examine the signal at the
> output of the equaliser? "".


could u please explain this?

i feel that the problem is concerning the noise modeling not the
equalization , casue i've said , when i simulated the system with no
noise at all. using the same channel effect and the same concept of
equalization, i get ZERO BER which means that the equalizer is workin
well, isn't it?


 THanks alot

Hany

cb135@hotmail.com (Col Brown) wrote in message news:<a254af6b.0406011110.1768eba7@posting.google.com>...

> Hany
> 
> That's an interesting way to model the system, but i'll come back to
> that.
> 
> What I think you're trying to do is this
> 
> 
> H*S / H_est = S
> 
> where H = channel and H_est is the channel estimate (therefore
> inv(H_est) represents the inverse channel which you're calling an
> equaliser).
> 
> Now, have you tried looking at the result of simply convolving the
> channel and the equaliser?  H / H_est = 1  What do you get out of
> this?
> 
> If that is difficult to obtain, then can you substitute the signal S
> for an impulse function (no noise) and examine the signal at the
> output of the equaliser?  What do you get?
> 
> The reason i say that it's an interesting way to model the system, is
> because you seem to be adding noise prior to the channel.  One form of
> noise is typically added to model the thermal noise effects at the
> receiver.  What you're doing it seems is receiving the signal
> noiselessly and are adding noise at the output of the transmitter. 
> Now, this may not be a bad thing, because you could be trying to model
> some form of interference, but you haven't told us anything about that
> yet, so I can only wait with bated breath until you explain what
> you're trying to do!
> 
> CB
> 
> 
> 
> 
> hany.kamal@ifrance.com (Hany) wrote in message news:<415b1c6f.0406010011.60e291f0@posting.google.com>...
> > Thanks Laki,
> > 
> > ur answer could be right if I apply the effect of the channel to the
> > signal then i add the noise, but what i do is adding the noise efect
> > first then applying the channel effect to the sum (signal+noise), to
> > get
> > 
> > channel*(signal+noise), so after equalization i'd have ---> 
> > (channel*(signal+noise))/channel -- > signal+noise.
> > 
> > 
> > hope u got what i mean .!
> > 
> > 
> > thanks alot Laki for ur participation
> > 
> > Hany
Reply by Col Brown ●June 1, 20042004-06-01
Hany

That's an interesting way to model the system, but i'll come back to
that.

What I think you're trying to do is this


H*S / H_est = S

where H = channel and H_est is the channel estimate (therefore
inv(H_est) represents the inverse channel which you're calling an
equaliser).

Now, have you tried looking at the result of simply convolving the
channel and the equaliser?  H / H_est = 1  What do you get out of
this?

If that is difficult to obtain, then can you substitute the signal S
for an impulse function (no noise) and examine the signal at the
output of the equaliser?  What do you get?

The reason i say that it's an interesting way to model the system, is
because you seem to be adding noise prior to the channel.  One form of
noise is typically added to model the thermal noise effects at the
receiver.  What you're doing it seems is receiving the signal
noiselessly and are adding noise at the output of the transmitter. 
Now, this may not be a bad thing, because you could be trying to model
some form of interference, but you haven't told us anything about that
yet, so I can only wait with bated breath until you explain what
you're trying to do!

CB




hany.kamal@ifrance.com (Hany) wrote in message news:<415b1c6f.0406010011.60e291f0@posting.google.com>...

> Thanks Laki,
> 
> ur answer could be right if I apply the effect of the channel to the
> signal then i add the noise, but what i do is adding the noise efect
> first then applying the channel effect to the sum (signal+noise), to
> get
> 
> channel*(signal+noise), so after equalization i'd have ---> 
> (channel*(signal+noise))/channel -- > signal+noise.
> 
> 
> hope u got what i mean .!
> 
> 
> thanks alot Laki for ur participation
> 
> Hany