In this subsection, the computational complexity of IMLD is investigated for each step of the algorithm by counting the number of complex operations.
It is observed that the proposed IMLD with K = 16 has an SNR gain of 2.
L] = 2 only takes two selected candidate solution vectors into account in each symbol interval, whereas IMLD needs to consider [?
t] = 4, the IMLD scheme achieves substantial gains over the conventional ML detector in the presence of CSI errors by taking the covariance matrix of the effective noise into account.
However, it is found that the ILSD scheme attains a significant reduction in the computational load with respect to the IMLD scheme.