Dimensioning Packet Based Fronthaul for Radio Base Stations: Network Requirements and Traffic Scenarios
Typ
Examensarbete för masterexamen
Program
Engineering mathematics and computational science (MPENM), MSc
Publicerad
2022
Författare
Abraham, Jimmy
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
A radio base station is a wireless communication station which is used for wireless
telephone systems such as GSM (2G), WCDMA (3G), LTE (4G) and NR (5G).
Today, radio base stations use circuit switched fronthaul and will continue to do so
for a foreseeable future, however packet based fronthaul will grow as 5G development
does as well. A radio base station for packet based communication consists of three
main units: radio, baseband and switch. The aim of the thesis was to implement a
simulator for packet based fronthaul and determine optimal network configurations
for a given traffic load through statistical methods.
First the radio base stations were defined mathematically as a small network through
graph theory. Then traffic generation was characterized as a two state Markov chain.
Algorithms were designed for both network and traffic generation. Packets inside
ethernet networks follow protocols which were simplified and adapted for radio base
stations. An algorithm was designed for simulating the packets movement in the
network. The simulator was implemented in Python and was evaluated through
statistical methods to determine the amount of time steps and iterations that is
required to have consistent simulation results. Two-sided one sample t-test showed
that five iterations for each simulation was enough to generate samples close enough
to the expected mean and 10000 time steps was shown to be enough for a low
standard deviation.
Finally, data was generated with the simulator for selected combinations of network
properties. Selected properties were number of radio units, buffer size, number
of switches, and traffic load. A Poisson regression model was used to correlate
properties of a given network with the amount of packet loss as response variable.
Results show that the D2 score is about 0.8 and prove that the model works with
a reasonably good score. This also shows that with more time and development a
more advanced analysis can be made, and better machine learning models can be
made with the help of the Ethernet Simulator.
Beskrivning
Ämne/nyckelord
Packets, Ethernet Frames, eCPRI, Telecommunication, Radio Base Station, RAN, 5G, Machine learning, Poisson Regression GLM, Fronthaul