The forthcoming advent of 5G networks is not only a technological revolution, it will also encompass new business models, whom service providers and network operators should be ready to adapt to. Differently from 4G, which enhanced the performance of the existing 3G systems, 5G will enable new communication paradigms such as device-to-device, inter- and intra- vehicular communications, as well as new services like remote healthcare and smart homes, just to name a few. 5G will dedicate different portions of the network to ad-hoc services with specific speed/latency requirements, taking advantage of the so-called ‘network slicing’ approach that allows to deploy several multi-service logically distinct networks over the same physical infrastructure. End-users and their expectations will be the focus of any provider. As such, the strategical key asset of 5G will no longer be Quality of Service (QoS), rather, Quality of Experience (QoE). To anticipate market needs, the definition of a trustable monitoring approach to QoE is a strategical action for Empirix, the corporation where I worked over my PhD years, as this company offers troubleshooting and diagnostics solutions to service providers and enterprises. This dissertation focuses on the exploration of 5G and QoE topics and in its final part, it proposes a novel technique to assess the quality that end-users will experience. Firstly, I followed 5G standardization with a special emphasis on investigating aspects related to Empirix core business. I transferred my knowledge to the whole company through trainings and lessons. After this phase of study, Empirix and I recognized in the delivery of voice one of the most promising 5G applications, worthy of an in-depth study. Recently, Forbes stated: “with 5G every object could soon have a voice”, and, according to Ericsson forecasting: “voice is the king of communication and in a 5G world it will be more important than ever […]. The network infrastructure used for Voice over LTE [VoLTE] today will also be used to enable 5G voice calls”. Thus, by exploiting passive network measurements, I performed a comparative study of the end-to-end quality degradation that several millions of real VoLTE calls underwent, when two popular codecs were employed, namely, Adaptive MultiRate (AMR) and Adaptive MultiRate WideBand (AMR-WB). This study revealed to what extent AMR-WB based calls are more robust against network impairments of voice services (e.g., jitter and packet loss rate) than their narrowband counterpart. In parallel, I also came to the conclusion that relying on empirical model to monitor QoE leads to several limitations, most notably, the lack of actual feedback from end-users. I addressed this shortcoming adopting a customer-driven approach, although in a much more confined environment. To this regards, I leveraged a pool of research participants that listened and scored AMR-WB calls generated by Hammer, an Empirix platform that emulates Voice over IP communications, therefore providing truly subjective evaluation scores. By comparing different state-of-the-art algorithms my goal was to exploit statistical approaches based on supervised Machine Learning to understand how QoE was related to network metrics and human-based features like age, gender and type of headset. The major findings were: i) Ordinal Logit Regression is the algorithm that best resembles the aforementioned relation. ii) Users usually agree to rate call quality as either ‘excellent’ or ‘bad’. iii) Conversely, when they are asked to rate a call whose quality can be considered between two opposites, the statistical prediction of QoE becomes a difficult task: the perception of 'intermediate' quality is deeply related to subjective personal traits and it may significantly vary among persons.

L’imminente avvento del 5G non rappresenta solo una rivoluzione tecnologica, ma definisce anche nuovi modelli di business, a cui i service providers e gli operatori di rete dovranno adattarsi. A differenza del 4G, che ha migliorato in termini di performance il preesistente 3G, il 5G promette di abilitare nuovi paradigmi di comunicazione come le comunicazioni device-to-device, intra e inter-veicolari, cosí come di diffondere nuovi servizi, la tele-medicina e le case intelligenti, per citarne alcuni. Coadiuvata dal cosiddetto network slicing, che consentirà lo sviluppo di multiple reti logiche sulla stessa infrastruttura fisica, il 5G allocherà porzioni di rete a servizi ad-hoc con specifici requisiti di latenza e velocità. L’utente focalizzerà l'attenzione di diversi providers, e con esso ogni sua aspettativa. Per tale motivo, il key asset del 5G non è più la Quality of Service (QoS) ma la Quality of Experience (QoE). Per anticipare e rimanere competitivi nel mercato, la definizione di un sistema affidabile di monitoraggio della QoE è cruciale per l’azienda in cui ho svolto il mio PhD, Empirix, la quale offre servizi di diagnostica e troubleshooting ad operatori di rete ed imprese. Questa tesi esplora il tema del 5G e della QoE, e nella parte finale propone un approccio innovativo volto a determinare la qualità percepita dall’utente finale. In una prima fase, ho seguito la standardizzazione del 5G, con particolare attenzione agli aspetti legati al core business dell’azienda. Ho svolto seminari e lezioni sulle principali caratteristiche del 5G. In seguito, assieme ad Empirix, ho individuato nei servizi voce una delle applicazioni più promettenti dell’ecosistema 5G, e meritevole di approfondimento. Recentemente, Forbes ha dichiarato che “with 5G every object could soon have a voice”, e secondo le previsioni di Ericsson “voice is the king of communication and in a 5G world it will be more important than ever […]. The network infrastructure used for Voice over LTE (VoLTE) today will also be used to enable 5G voice calls”. Ho dapprima sfruttato misure passive di rete per analizzare e confrontare la qualità di diversi milioni di chiamate VoLTE di una rete reale, codificate dagli odierni Adaptive Multi-Rate (AMR) e Adaptive Multi-rate Wide-Band (AMR-WB) codecs. Questa analisi ha permesso di verificare la maggiore robustezza delle chiamate AMR-WB rispetto agl’impariments dei servizi voce, quali jitter e packet loss rate. In parallelo, sono giunta alla conclusione che fare affidamento ad un modello empirico per monitorare la QoE porti a diverse limitazioni, prima fra tutte, la mancanza feedback dell’utente finale. Per ovviare a questa lacuna, ho adottato un approccio customer-driven, seppure in un contesto molto piú confinato del precedente. A tale proposito, ho sottoposto ad un gruppo di volontari un esperimento di valutazione di qualità chiamate AMR-WB, generate in ambiente virtuale da Hammer, una piattaforma proprietaria di Empirix volta ad emulare comunicazioni Voice over IP. Confrontando le caratteristiche di diversi algoritmi dello stato dell’arte, ho allenato un modello di Machine Learning supervisionato per comprendere come la QoE sia in relazione con le metriche di rete e le caratteristiche dell’utente (età, genere, tipo di dispositivo impiegato). Ho constatato che: i) l’algoritmo che meglio approssima la relazione di cui sopra é l’Ordinal Logit Regression; ii) gli utenti spesso convengono nel definire una chiamata di qualità ‘eccellente’ o ‘pessima’ iii) per contro, quando valutano una chiamata di qualità definibile ‘intermedia’, entrano in campo variabili soggettive che difficilmente possono essere interpretate in termini statistici, rendendo pertanto ardua la predizione di QoE.

Il ruolo della Quality of Experience e della Voce nelle reti 5G / Elena Cipressi , 2020 Mar 09. 32. ciclo, Anno Accademico 2018/2019.

Il ruolo della Quality of Experience e della Voce nelle reti 5G

CIPRESSI, ELENA
2020

Abstract

The forthcoming advent of 5G networks is not only a technological revolution, it will also encompass new business models, whom service providers and network operators should be ready to adapt to. Differently from 4G, which enhanced the performance of the existing 3G systems, 5G will enable new communication paradigms such as device-to-device, inter- and intra- vehicular communications, as well as new services like remote healthcare and smart homes, just to name a few. 5G will dedicate different portions of the network to ad-hoc services with specific speed/latency requirements, taking advantage of the so-called ‘network slicing’ approach that allows to deploy several multi-service logically distinct networks over the same physical infrastructure. End-users and their expectations will be the focus of any provider. As such, the strategical key asset of 5G will no longer be Quality of Service (QoS), rather, Quality of Experience (QoE). To anticipate market needs, the definition of a trustable monitoring approach to QoE is a strategical action for Empirix, the corporation where I worked over my PhD years, as this company offers troubleshooting and diagnostics solutions to service providers and enterprises. This dissertation focuses on the exploration of 5G and QoE topics and in its final part, it proposes a novel technique to assess the quality that end-users will experience. Firstly, I followed 5G standardization with a special emphasis on investigating aspects related to Empirix core business. I transferred my knowledge to the whole company through trainings and lessons. After this phase of study, Empirix and I recognized in the delivery of voice one of the most promising 5G applications, worthy of an in-depth study. Recently, Forbes stated: “with 5G every object could soon have a voice”, and, according to Ericsson forecasting: “voice is the king of communication and in a 5G world it will be more important than ever […]. The network infrastructure used for Voice over LTE [VoLTE] today will also be used to enable 5G voice calls”. Thus, by exploiting passive network measurements, I performed a comparative study of the end-to-end quality degradation that several millions of real VoLTE calls underwent, when two popular codecs were employed, namely, Adaptive MultiRate (AMR) and Adaptive MultiRate WideBand (AMR-WB). This study revealed to what extent AMR-WB based calls are more robust against network impairments of voice services (e.g., jitter and packet loss rate) than their narrowband counterpart. In parallel, I also came to the conclusion that relying on empirical model to monitor QoE leads to several limitations, most notably, the lack of actual feedback from end-users. I addressed this shortcoming adopting a customer-driven approach, although in a much more confined environment. To this regards, I leveraged a pool of research participants that listened and scored AMR-WB calls generated by Hammer, an Empirix platform that emulates Voice over IP communications, therefore providing truly subjective evaluation scores. By comparing different state-of-the-art algorithms my goal was to exploit statistical approaches based on supervised Machine Learning to understand how QoE was related to network metrics and human-based features like age, gender and type of headset. The major findings were: i) Ordinal Logit Regression is the algorithm that best resembles the aforementioned relation. ii) Users usually agree to rate call quality as either ‘excellent’ or ‘bad’. iii) Conversely, when they are asked to rate a call whose quality can be considered between two opposites, the statistical prediction of QoE becomes a difficult task: the perception of 'intermediate' quality is deeply related to subjective personal traits and it may significantly vary among persons.
The role of Quality of Experience and Voice in 5G Networks
9-mar-2020
MERANI, Maria Luisa
File in questo prodotto:
File Dimensione Formato  
thesis_ECipressi.pdf

Open Access dal 09/09/2021

Descrizione: tesi di dottorato
Dimensione 3.25 MB
Formato Adobe PDF
3.25 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

Licenza Creative Commons
I metadati presenti in IRIS UNIMORE sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono rilasciati con licenza Attribuzione 4.0 Internazionale (CC BY 4.0), salvo diversa indicazione.
In caso di violazione di copyright, contattare Supporto Iris

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11380/1200608
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact