The question is based on a research article titled, "Novel Robust Blind Equalizer for QAM Signals Using Iterative Weighted-Least-Mean-Square Algorithm"

The Authors have proposed an equalization scheme for channel equalization for QAM symbols. The Authors in the last page, Table I present their results. The first column result is for Constant Modulus lagorithm given below:

5 (dB) 6.36 ; 10 (dB) 9.90 ;15 (dB) 15.12; 25 (dB) 18.0

with the plot img

The Authors say that the values are in dB, but I don't undertstand why the SIR curve is increasing. In general, lesser the value of SIR, better is the perofrmance. But the curve shows the opposite. So, does CMA not work for QAM?

Moreover, their proposed method the values for which are also higher for high SNR. So, when the noise level is low both CMA and their method is perfroming badly?


The constant modulus algorithm does not work for QAM because QAM is not a constant modulus signal (amplitude at the symbol decisions is not constant), such as BPSK and QPSK.

SIR is the "Signal to Interference Ratio" and the greater the SIR the better the performance, not the other way around as you describe. With a higher SIR, the signal is stronger relative to the interference.

  • $\begingroup$ Thank you for your reply. Then what algorithms are used for QAM and in practical (industry) in equalization (blind). I have not found any reference or a text book which algorithm is commonly used for blind equalization for QAM signals. Another thing, in the paper, the Table shows that the values for high SNR is lower than the values at low SNR. I cannot understand this trend, is this the usual case? would you please update your answer witht the usual trend that is observed? $\endgroup$ – Srishti M May 28 '17 at 16:50
  • $\begingroup$ The trend looks right to me so I am not sure I understand your question. The SIR is increasing as the SNR is increasing, is that not what you would expect? Decision directed equalization is a reasonable approach for blind equalization as long as the threshold error rate is high enough so that false decisions to not corrupt the convergence. I do not readily have the exact threshold but believe anything higher than a BER of 10^-2 would be sufficient. $\endgroup$ – Dan Boschen May 28 '17 at 17:58
  • $\begingroup$ Another approach is to use a QPSK pattern in the header where CMA can be done, and in the process improve the BER to below threshold where then a decision directed adaptive equalization can be done on the remaining payload with full QAM constellation. $\endgroup$ – Dan Boschen May 28 '17 at 17:59
  • 1
    $\begingroup$ @SrishtiM See this post I have made on common equalization algorithms used in industry: dsp.stackexchange.com/questions/40939/…. You either have a training sequence that is used or you make bit decisions on your received signal and use that as your training sequence (which is "Decision Directed" with the constraints I described above). For frequency selective fading conditions (when you have delay spreads that exceed the symbol duration) an alternate decision-feedback equalization (DFE) approach is recommended. $\endgroup$ – Dan Boschen May 28 '17 at 18:55
  • $\begingroup$ Thank you for the link, very useful. I will read through it carefully. I have 3 quick questions - (1) is least mean square based equalization blind or training based (2) is decision directed blind or training based (non-blind)? (3) Are these 2 algorithms applicable to QAM ? $\endgroup$ – Srishti M May 28 '17 at 21:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.