Mean Opinion Score(MOS) KPI

Network Quality & Performance Division

“Monitoring the End User Quality of “Monitoring Experience using the derivative Mean Opinion Score (MOS) KPI”

Paper Authors: Konstantinos Vlahodimitropoulos PhD, Evangelos Katsaros

Presentation: Filippos Kyriazidis

Effect of transmission performance on Multimedia Quality of Service

17 19 June 2008 Prague Czech Republic


Objectives: 

Indicate the importance of MOS to the end user perceived speech quality.



Introduce the MOS derivative Key Performance Indicator (KPI) which comes from Drive Tests Data and Network Statistics.



Implement a mathematical formalism in order systemize the MOS calculation procedure.



Compare different speech codecs implementations in BSC areas using MOS distribution patterns.




Contents: • MOS concept & dependencies. • Average MOS calculation in Cell & BSC level combining drive tests data & OSS measurements. • Cell MOS distribution patterns for different speech codecs. EFR vs AMR codec comparison. • Effect of MS technology on different speech codecs share. • Evaluating speech codec effectiveness using quality and speech codec distribution patterns Effect of transmission performance on Multimedia Quality of Service



Mean Opinion Score (MOS) concept description: • A common benchmark used to determine sound quality. • In voice telephony, represents a measure of human speech quality, at the destination end of the circuit. • It is an important factor in determining the QoS. MOS Measurement: 1. A pool of listeners rate a series of audio files using a five-grade scale ranging from 1 to 5. 2. The avera e or Mean O inion Score MOS for each audio file is calculated. Conditions: • 16 or more listeners (adequate sample). • Quiet environment.

To

reduce variability, tests commonly include reference files that have "industry accepted" MOS scores.




MOS dependencies in a mobile network:

Speech

Radio

Codec

Full Rate (FR) Half Rate (HR) Enhanced Full Rate (EFR) Adaptive Multi Rate (AMR)

Interface Conditions

Signal Quality i.e.: BER, FER

MOS classification according to ASCOM scale 0

1.5 Bad

2.1

Poor

2.7 Fair


3.5 Good

5 Excellent



• MOS concept & dependencies.

• Average MOS calculation in Cell & BSC level combining drive tests data & OSS measurements

• Cell MOS distribution patterns for different speech c EFR vs AMR codec comparison. • Effect of MS technology on different speech codec


• Basic principle of extracting a MOS derivative KPI: Perform Drive Tests

Query OSS

Process results & create MOS Quality & Speech Codec Matrix

Export Speech Codec usage Matrix Export Quality class statistics Matrix

Data Merge

Perform Matrix Operations to calculate Cell/BSC MOS value Effect of transmission performance on Multimedia Quality of Service



Drive Tests Reference Table : Quality degradation direction (8 quality classes) Speech Codec / DL Quality clas AMR FR MOS AMR HR MOS EFR MOS FR MOS HR MOS

0

1

2

3

4

5

6

7

3,71 3,36 3,76 3,19 2,92

3,56 3,24 3,68 3,14 2,83

3,31 3,04 3,43 3,03 2,68

3,05 2,88 3,08 2,77 2,59

2,72 2,71 2,72 2,58 2,43

2,49 2,47 2,34 2,22 2,13

2,12 2,11 2,01 1,82 1,66

1,70 1,40 1,44 1,45 1,41

Speech Codec degradation direction (5 speech codecs)

OSS Derived Tables Quality Class Share of BSC Speech Share of BSC Table Speech Codec Share AMR FR AMR HR EFR FR

HR

33.95% 13.96% 39.46% 0.39% 12.23%


DL Quality QUAL_0 QUAL_1 QUAL_2 QUAL_3 QUAL_4 QUAL_5 QUAL_6 QUAL_7

Samples Share

80.17% 3.73% 3.68% 3.67% 3.22% 2.72% 1.89% 0.91%



BSC MOS calculation example: Spech Codec / DL Quality class AMR FR MOS AMR HR MOS EFR MOS FR MOS HR MOS

0

1

2

3

4

5

6

7

3.71 3.36 3.76 3.19 2.92

3.56 3.24 3.68 3.14 2.83

3.31 3.04 3.43 3.03 2.68

3.05 2.88 3.08 2.77 2.59

2.72 2.71 2.72 2.58 2.43

2.49 2.47 2.34 2.22 2.13

2.12 2.11 2.01 1.82 1.66

1.70 1.40 1.44 1.45 1.41

X

Dimension: 5X8

DL Quality QUAL_0 QUAL_1 QUAL_2 QUAL_3 QUAL_4 QUAL_5 QUAL_6 QUAL_7

Samples Share

80.17% 3.73% 3.68% 3.67% 3.22% 2.72% 1.89% 0.91%

=

Speech Codec Weighted MOS Value AMR FR MOS 3.55 3.24 AMR HR MOS EFR MOS 3.59 FR MOS 3.08 HR MOS 2.82

Dimension: 5X1

Dimension: 8X1

BSC MOS =

Speech Codec Weighted MOS Value 3.55 AMR FR MOS 3.24 AMR HR MOS EFR MOS 3.59 FR MOS 3.08 2.82 HR MOS

X

Speech Codec Share AMR FR AMR HR EFR FR

HR

33.95% 13.96% 39.46% 0.39% 12.23%

= 3.43

Dimension: 1X5

Dimension: 5X1




Generalized calculation procedure: The (m x n) (m codecs, n quality classes) matrix is multiplied with the (n x1) quality classes matrix as in (1):  Q0  ... ... M 0 n    Q1 ... ... ...     ×  ...  ... ... ...    ... ... ... M mn  Qm 

 M  ...   ...

00

 M m 0

 SC   SC    =  ...    0

1

(1)

...

 SC  m

The product of (1), (m x 1) is further multiplied with the (1 x m) speech codecs matrix to obtain a single average MOS value, as in (2):  SC  ...   ...   ...  SC

0

m

    × [C   

0

... ... ... C m ] = Average _ MOS


(2)



Key issues: • • • •

Process can be easily implemented in a performance monitoring system. Can be applied in Cell, BSC or any other network structure. Cells with Poor or Bad MOS can be isolated and be further nves ga e . A single average MOS value of a cell group (i.e. BSC area) may not be essential. The comparison of different network areas provides QoS indicative results.




• MOS concept & dependencies. • Average MOS calculation in Cell & BSC level combining drive tests data & OSS measurements. • Cell MOS distribution patterns for different speech codecs. EFR vs AMR codec comparison. • Effect of MS technology on different speech codecs share. • Evaluating speech codec effectiveness using quality and speech codec distribution patterns Effect of transmission performance on Multimedia Quality of Service



MOS distribution: •Derives from cell level MOS calculation of a group of cells •Gives a brief description of the end user perceived speech quality in a specific network area.

70 0

1.5 Bad

Poor

2.1

2.7 Fair

3.5 Good

5 Excellent

60

s50 l l e c f 40 o r e b30 m u N 20 10 0

5 8 6 6 7 4 8 2 9 8 0 6 1 4 2 2 3 3 8 4 6 5 4 6 2 7 8 6 4 2 , 2 5 , 2 , 2 , 2 , 2 , 2 9 , 3 , 3 , 3 , 3 , 3 , 3 , 3 7 , 3 8 , 3 9 , 3 , 3 , 3 , MOS classification




MOS improvement pattern: •The distance between distributions shows the MOS improvement between two different states of a group of cells. 70

FR/HR codec 60

AMR codec (together with EFR & FR/HR)

s l l e c f 40 o r e b30 m u N

0

1.5 Bad

2.1

Poor

2.7 Fair

3.5 Good

5 Excellent

20

10

0

5 8 6 6 7 4 8 2 9 8 0 6 1 4 2 2 , 3 3 8 4 6 5 4 6 2 7 8 6 4 2 , 2 5 , 2 , 2 , 2 , 2 , 2 9 , 3 , 3 , 3 , 3 3 , 3 , 3 7 , 3 8 , 3 9 , 3 , 3 , 3 ,

> 4

MOS classification




EFR versus AMR: •EFR achieves highest but also lowest MOS values in comparison to AMR. •EFR distribution has a larger standard deviation than AMR. •AMR is subsequently, a preferable speech codec. 160

EFR codec (together with FR/HR) 140

AMR codec (together with EFR/FR/HR)

120

s l l e 100 c f o r 80 e b m u 60 N

. Bad

Poor

.

.

. Fair

Good

Excellent

40 20 0

5 8 6 6 7 4 8 2 9 8 0 6 1 4 2 2 3 3 8 4 6 5 4 6 2 7 8 8 6 9 4 , 7 2 , 2 5 , 2 , 2 , 2 , 2 , 2 9 , 3 , 3 , 3 , 3 , 3 , 3 , 3 , 3 , 3 3 , 3 , 3 ,

> 4

MOS classification




• MOS concept & dependencies. • Average MOS calculation in Cell & BSC level combining drive tests data & OSS measurements. • EFR vs AMR codec comparison. • Effect of MS technology on different speech codecs share. • Evaluating speech codec effectiveness using quality and speech codec distribution patterns Effect of transmission performance on Multimedia Quality of Service



The role of Mobile Station technology 

The following speech codecs share matrix is involved in all MOS calculations:

SC = [C 0

... .... ... C m ]

According to the previous example: AMRFR

AMRHR

SC = [33.95% 13.96%

EFR

39.46%

FR

HR

0.39% 12.23%]



The existence of old-technology mobiles in a network area, which do not support ‘modern’ speech codecs, will affect codecs’ shares against EFR & AMR percentages. This will degrade average MOS of BTSs.



Promotion of latest technology MSs to existing subscribers could be a key issue of marketing strategy in order to optimize speech quality.




• MOS concept & dependencies. • Average MOS calculation in Cell & BSC level combining drive tests data & OSS measurements. • EFR vs AMR codec comparison. • Effect of MS technology on different speech codecs share. • Evaluating speech codec effectiveness using quality and speech codec distribution patterns




Quality & Speech codec distribution patterns:  An

equivalent way of deciding speech codecs’ effectiveness is to calculate the percentage of samples in each MOS class, i.e. create a graph with the following information: 45%

40%

35%

30%

e r a h 25% s s e l p m20% a S 15%

10%

5%

0% Excellent

Good

Fair

Poor

Bad

To achieve this aim drive tests data should be post-processed differently with regard to the original data. Effect of transmission performance on Multimedia Quality of Service



Reference Table Quality & Speech codec distribution patterns: Post-Processed Drive Tests Data SC MOS Class AMR FR E AMR FR G AMR FR F AMR FR P AMR FR B AMR HR E AMR HR G AMR HR F AMR HR P AMR HR B EFR E EFR G EFR F EFR P EFR B FR E FR G FR F FR P FR B HR E HR G HR F HR P HR B

0 87.79% 7.33% 4.31% 0.57% 0.00% 38.71% 53.12% 7.10% 1.08% 0.00% 91.48% 7.05% 1.25% 0.00% 0.23% 0.00% 96.14% 3.34% 0.39% 0.13% 0.16% 92.54% 5.40% 1.90% 0.00%

1 75.78% 13.66% 7.45% 3.11% 0.00% 20.81% 68.79% 7.72% 2.68% 0.00% 87.06% 9.90% 2.08% 0.80% 0.16% 0.21% 92.31% 6.44% 1.04% 0.00% 0.42% 81.89% 13.63% 3.95% 0.10%

2 55.38% 22.58% 16.94% 4.30% 0.81% 15.63% 58.75% 18.13% 6.88% 0.63% 68.26% 19.90% 8.82% 2.77% 0.25% 0.00% 84.43% 12.11% 3.11% 0.35% 1.31% 60.93% 26.24% 11.08% 0.44%


3 35.41% 31.91% 21.40% 9.73% 1.56% 10.59% 54.12% 23.53% 11.76% 0.00% 41.13% 30.65% 21.77% 5.24% 1.21% 0.00% 63.30% 25.00% 9.57% 2.13% 1.72% 49.68% 32.47% 14.41% 1.72%

4 15.51% 33.16% 32.09% 17.11% 2.14% 12.50% 35.00% 35.00% 12.50% 5.00% 15.69% 39.87% 29.41% 7.84% 7.19% 0.67% 43.33% 38.00% 16.67% 1.33% 1.32% 32.34% 39.27% 24.09% 2.97%

5 7.77% 30.10% 33.01% 24.27% 4.85% 10.81% 18.92% 29.73% 37.84% 2.70% 6.15% 18.46% 43.08% 23.85% 8.46% 0.00% 18.92% 36.94% 34.23% 9.91% 0.00% 9.83% 43.35% 37.57% 9.25%

6 0.00% 19.28% 30.12% 31.33% 19.28% 0.00% 12.50% 31.25% 43.75% 12.50% 0.00% 16.67% 21.43% 38.10% 23.81% 0.00% 4.00% 24.00% 45.33% 26.67% 0.00% 1.82% 13.64% 44.55% 40.00%

7 0.00% 0.00% 22.86% 34.29% 42.86% 0.00% 0.00% 11.11% 11.11% 77.78% 0.00% 0.00% 15.00% 15.00% 70.00% 0.00% 0.00% 10.00% 25.00% 65.00% 0.00% 0.00% 8.51% 19.15% 72.34%



Calculation example in a BSC area with AMR /EFR/FR/HR enabled: 

Calculation Procedure:  Drive Tests data are weighted according to quality class share, that is the following table: _ 83.58%



_ 3.12%

_ 2.97%

_ 2.93%

_ 2.52%

_ 2.23%

_ 1.71%

_ 0.94%

The resulting table is further transformed according to speech encoding share: AMRFR AMRHR EFR FR HR


37.43% 8.73% 46.12% 0.44% 7.28%



Calculation example in a BSC area with AMR /EFR/FR/HR enabled: 

After all transformations (multiplications) the Drive tests table becomes: SC MOS Class AMR FR E AMR FR G AMR FR F AMR FR P AMR FR B AMR HR AMR HR AMR HR AMR HR EFR EFR EFR EFR EFR FR FR FR FR FR HR HR HR HR HR

G F P B E G F P B E G F P B E G F P B

0 27.213% 2.292% 1. 348% 0. 180% 0. 000% . 3.876% 0. 518% 0.078% 0.000% 35.261% 2. 716% 0. 482% 0. 000% 0. 088% 0.000% 0.354% 0.012% 0.001% 0.000% 0.010% 5.634% 0.329% 0.116% 0.000%

1 0.897% 0.159% 0.087% 0.036% 0.000% . 0.187% 0.021% 0.007% 0.000% 1.251% 0.142% 0.030% 0.011% 0.002% 0.000% 0.013% 0.001% 0.000% 0.000% 0.001% 0.186% 0.031% 0.009% 0.000%

2 0.663% 0.251% 0.188% 0.048% 0.009% . 0.152% 0.047% 0.018% 0.002% 0.936% 0.273% 0.121% 0.038% 0.003% 0.000% 0.011% 0.002% 0.000% 0.000% 0.003% 0.132% 0.057% 0.024% 0.001%

3 0.425% 0.350% 0. 235% 0. 107% 0. 017% . 0.138% 0. 060% 0.030% 0.000% 0. 555% 0. 414% 0. 294% 0. 071% 0. 016% 0.000% 0.008% 0.003% 0.001% 0.000% 0.004% 0.106% 0.069% 0.031% 0.004%


4 0.160% 0.312% 0. 302% 0. 161% 0. 020% . 0.077% 0. 077% 0.027% 0.011% 0. 182% 0. 463% 0. 342% 0. 091% 0. 083% 0.000% 0.005% 0.004% 0.002% 0.000% 0.002% 0.059% 0.072% 0.044% 0.005%

5 0.068% 0.251% 0.275% 0.202% 0.040% . 0.037% 0.058% 0.074% 0.005% 0.063% 0.190% 0.443% 0.245% 0.087% 0.000% 0.002% 0.004% 0.003% 0.001% 0.000% 0.016% 0.070% 0.061% 0.015%

6 0.000% 0.123% 0.193% 0.201% 0.123% . 0.019% 0.047% 0.065% 0.019% 0.000% 0.132% 0.169% 0.301% 0.188% 0.000% 0.000% 0.002% 0.003% 0.002% 0.000% 0.002% 0.017% 0.056% 0.050%

7 0.000% 0.000% 0.081% 0.121% 0.151% . 0.000% 0.009% 0.009% 0.064% 0.000% 0.000% 0.065% 0.065% 0.304% 0.000% 0.000% 0.000% 0.001% 0.003% 0.000% 0.000% 0.006% 0.013% 0.050%

Total 29.427% 3.740% 2.710% 1.056% 0.361% . 4.487% 0.837% 0.309% 0.101% 38. 248% 4.329% 1.945% 0.822% 0.772% 0.000% 0.392% 0.028% 0.013% 0.007% 0.020% 6.136% 0.651% 0.354% 0.125%



Calculation example in a BSC area with AMR /EFR/FR/HR enabled: 

A grouping by the MOS class yields the MOS class distribution of samples within the BSC area: MOS Class Samples Share Excellent 70.69% Good 19.08% Fair 6.17% Poor 2.55% .

80%

Samples Share 70%

60%

e g a t n50% e c r e p40% s e l p30% m a S 20%

10%

0% Excellent

Good

Fair

Poor

Bad

MOS Class




Comparative evaluation of three different network states: MOS Class Excellent Good Fair Poor Bad

ΑΜR/EFR/FR/HR

70.69% 19.08% 6.17% 2.55% 1.37%

EFR/FR/HR 56.98% 33.65% 5.32% 2.47% 1.58%

FR/HR 0.11% 87.36% 7.23% 3.61% 1.59%

100%

ΑΜR/EFR/FR/HR

EFR/FR/HR

FR/HR

90% 80%

e 70% g a t n 60% e c r e p 50% s e l p 40% m a S 30% 20% 10% 0% Excellent

Good

Fair

Poor

Bad

MOS Class






Mean Opinion Score(MOS) KPI

Recommend Documents