Machine learning algorithms for 5G optical networks
(English) Throughout this thesis we addressed 5G network challenges. The non-coherent optical modulation tradeoff has been addressed, as we proposed FBMC and artificial intelligence-based schemes. The proposed schemes have been proved to solve the spectral and the power efficiency trade off in non-...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Estado: | Versión publicada |
| Fecha de publicación: | 2023 |
| País: | España |
| Institución: | CBUC, CESCA |
| Repositorio: | TDR. Tesis Doctorales en Red |
| OAI Identifier: | oai:www.tdx.cat:10803/690170 |
| Acceso en línea: | http://hdl.handle.net/10803/690170 https://dx.doi.org/10.5821/dissertation-2117-403554 |
| Access Level: | acceso abierto |
| Palabra clave: | Àrees temàtiques de la UPC::Enginyeria de la telecomunicació 621.3 |
| Sumario: | (English) Throughout this thesis we addressed 5G network challenges. The non-coherent optical modulation tradeoff has been addressed, as we proposed FBMC and artificial intelligence-based schemes. The proposed schemes have been proved to solve the spectral and the power efficiency trade off in non- coherent optical modulation schemes. Then we addressed the problem of interference management for indoor and outdoor networks. We evaluated the proposed scheme for VLC network and for the front haul optical network. The proposed schemes show significant enhancement in the network behavior regarding power saving and network management. Chapter 2 presented a non-coherent optical modulation transceiver based on artificial neural network and decision regression tree. It is shown that nonlinear regression provides the best system performance, over linear and polynomial regression, as it offers better feature extraction. While the transceiver based on decision regression tree eliminates the clipping distortion produced by clipping the transmitted signal over optical channel. The proposed transceiver overcomes the tradeoff between the spectral and the power efficiency of the noncoherent optical modulation, moreover it shows better performance and enhanced BER for high subcarrier spacing. On the other side, reducing the subcarrier spacing highly influences the RDT scheme, as it increases the BER and the complexity of the scheme. It is shown that the AI based transceiver enhances the BER of the DCO-OFDM, as the BER reaches 10^-3 at 5 db SNR with the same spectral efficiency. While in chapter 3, the optical ACO-FBMC scheme for direct detection modulation was presented. We have shown that the proposed scheme eliminates self-frame interference and suffers from inter-frame interference. Interframe interference is eliminated using the proposed iterative receiver. Also, the ACO-FBMC enhances the BER performance of the ACO-OFDM with the perfect rectangular pulse shaping and eliminating the emission of the filter bank out of the band. Our system model has higher spectral efficiency than the classic ACO-OFDM, as FBMC shows better spectral efficiency by removing the guard interval. Moreover, it has been shown that the ACO-FBMC with 8 overlapping factors improves the BER performance of the ACO-OFDM by 4 dB due to the perfect rectangular pulse shaping and the elimination of the out-band emission of the filter bank. In chapter 4, we addressed the problem of interference management in a VLC indoor system with multi-user access. The performance of the system applying the SRF technique compared to UFR and PFR. The results showed that the proposed approach improves system performance in terms of total system throughput, outage probability, and SINR. It solves the problem of high SINR at the cell edge as experienced by UFR and improved the overall system throughput as compared with PFR. In chapter 5, we proposed RRH-BBU assignment based on clustering algorithm that targets minimizing the power consumption and the inter BBU handover. The proposed algorithm computes the required number of installed BBUs to accommodate the maximum traffic load, deploys time series clustering as temporal clustering method, and applies DBSCAN algorithm to divide each temporal cluster into several spatial cluster based on the cell location. Then the problem of assigning RRH of each cluster is described as a bin packing optimization problem to find the optimum number of BBUs for each cluster. The proposed algorithm has been validated using real world CDR of Milan city and it is verified the published Milan land use map. The inter BBU handover signals have been enhanced by assigning near RRHs to the same spatial cluster and same BBU, to avoid inter BBU handover. It is shown that the algorithm reduces the total power consumption of the current deployed network by 28.8 %, by assigning all the active RRHs tocertain BBUs based on the traffic load and switching off the unassigned BBUs. |
|---|