QAM With turbo coding at High SNR outputting all zeros

Hello,

I've been having trouble implementing 16 and 32-QAM with turbo coding. I've been using the qammod and qamdemod functions as well as the Comm.TurboEncoder and Comm.TurboDecoder functions from the communications toolbox to do this. At low SNR it gives me no issues. However, when the SNR gets high enough (i.e. 15 dB), I notice that the ouptut of the qamdemodulate function contains multiple values that are registered as infinite. Additionally, when this output is fed into the Comm.TurboDecoder functions, I notice that the output is all zeros which is incorrect. I'm thinking that it's related to the noise variance (i.e. 10^(-SNR/10)) I pass into the qamdemod function in order to perform soft demodulation. Can anyone help me understand why this happening as well as why this isn't happening at low SNR?

Answer 1

[Sai Sri Pathuri]

When output type is LLR and the NoiseVariance value is very small (i.e., SNR is very high), LLR computations may yield Inf or -Inf because the LLR algorithm would involve computing exponentials of very large or very small numbers using finite precision arithmetic. In such cases, use of approximate LLR (approxllr) is recommended, as its algorithm does not involve computing exponentials.

You may refer the 'Tip' mentioned in 'Noise Variance' section in documentation of qamdemod

https://in.mathworks.com/help/comm/ref/qamdemod.html#mw_8898ce8d-4c00-4a4a-a7cc-bb1af17a020c