Dear Balaji,

For solving your problem i suggest to you links and attached files in topics.

- Equalize using maximum likelihood sequence estimation - MATLAB

https://www.mathworks.com › ... › Equalization

-wireless communications - UC Davis CS

www.cs.ucdavis.edu/~liu/289I/Material/book-goldsmith.pdf

Best regards

Conference Paper Adaptive equalization and interference cancellation for wire...

Thesis Electronic Equalization by Maximum-Likelihood Sequence Estim...

Hi,

Probably the equalizer should reset each SNR loop. I did it through a workaround but you can check the MLSE structure and do reset each snr loop.

check this code, it works!

clc;clear;

close all;

ii=1;

snr_vec=-10:1:30;

for snr_i=1:length(snr_vec)

snr=snr_vec(snr_i);

N = 100000;

%chCoeffs = [0.986+0.5i;0.845;0.337;0.32345+0.31i;0.2];

chCoeffs = [0.5;0.00000000001];

mlse = comm.MLSEEqualizer('TracebackDepth',10,'Channel',chCoeffs,'Constellation',(1/sqrt(2))*qammod(0:3,4));

errorate = comm.ErrorRate;

Tx_Data = randi([1 2],N,1)-1;

modData = (1/sqrt(2))*qammod(Tx_Data,4,'InputType','bit');

chanOutput = conv(chCoeffs,modData);%filter(chCoeffs,1,modData);

chanOutput = chanOutput(1:end-length(chCoeffs)+1);

noise = (randn(1,length(chanOutput))+1i*randn(1,length(chanOutput)))/sqrt(2)/sqrt(db2pow(snr));

Rx_data = chanOutput+noise.';%awgn(chanOutput,snr);

equalized_Data = mlse(Rx_data);

% clear mlse

% mlse = comm.MLSEEqualizer('TracebackDepth',10,'Channel',chCoeffs,'Constellation',(1/sqrt(2))*qammod(0:3,4));

% reset(mlse)

demodData = qamdemod(equalized_Data,4,'OutputType','bit');

errorStats = errorate(Tx_Data,demodData);

% errorStats(2)

ber(snr_i) = errorStats(1);

save ber ber snr_vec;

NumofErr(snr_i) = errorStats(2);

% ii=ii+1;

clear;

load ber;

end

semilogy(1:length(snr_vec),ber)

grid on;

A first simple test to make sure the equalization is well performed is to delete the AWGN box (SNR=+inf in MMSE equalizer), and compare the transmitted complex symbols to the received ones. Normally, they should be equal, if not... the problem comes from the equalizer.

I have done similar simulations many years ago. A very important factor is to ensure synchronism between the generated and received symbols. If you compare inappropriate symbols with each other you will never get low BER (or SER). I propose to remove multipath temporarily or to select it so small that for high SNR MLSE will produce good symbol estimates. Observe both  input and output symbol sequences and check if they are in sychronism or are shifted by a certain number of symbols. The MLSE equalizer introduces delay in decisions!   

Dear Mahmoud,

I have used your idea of resetting MLSE equalizer for each snr loop. I have done this by initiating a new equalizer object for each snr loop. But when I run my code and your code, the results do not match.

what could be the reason?

Only difference between our code is, I have used some inbuilt functions like "AWGN" - to add noise, "Filter" instead of "Convolution" etc. But I don't think it is because of this inbuilt functions.

I think the problem is in calculating the receive SNR.

Should I be considering SINR instead of just snr?

While plotting the graphs I haven't considered the effect of ISI-channel in calculating the SNR.

Correct me please.

I am attaching the graph for your reference.

Thanks every one for replying to my query.

Hi Balaji,

Thanks for follow up!

I have something wrong in my code which is the plotting,

it should be "semilogy(snr_vec,ber)" not "semilogy(1:length(snr_vec),ber)".

And this will make your curve exactly as mine.

However, both of them are not correct!!!!

The reason is that the noise addition in both your code and in my code assumes that the transmitted data is power-normalized which is not correct. The channel you used is not power-normalized (it actually multiplies the amplitude by half, equivalently scale the power by 0.25). This factor makes 6 dB difference in the BER.

- To fix that in your code, you should use:

"Rx_data = awgn(chanOutput,snr,'measured');"

The option 'measured' measures the signal power first then scale the noise based on both SNR and the input transmitted power.

Alternatively, to fix in my code, you should use:

" noise = (randn(1,length(chanOutput))+1i*randn(1,length(chanOutput)))/sqrt(2)/sqrt(db2pow(snr))*sqrt(var(chanOutput));

Rx_data = chanOutput+noise.';%awgn(chanOutput,snr);"

This way, I scale the noise with the transmitted signal power.

I checked the curve after that correction and it should match the theoretical BER for the QPSK.

Please let me know if you still have a problem.

Thanks,

Mahmoud

1. I'm using comm.MIMOChannel system object to generate SUI channel implies response, h in SISO system. Then I take channel coefficients like this:

[channel_out_signal, pathGains] = step(h, SIGNAL);

I have made the channel time invariant so I take only 3-taps of it and use them as:

signal_1 = filter(channel_coefficients, 1, SIGNAL);

At the receiver, I use comm.MLSEEqualizer system object as this:

mlse = comm.MLSEEqualizer('Channel', channel_coefficients, ...

'TracebackDepth', 16);

equalized_signal = step(mlse, rcvd);

But BER curves do not converge. What am I doing wrong?

2. I don't understand the reason of taking channel coefficients like this:

%chCoeffs = [0.986+0.5i;0.845;0.337;0.32345+0.31i;0.2];

chCoeffs = [0.5;0.00000000001];