Transcript Master`s Thesis Presentation

Frame Header Based Speech Quality
Analysis Method
in a Circuit-Switched Media Gateway
Master’s Thesis Presentation 18.10.2005
Author:
Supervisor:
Instructor:
Mika Väisänen
Prof. Raimo Kantola
Ph.D. Peter Jungner
Contents








Introduction
Circuit-Switched Media Gateway
Speech Coding
Iu and Nb User Plane Protocols
Speech Quality Measurement
Estimation Method development
Analysis of the Method
Conclusions
© Ericsson AB 2005
Mika Väisänen
2
Master's Thesis Presentation
2005-10-18
Introduction

Background
–
–
–

Problem
–
–
–

On UMTS networks coded speech is transported in frames
On ideal situation only the used speech coding method degrades
the speech quality of a call
In practise, frames are damaged on air-interface and lost on core
network congestion
Operator may not know, how customers are perceiving the quality
of the network
Operator will lose customers, if speech quality in the network
drops
Operator must be able to monitor the speech quality in the
network in real time
Objectives
–
© Ericsson AB 2005
To develop a method that can estimate speech quality of calls in
UMTS Core Network by analysing only the speech frame headers
Mika Väisänen
3
Master's Thesis Presentation
2005-10-18
Circuit-Switched Media Gateway
(CS-MGW)
 Adapts different Access Networks to the Core Network
 Main functions:
– Media conversion (ATM, IP, TDM)
– Bearer control (Resource reservation)
– Payload processing (Transcoding, echo cancelling, …)
© Ericsson AB 2005
Mika Väisänen
4
Master's Thesis Presentation
2005-10-18
Speech Coding
 Adaptive Multi-Rate (AMR) coding used in UTRAN
– Variable bit-rate modes from 4.75 to 12.2 kbps
– Source Controlled Rate of operation
 During silence only Silence Descriptor (SID) frames
are sent with low bit-rate
– Uses efficient error concealment
 Lost or damaged frames are “faded away”
 Frame substitution and muting
– AMR end-to-end = Transcoder Free Operation (TrFO)
 Pulse Code Modulation (PCM) possibly used in CN
– Compressed, 64 kbps
– No error concealment
– AMR-PCM-AMR = Coder tandeming, transcoding
© Ericsson AB 2005
Mika Väisänen
5
Master's Thesis Presentation
2005-10-18
Iu and Nb User Plane Protocols

Speech is carried in User Plane frames
–
–

Besides speech the Iu/Nb frames contain information
–
–
–

1 AMR frame in each Iu/Nb frame
40 PCM samples in each Nb frame
Frame numbering to detect lost frames
Frame Quality Classification (FQC)
Information of the frame type (AMR bit-rate, SPEECH/SID)
Transcoding in Tandem call cases re-creates the frame stream
–
© Ericsson AB 2005
All information regarding quality in the frame headers is lost
Mika Väisänen
6
Master's Thesis Presentation
2005-10-18
Speech Quality Measurement

Listening tests
–
–

Absolute Category Rating (ACR), scale 1-5
Mean Opinion Score (MOS)
Objective methods
–
–
–
© Ericsson AB 2005
Emulate listening tests
Speech signal based
 Resource consuming
 Perceptual Evaluation of Speech Quality (PESQ)
- PESQ score, ranging from -0.5 to 4.5.
- Correlation against listening tests 0.935.
Parameter based
 Light, but not as accurate
 ITU E-Model
 PsyVoIP, VQMon
Mika Väisänen
7
Master's Thesis Presentation
2005-10-18
Estimation Method Development

Establish a model between frame loss/damage and
speech quality
– Frame losses and damages in simulated environment
– Lost SID frames ignored, because they are 100 times less
important than speech frames
– Speech quality analysis with PESQ

Find out a way to determine types of lost frames
– In PCM case simple, as all frames can be considered
equal.
– In AMR case SID frames complicate the determination

Create a method implementation to be run in CS-MGW
© Ericsson AB 2005
Mika Väisänen
8
Master's Thesis Presentation
2005-10-18
Analysis of the Method
 AMR TrFO case (AMR 12.2 kbps all the way)
– Correlation of 0.90 was established between the method
and real PESQ scores
 Mean estimation error 0.14 PESQ-MOS units
© Ericsson AB 2005
Mika Väisänen
9
Master's Thesis Presentation
2005-10-18
Analysis of the Method
 Tandem case (AMR 12.2 - PCM – AMR 12.2)
– Correlation of 0.83 was established between the method
and real PESQ scores
 Mean estimation error 0.19 PESQ-MOS units
© Ericsson AB 2005
Mika Väisänen
10
Master's Thesis Presentation
2005-10-18
Conclusions
 The method proven to be surprisingly accurate, despite
its simple implementation
– PESQ-MOS differences < 0.5 are barely audible
 Being able to determine the frame content
(silence/speech) helps to improve the estimation
 Ideal solution for operators using a leased RAN
– In addition to price, also speech quality can be used to
compare alternative networks
© Ericsson AB 2005
Mika Väisänen
11
Master's Thesis Presentation
2005-10-18
© Ericsson AB 2005
Mika Väisänen
12
Master's Thesis Presentation
2005-10-18