Post on 10-Aug-2020
Fig. 1 Optimization strategy
Data-driven network management for 5G systems based on QoE criteriaCarolina Gijón-Martín1 (supervised by Matías Toril2 and Salvador Luna-Ramírez3)
Communications Engineering Department, University of Málaga (Spain)
e-mails: {cgm@ic.uma.es 1, mtoril@ic.uma.es2, sluna@ic.uma.es3}
WORK LINE 1: data-driven traffic steering for optimizing QoE in multi-tier networks [1]
Goal: optimize traffic sharing among carriers.
1) Activation of RSRQ-based inter-frequency handovers.
2) Self-tuning algorithm to adjust handover margins with
QoE criteria.
Method: indicator (∆𝑄𝑜𝐸𝑇(𝑖, 𝑗) ) derived from connection
traces as a driver.
Performance assessment: dynamic system-level simulator
emulating a live 2-tier LTE network with video, FTP, web
browsing and VoLTE services.
Results: proposed algorithm (QBHO+OE) outperforms
classical traffic steering algorithms in terms of QoE.
WORK LINE 2 (ongoing): dimensioning cell capacity in 5G systems with network slicing based on QoE criteria
Basic cell performance models in the absence of QoE measurements.
Models trained with few labeled cases of congestion using month-BH stats.
Prediction of cell performance independent of other cells.
Need for short-term prediction in systems with network slicing.
MOTIVATION & GOALS
1. LEGACY APPROACH
Fig. 2 Performance comparisonFig. 1 Optimization strategy
a) Proposed inter-frequency handover scheme b) Self-tuning algorithm
2. USE CASE
3. PROPOSED APPROACH
WHAT DOES 5G IMPLY?
New services and use cases
New functions and features (e.g., network slicing, network
virtualization, multi-connectivity)
High user expectations
Complex management Self-Organizing Networks (SON)
CURRENT SITUATION
• Need to adapt SON techniques to 5G features
• Need to change network management approach to a
user centric approach (Quality of Experience, QoE)
• Massive data (connection traces, PM, CM) stored but
not exploited
PHD GOAL
Develop self-planning and self-
optimization techniques for cellular
networks based on QoE criteria by
using big data analytics over traces
OTT service provider 1
-Time
- Services
- QoE requirements
OTT service provider 2
-Time
- Services
- QoE requirements
Virtual Network Operator
- Time
- Services
- QoE requirements
Traffic classification
(clustering)
Traffic forecasting
(time series analysis,
supervised learning)
QoE modeling
(MLR, supervised
learning)
Bottleneck detection
(with QoE constraints)
…
PMs
CMs
Traces
Purpose: classify encrypted connections in the radio access network per application
type (e.g., immersive video, IoT, app download…).
Methodology:
1) Feature selection/extraction of relevant traffic descriptors from connection traces.
2) Unsupervised learning (k-means, k-medoids, DBSCAN) over selected descriptors.
3) Check results against typical traffic mix reported by vendors/operators.
More slices
Mobile Network operator (MNO)[2]
Dynamic resource
allocation per slice
Classical dimensioning tool
Dynamic resource
reservation
Cell/slice classification
(clustering)
Proactive QoE-based dimensioning tool
REFERENCES[1] C. Gijón, et al. A data-driven traffic steering algorithm for optimizing user experience in multi-tier LTE networks, IEEE Transactions on Vehicular Technology, vol.68, nº10, pp.9414-9424, 2019.
[2] Da Silva, Icaro, et al. Impact of network slicing on 5G Radio Access Networks. En 2016 European conference on networks and communications (EuCNC), pp. 153-157, 2016.
Work funded by the Spanish Ministry of Science, Innovation and
Universities (RTI2018-099148-B-I00) and the Spanish Ministry of
Education, Culture and Sports (FPU grant FPU17/04286).
Purpose: forecast traffic a per-slice basis in the long-/short-term to avoid
future capacity problems.
Methodology: supervised learning (e.g., SVR, ANN, RF) over historical
performance data collected per slice.
QoE prediction
(time series analysis, supervised learning)
Fig. 3 Long-term cell traffic forecasting exampleFeature
selection/
extraction
(PCA, wrapper,
filtering)
1) Will the network experience capacity problems in the future with existing tenants?
2) Can the network accept new tenants?