The research during my Ph.D. studies at University of California, Irvine.
The system will generate random sequence of complex symbols (constellations of BPSK, QPSK, 4QAM, 16QAM, and 64QAM). There will be a precoder to encode the sequences in order to transmit the symbols
After receiving the symbols, the optimizer categorizes the sequence in order to detect each individual symbol
Using statistics methods (such as maximum likelihood and maximum a posteriori), supervised machine learning techniques (logistic regression, nearest neighbor, decision trees, support vector machines, naive Bayes, and random forests), and eventually, unsupervised machine learning techniques (clustering and k-means), the system can detect which symbol was originally transmitted.