Accurate monitoring of received signal quality provides a precious source of information to secure optical networks performance and to guarantee transmission quality. During network operation, several kinds of soft failures (as opposed to hard failures, where signal is totally disrupted) can affect signal quality and induce anomalies in the BER at the receiver, ultimately leading to packet losses or even service disruption. Hence, a solid mechanism for soft failure detection (i.e., recognize anomalies due to failure occurrences), localization (i.e., identify where in the network the failure occurred), and identification (i.e., understand the actual cause of the failure) is crucial, as it may be used by operators to perform traffic re-routing and rapid failure recovery. New generation coherent transponders give the opportunity to monitor several parameters associated to optical signal transmission, such as Optical Signal-to-Noise-Ratio (OSNR), Q-factor, or pre-Forward Error Connection Bit Error Rate (pre-FEC BER). The amount of generated data is enormous and requires advanced data analysis techniques to extract useful information for these large data-sets. In this context, techniques from machine learning (ML) discipline are regarded as a strong candidate to address this issue, as ML enables automatized network self-configuration and fast decision-making by effectively leveraging the plethora of data that can be retrieved via network monitors. Previous works have already leveraged ML to perform monitoring and even failure detection in optical networks. Even though some initial research works have appeared on this topic, several questions are still pending regarding, e.g., which information is more important to be used, how often this information shall be sampled/collected by the monitors, and which ML technique (among the large set of already existing and well established tools) is better suited for soft-failure detection based on BER analysis at signal receiver. To address some of these pending questions, in this thesis we provide the following contributions: 1) we define a framework for BER anomaly detection based on monitored BER data; here we employ different ML algorithms, for which we assess the trade-off between complexity and prediction accuracy; 2) we perform a sensitivity analysis of the accuracy of the various ML techniques in order to identify the right BER sampling time, i.e., how often BER values should be collected and analyzed; 3) we propose, to the best of our knowledge for the first time in literature, a ML-based methodology to discriminate among different sources of soft failure, enabling the capability to distinguish if a BER anomaly is determined by narrow filtering (e.g., due to filters misalignment) or reduced amplification (e.g., due to amplifier malfunctioning, impacting the OSNR performance at intermediate span). In this context, ML provides a huge potential in extracting hidden patterns from BER data, while using other approaches would require the availability and analysis of additional parameters, e.g., OSNR and/or Q-factor. The whole analysis is performed using real BER traces obtained over an optical network testbed including a 380km optically amplified link and a commercial coherent polarization multiplexed quadrature phase shift keying (PM-QPSK) 100 Gb/s transmission system.
Nei sistemi di comunicazione basati su canali di tipo diffusivo, come nel caso di sistemi molecolari, il massimo ritmo di trasmissione raggiungibile è limitato dalla lenta natura della diffusione e dall'esistenza di una severa interferenza intersimbolica (ISI). Sistemi di comunicazione che utilizzano una schiera di antenne sia al trasmittitore che al ricevitore (MIMO) posso essere utilizzati per migliorare il ritmo di trasmissione. Tuttavia, i sistemi MIMO, oltre all'ISI, introducono ulteriore interferenza tra i vari canali spaziali (ILI). Quest'ultima può essere mitigata utilizzando tecnica di interleaving nel tempo. La conoscenza della risposta all'impulso del canale (CIR) è fondamentale per l'equalizzazione e la rivelazione nei sistemi MIMO. Questa tesi presenta una tecniche di stima della CIR per canali MIMO diffusivi (D-MIMO) basata su sequenze di training. Stimatori a massima verosimiglianza (ML) e ai minimi quadrati (LS) sono stati derivati per la stima delle CIR e le sequenza di training sono state ottimizzate per minimizzare il corrispondente limite di Cramer-Rao. Le prestazioni di stimatori delle CIR sub-ottimi sono stati confrontati con il limite di Cramer-Rao per validarne l’efficacia. In più, varie tecniche di equalizzazione e rivelazione sono state proposte per analizzare le prestazioni dei sistemi D-MIMO molecolari. I parametri che caratterizzano i sistemi D-MIMO sono stati studiati in termini di probabilità d’errore sul bit (BER) a ritmo di trasmission nel caso in cui si assume completa conoscenza del canale (CSI) al ricevitore. Infine, la comunicazione a blocchi è stata enorcare con le sequenze di training che trasmesse all’inizio di ogni blocco per la stima delle CIR. Tali CIR stimate sono poi utilizzate per equalizzare e rivelare i segnali ricevuti per la rimanente durata del blocco. L’impatto della lunghezza delle sequenze di training sulla probabilità di errore sul simbolo e sul ritmo di trasmissione è stata studiata, dimostrando che anche per sequenza di lunghezza relativamente corta, le prestazioni del sistema approcciano il caso ideale in cui il ricevitore dispone di una perfetta conoscenza delle infromazioni di canale.
Machine-learning-based soft-failure detection and identification in optical networks
SHAHKARAMI, SHAHIN
2017/2018
Abstract
Accurate monitoring of received signal quality provides a precious source of information to secure optical networks performance and to guarantee transmission quality. During network operation, several kinds of soft failures (as opposed to hard failures, where signal is totally disrupted) can affect signal quality and induce anomalies in the BER at the receiver, ultimately leading to packet losses or even service disruption. Hence, a solid mechanism for soft failure detection (i.e., recognize anomalies due to failure occurrences), localization (i.e., identify where in the network the failure occurred), and identification (i.e., understand the actual cause of the failure) is crucial, as it may be used by operators to perform traffic re-routing and rapid failure recovery. New generation coherent transponders give the opportunity to monitor several parameters associated to optical signal transmission, such as Optical Signal-to-Noise-Ratio (OSNR), Q-factor, or pre-Forward Error Connection Bit Error Rate (pre-FEC BER). The amount of generated data is enormous and requires advanced data analysis techniques to extract useful information for these large data-sets. In this context, techniques from machine learning (ML) discipline are regarded as a strong candidate to address this issue, as ML enables automatized network self-configuration and fast decision-making by effectively leveraging the plethora of data that can be retrieved via network monitors. Previous works have already leveraged ML to perform monitoring and even failure detection in optical networks. Even though some initial research works have appeared on this topic, several questions are still pending regarding, e.g., which information is more important to be used, how often this information shall be sampled/collected by the monitors, and which ML technique (among the large set of already existing and well established tools) is better suited for soft-failure detection based on BER analysis at signal receiver. To address some of these pending questions, in this thesis we provide the following contributions: 1) we define a framework for BER anomaly detection based on monitored BER data; here we employ different ML algorithms, for which we assess the trade-off between complexity and prediction accuracy; 2) we perform a sensitivity analysis of the accuracy of the various ML techniques in order to identify the right BER sampling time, i.e., how often BER values should be collected and analyzed; 3) we propose, to the best of our knowledge for the first time in literature, a ML-based methodology to discriminate among different sources of soft failure, enabling the capability to distinguish if a BER anomaly is determined by narrow filtering (e.g., due to filters misalignment) or reduced amplification (e.g., due to amplifier malfunctioning, impacting the OSNR performance at intermediate span). In this context, ML provides a huge potential in extracting hidden patterns from BER data, while using other approaches would require the availability and analysis of additional parameters, e.g., OSNR and/or Q-factor. The whole analysis is performed using real BER traces obtained over an optical network testbed including a 380km optically amplified link and a commercial coherent polarization multiplexed quadrature phase shift keying (PM-QPSK) 100 Gb/s transmission system.| File | Dimensione | Formato | |
|---|---|---|---|
|
Shahkrami(Thesis).pdf
non accessibile
Dimensione
2.87 MB
Formato
Adobe PDF
|
2.87 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/139956