Time-varying channel

Hello People,

Basically I am stuck at this problem for quite a long time. Basically I
have got two problems:

1. I would like to model time varying channel.

Now, is it possible to model time varying multipath channel at baseband
where I have just one sample per symbol, no pulse shaping, no carrier
modulation.

2. And recently I did one simulation to replicate proakis resluts for
adaptive equaliser. I am getting exactly what I should get.

But when I tried to do the same using pulse shaping(square pulse) and
carrier modulation, I am not getting the same plot. If I remove the noise,
I am getting zero error. So I think there is some problem with noise
scaling. 

Now the way I scale the noise is:
sigma=sqrt(Eb/(2*R*EbNO_lin)), where R is my code rate.

Now, again to cross check, I did simple AWGN simulation at passband (no
multipath), keeping the same parameters, I am getting exact plot.

This formula is correct for one sample per symbol, but is this same for
passband as well.?

My sampling freq is 48 KHz. Also my pulse shaping filter has unit energy.

Your guidance will be greatly appreciated. 

Thanks,

Chintan

On Sep 3, 12:27&#4294967295;pm, "cpshah99" <cpsha...@rediffmail.com> wrote:
> Hello People,
>
> Basically I am stuck at this problem for quite a long time. Basically I
> have got two problems:
>
> 1. I would like to model time varying channel.
>
> Now, is it possible to model time varying multipath channel at baseband
> where I have just one sample per symbol, no pulse shaping, no carrier
> modulation.
>
> 2. And recently I did one simulation to replicate proakis resluts for
> adaptive equaliser. I am getting exactly what I should get.
>
> But when I tried to do the same using pulse shaping(square pulse) and
> carrier modulation, I am not getting the same plot. If I remove the noise,
> I am getting zero error. So I think there is some problem with noise
> scaling.
>

How about trying it at baseband, with pulse shaping?  Does it work?

What "SNR" do you need to get your passband system to "work"?
Does any small amount of noise make it not work?

>On Sep 3, 12:27=A0pm, "cpshah99" <cpsha...@rediffmail.com> wrote:
>> Hello People,
>>
>> Basically I am stuck at this problem for quite a long time. Basically
I
>> have got two problems:
>>
>> 1. I would like to model time varying channel.
>>
>> Now, is it possible to model time varying multipath channel at
baseband
>> where I have just one sample per symbol, no pulse shaping, no carrier
>> modulation.
>>
>> 2. And recently I did one simulation to replicate proakis resluts for
>> adaptive equaliser. I am getting exactly what I should get.
>>
>> But when I tried to do the same using pulse shaping(square pulse) and
>> carrier modulation, I am not getting the same plot. If I remove the
noise=
>,
>> I am getting zero error. So I think there is some problem with noise
>> scaling.
>>
>
>How about trying it at baseband, with pulse shaping?  Does it work?
>
>What "SNR" do you need to get your passband system to "work"?
>Does any small amount of noise make it not work?
>
%%%

Hi Julius

Thanks for replying.

I am sorry but I did not get ur first que.

And I need to get the BER plot for the SNR range of 0 to 10 dB. And for
very small amount of noise, it gives me zero error, which says that my
receiver is working good.

Thanks again.

Chintan

On Sep 4, 3:24&#4294967295;am, "cpshah99" <cpsha...@rediffmail.com> wrote:
>
> Hi Julius
>
> Thanks for replying.
>
> I am sorry but I did not get ur first que.
>
> And I need to get the BER plot for the SNR range of 0 to 10 dB. And for
> very small amount of noise, it gives me zero error, which says that my
> receiver is working good.
>
> Thanks again.
>
> Chintan

Sounds like you have a scaling error in computing the
noise variance that you need.  I don't know how to check
this since you speak in non-quantitative terms :-P.

Let me try to be very specific:  how much is the SNR shift
between the baseband and passband simulation?

>On Sep 4, 3:24=A0am, "cpshah99" <cpsha...@rediffmail.com> wrote:
>>
>> Hi Julius
>>
>> Thanks for replying.
>>
>> I am sorry but I did not get ur first que.
>>
>> And I need to get the BER plot for the SNR range of 0 to 10 dB. And
for
>> very small amount of noise, it gives me zero error, which says that my
>> receiver is working good.
>>
>> Thanks again.
>>
>> Chintan
>
>Sounds like you have a scaling error in computing the
>noise variance that you need.  I don't know how to check
>this since you speak in non-quantitative terms :-P.
>
>Let me try to be very specific:  how much is the SNR shift
>between the baseband and passband simulation?
>
%%%%


HI Julius

Sorry for this. The shift is very huge, for baseband(one sample per
symbol) my BER is 10^-2 at 4 dB and for passband(12 samples per symbol) it
is at 20dB!!!!!!!!!!!

Thanks again and sorry for not talking in technical terms.

Chintan

On Sep 4, 7:26&#4294967295;am, "cpshah99" <cpsha...@rediffmail.com> wrote:
>
> HI Julius
>
> Sorry for this. The shift is very huge, for baseband(one sample per
> symbol) my BER is 10^-2 at 4 dB and for passband(12 samples per symbol) it
> is at 20dB!!!!!!!!!!!
>
> Thanks again and sorry for not talking in technical terms.
>
> Chintan

So if you plot BER vs. SNR for baseband vs. passband, they are
identical,
except for the 16 dB offset in SNR, correct?

If this is the case, most likely it's a scaling error somewhere.
Let's
call your baseband simulation A and passband simulation B.  For sure
you also have a baseband part for B, right?

1.  Set the Eb to be the same for A and B, and set noise to 0 for
both.
     Then look at the baseband receiver part of both A and B.  Are
they
     the same amplitude-wise?

2.  Set the SNR to be the same for both A and B, and look at the
variance
     of the noise part of both A and B in the baseband receiver part.
Are
     they the same?

Hope this helps.  Let's not get too worked up and frustrated before
trying a side-by-side comparison.

Julius

>On Sep 4, 7:26=A0am, "cpshah99" <cpsha...@rediffmail.com> wrote:
>>
>> HI Julius
>>
>> Sorry for this. The shift is very huge, for baseband(one sample per
>> symbol) my BER is 10^-2 at 4 dB and for passband(12 samples per symbol)
i=
>t
>> is at 20dB!!!!!!!!!!!
>>
>> Thanks again and sorry for not talking in technical terms.
>>
>> Chintan
>
>So if you plot BER vs. SNR for baseband vs. passband, they are
>identical,
>except for the 16 dB offset in SNR, correct?
>
>If this is the case, most likely it's a scaling error somewhere.
>Let's
>call your baseband simulation A and passband simulation B.  For sure
>you also have a baseband part for B, right?
>
>1.  Set the Eb to be the same for A and B, and set noise to 0 for
>both.
>     Then look at the baseband receiver part of both A and B.  Are
>they
>     the same amplitude-wise?
>
>2.  Set the SNR to be the same for both A and B, and look at the
>variance
>     of the noise part of both A and B in the baseband receiver part.
>Are
>     they the same?
>
>Hope this helps.  Let's not get too worked up and frustrated before
>trying a side-by-side comparison.
>
>Julius
>
%%%%

Hi Julius,

I did one experiment, just to verify something. The system is uncoded and
just AWGN, no multipath. 

Transmitter-1:

data symbols(BPSK), pulse shaping(square pulse), carrier modulation,
noise.

Receiver-1:

Carrier Demodulation, Matched filtering, detection(+1 or -1), BER

I am getting exactly what I should get, i.e. theoratical BER=simulated
BER.


Now, I am changing my transmitter, in the sense that, along with data
symbols, I am putting some training sequence.

And then, at receiver, after matched filtering, I am doing correlation of
training sequence with the o/p of MF to find the start of packet. After
finding the start, rest of the processing is same.

But this time, the BER is 3 dB away from the theoratical BER.


Any hint?


Thanks again.

Chintan

On Sep 4, 10:04 am, "cpshah99" <cpsha...@rediffmail.com> wrote:

> Hi Julius,
>
> I did one experiment, just to verify something. The system is uncoded and
> just AWGN, no multipath.
>
> Transmitter-1:
>
> data symbols(BPSK), pulse shaping(square pulse), carrier modulation,
> noise.
>
> Receiver-1:
>
> Carrier Demodulation, Matched filtering, detection(+1 or -1), BER
>
> I am getting exactly what I should get, i.e. theoratical BER=simulated
> BER.
>
> Now, I am changing my transmitter, in the sense that, along with data
> symbols, I am putting some training sequence.
>
> And then, at receiver, after matched filtering, I am doing correlation of
> training sequence with the o/p of MF to find the start of packet. After
> finding the start, rest of the processing is same.
>
> But this time, the BER is 3 dB away from the theoratical BER.
>
> Any hint?
>
> Thanks again.
>
> Chintan

Can't we stay on topic, please.  It's clear to me that you are making
normalization mistakes.  Let's solve one problem at a time.  Perfect
synchronization, uncoded, simplest channel possible.

There are so many things to check, even including how you define
your EbNo.  Things get more complicated when you have training
sequence, etc.

Divide and conquer, man.

>On Sep 4, 10:04 am, "cpshah99" <cpsha...@rediffmail.com> wrote:
>
>> Hi Julius,
>>
>> I did one experiment, just to verify something. The system is uncoded
and
>> just AWGN, no multipath.
>>
>> Transmitter-1:
>>
>> data symbols(BPSK), pulse shaping(square pulse), carrier modulation,
>> noise.
>>
>> Receiver-1:
>>
>> Carrier Demodulation, Matched filtering, detection(+1 or -1), BER
>>
>> I am getting exactly what I should get, i.e. theoratical BER=simulated
>> BER.
>>
>> Now, I am changing my transmitter, in the sense that, along with data
>> symbols, I am putting some training sequence.
>>
>> And then, at receiver, after matched filtering, I am doing correlation
of
>> training sequence with the o/p of MF to find the start of packet.
After
>> finding the start, rest of the processing is same.
>>
>> But this time, the BER is 3 dB away from the theoratical BER.
>>
>> Any hint?
>>
>> Thanks again.
>>
>> Chintan
>
>Can't we stay on topic, please.  It's clear to me that you are making
>normalization mistakes.  Let's solve one problem at a time.  Perfect
>synchronization, uncoded, simplest channel possible.
>
>There are so many things to check, even including how you define
>your EbNo.  Things get more complicated when you have training
>sequence, etc.
>
>Divide and conquer, man.
>

%%%%

Hi

Ok, I tried to do what u said earlier. Simulation A and B.

As you said, in simulation B, I do have baseband part.

Now, the system parameters are:

uncoded, BPSK, perfect synchronisation and simplest channel AWGN (no
multipath, no training).

1. Eb=1, Noise=0;
   In both the cases, they are same amplitude wise.

2. Set same SNR for both the cases and check variance of noise
   It is same in both the cases.

And when I run both simulations, the BER plot is exactly same.

Now, instead of AWGN, if I use simplest frequency selective channel, the
BER is not same. As I am using adaptive DFE, I need to use training
sequence and I must correlate find the start of the training.

Thanks again. Will definately solve the problem one by one.

But I think the problem is with correlation.

Thanks

Chintan

julius  <juliusk@gmail.com> wrote:

>Sounds like you have a scaling error in computing the
>noise variance that you need.  I don't know how to check
>this since you speak in non-quantitative terms :-P.

That is one highly likely problem (although it does not
preclue other parallel problems).

I have learned that when injecting noise in a channel
model, it is best to also measure the SNR at the input
of the demodulator or decoder within the modem.
Turn off the noise, store a signal sample, turn the
noise back on and store the noise signal, and measure
SNR by calculating the RMS value of these signals
and their difference signal.

Steve