Ημερολόγιο Εκδηλώσεων

Περιγραφή:
ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ Σχολή Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών Πρόγραμμα Μεταπτυχιακών Σπουδών ΠΑΡΟΥΣΙΑΣΗ ΜΕΤΑΠΤΥΧΙΑΚΗΣ ΕΡΓΑΣΙΑΣ ΝΙΚΟΛΑΟΥ ΨΑΡΟΜΑΝΩΛΑΚΗ με θέμα Approximate Message Passing for Joint Data and Channel Estimation in Multi-cell Massive MIMO Εξεταστική Επιτροπή Αναπληρωτής Καθηγητής Άγγελος Μπλέτσας (επιβλέπων) Καθηγητής Μιχαήλ Πατεράκης (Σχολή ΗΜΜΥ, Π.Κ.) Αναπληρωτής Καθηγητής Γεώργιος Καρυστινός (Σχολή ΗΜΜΥ, Π.Κ.) Abstract This work studies joint channel estimation and data detection in uplink, multi-cell (interference-limited), large-scale antenna (i.e., massive multiple input multiple output - MIMO) networks. With massive MIMO, future 5G networks will be able to communicate more data within a given spectrum. Realistic assumptions are considered, accounting for path-loss and pilot contamination from adjacent cells. First, a baseline prior art algorithm is presented, that initially utilizes minimum mean-squared error (MMSE) channel estimator and then linear MMSE detector (given estimated channel) for data detection (abbreviated as SCED). Second, a novel iterative algorithm for joint channel estimation and linear data detection (JCED-Iterative) is proposed, where estimation and detection are done iteratively, until convergence. Third, based on joint data/channel estimation bilinear generalized approximate message passing (JCD-GAMP) proposed recently, which is an approximation of the sum-product belief propagation algorithm and offers scalar computation operations, two inference-based algorithms for joint MMSE channel and data estimation are developed; the first algorithm is an extension of JCD-GAMP to the multi-cell setting, exploiting bilinear approximate message passing and pilot symbols used both in the initialization, as well as the message passing phase (MC-JCD-GAMP); the second inference algorithm uses pilot symbols only for initialization (MC-JCD-GAMP-D). Simulation results underline the critical importance of damping and initialization fine-tuning in the convergence of the inference-based algorithms. It is found that joint channel estimation and detection algorithms, in few iterations, can achieve better performance compared to SCED. Moreover, it is shown that pilot symbols are helpful in the initialization phase and not in the subsequent message passing procedure, i.e. MC-JCD-GAMP and MC-JCD-GAMP-D offer similar results. The two latter have better performance compared to compared to JCED-Iterative at high SNR, but JCED-Iterative algorithm converges faster in all cases. Hence, there exist interesting convergence-accuracy and tuning-complexity trade-offs between the above practical algorithms for massive MIMO uplink networks.