Could you please send me the simulation code of Fig. 1 through email? I wanna reproduce this graph. ]]>

I would like to point out one caveat in the description of the second point ‘2.’:

Although it is true that the expectation in E[log(1+”SINR”)] is over the quantities that are random within the duration of a codeword (such as the small-scale channel features), it does not include the averaging over ‘the randomness incurred by noise (AWGN)’ because that is already subsumed in the mutual information form log(1+”SINR”). Specifically, for a given “SINR” value, i.e., with the large- and small-scale channel features conditioned upon,

\log(1+”SINR”) = I(s, \sqrt{“SINR”} \, s + z)

with z being AWGN and the signal s being Gaussian as well. This mutual information involves the expectation over the ‘the randomness incurred by noise (AWGN)’.

Thank you very much for the new post,

I got a little bit confused about “2.”,

if I understand correctly, first we should make an average with respect to variables of the first level of randomness that we have for the input-output of the system, given the variables of the second level of randomness, which are constant within the particular intervals of interest. Then, you explained that the CDF curves of the second level of randomness variables can be used to evaluate the data-rates.

What I don’t understand, is that why we can’t make an average for the second level of randomness instead of CDF plot…?

For instance, I also see in “Massive MIMO With Max-Min Power Control in Line-of-Sight Propagation Environment”, that they use the CDF plot for the SINR evaluation of LOS scenarios (Fig 3. and 4.), and not the average.

Best Regards,

Ashkan