• No results found

on QoE and inter-parameter relationship

2. We were able to identify parameter quantization and parameter buffer size to be the most influential parameters of the content and CPP domains respectively. Whereas parameter packet loss, parameters packet reorder and parameter jitter were found to be equally important within the network domain.

3. We were able to identify the effectiveness of the Taguchi DoE approach in con- ducting our subjective experiments. Such utilization of Taguchi DoE significantly reduced the time and cost of experimentation without compromising experiment’s evaluation capability.

4. We were able to utilize and (independent) HoQ method, in parallel to the Taguchi DoE method, for identifying order of effect of parameters and validating our Taguchi method findings.

5. We developed a model based on 5 parameters for predicting end to end video QoE, which has been demonstrated, gives good and accurate indications of this quality across the three key domains over which services are transmitted to users in the video IPTV environment.

9.4

Future Work

The thesis discussed the need for developing model based on parameters from the three domains of content, network and CPP. We were able to include 11 control parameters and 3 noise factors in 3 experiments. We identified the most influential parameters and then identified the order of effect of these most influential parameters on end to end MOS of video QoE.

We were able to identify the following weaknesses in our research 1. Basic selection of parameters were on the existing literature

2. Customer premises equipment experiments were restricted due to limitation of hardware combinations available for experimentation

3. Human psychology depends on multiple parameters and those were not studied in this work

4. Wireless communication is not an integral part of home network and we assumed all communication to be over wired medium

One of the identified weakness of this research was that instead of an exhaustive parameter research we made an effort to reduce the initial set of parameters to a

manageable number. For this reason we included the parameters which were repeatedly reported in the literature. A further study is required to investigate an even larger set of parameters. Especially for the CPP domain there is a need to investigate a larger set of parameters. This require development of special equipment which could be configured in combinations of the selected parameters. A special test bed needs to be developed to help enable the experiment execution.

Human psychology play an important role in evaluation of services being offered to them. For this study we reduced the parameters to the engineering domain and only discussed few psychological effects. A further study is required to incorporate important aspect of culture and impact of socio-economic condition. For this research we tried to develop interest, of volunteers/viewers, in the experimentation process. For this we used movie clips, which helped us in increasing the viewer’s interest in the experiment. These clips were of 10 second length, which were very short. Further study is required to investigate the effects of using a longer video clip. In addition to the video length we also want to investigate the effect of varied genre of movies. Use of full length movie and option of selecting this full length movie from a pool of selected movies will simulate the actual scenario of an IPTV video on demand service deployment.

For the CPP domain we assumed the communication to be over the wired medium. There is a need to quantify the effect of wireless communication for end to end QoE. In addition we also want to quantify the end to end effect of parameters on interactive communication applications. Such an evaluation will require integration of further parameters which affect quality of an interactive application.

We also utilized HoQ method of QFD and we found that QFD failed to deliver the results, for analyzing the best parameter from the group of most influential parameters. There is need of further investigation of this fact and an improvement in HoQ is required to be proposed.

The outcome of our work can be utilized by content producers, content distributors, network provider or service providers and by the companies who are selling equipment for the CPP. We proposed the configurations which should be avoided in order to provide a minimal acceptable service as well as the configuration which will ensure at least good or excellent feedback from viewers. The content producers or distributors can use this knowledge in producing content which is better suited to constraints in the network and CPP domains. The knowledge of network domain parameters will help us in avoiding scenarios where the service provider will lose customer without knowing about it. If they can avoid the pitfalls they will be able to provide an acceptable service. The service provider will also be able to predict expected QoE and can fix the issue even before they become an issue. The model can be used by network planner for identifying minimum requirement within the system for a successful system deployment

9.4. FUTURE WORK 145

Appendix A

Code Fragments and additional

tables

A.1

Code Fragments

1 2 ####R−Programme 3 ####Program f o r s i g n i f i c a n c e t e s t u s i n g p e r m u t a t i o n s a m p l i n g#### 4 y=c( 3 , 0 , 9 , 0 , 1 , 1 , 1 , 3 , 1 , 1 , 1 , 3 , 3 , 1 , 1 , 9 , 3 , 3 , 3 , 9 , 3 , 3 , 9 , + 5 9 , 9 , 3 , 3 , 3 , 3 , 3 , 3 , 0 , 3 , 3 , 3 , 0 , 3 , 3 , 9 , 0 , 1 , 3 , 1 , 0 ) 6 imp r a t i n g=c( 5 , 3 , 3 , 4 , 4 , 3 , 5 , 2 , 5 , 3 , 5 ) 7 R=11; C=4

8 hoq=m a t r i x( y ,nrow=R, n c o l=C, byrow=TRUE)

9 hoq ; imp r a t i n g 10 t e c h r a t i n g mat=hoq∗imp r a t i n g 11 i n i t i a l t e c r a t i n g=colSums ( t e c h r a t i n g mat) 12 i n i t i a l t e c r a t i n g 13 s r t=s o r t( i n i t i a l t e c r a t i n g , d e c r e a s i n g = FALSE) 14 s r t ; i n i t i a l t e c r a t i n g=s r t 15 d=NA 16 f o r( i i n 1 :C) 17 { 18 d [ i ]= i n i t i a l t e c r a t i n g [ i ] 19 } 20 r e p l i c =5000

21 permute=m a t r i x(NA,nrow=r e p l i c ,n c o l=C)

22 d i f f=m a t r i x(NA,nrow=r e p l i c ,n c o l=C) 23 f o r( i i n 1 : r e p l i c ) 24 { 25 y=sample( y , r e p l a c e=F) 26 #y=r p o i s (R∗C, 3 ) 27 hoq=m a t r i x( y ,nrow=R, n c o l=C) 28 t e c h r a t i n g mat=hoq∗imp r a t i n g 147

29 pseudo .t=colSums ( t e c h r a t i n g mat)

30 permute [ i , ] = pseudo .t

31 }

32 p v a l u e mat=m a t r i x(NA,nrow=C,n c o l=C)

33 d i f=NA; p v a l u e=NA 34 d i f f=c(r e p(NA, r e p l i c ) ) 35 f o r( j i n 1 :C) 36 { 37 f o r ( i i n 1 :C) 38 { 39 d i f [ i ]=a b s( d [ j ]−d [ i ] ) 40 d i f f=a b s( permute [ , j ]−permute [ , i ] ) 41 i f ( i==j ) 42 { 43 p v a l u e [ i ]=NA 44 } 45 e l s e 46 { 47 i f( i<j ) 48 { 49 p v a l u e [ i ]=NA 50 } 51 e l s e 52 { 53 p v a l u e [ i ]=l e n g t h(d i f f[ (d i f f)>( d i f [ i ] ) ] )/r e p l i c 54 } 55 } 56 } 57 p v a l u e mat[ j , ] = p v a l u e 58 } 59 p v a l u e mat

A.2

Tables

Table A.1: Calculation for Network Domain Fleiss´ Kappa

1 2 3 4 5 Pi

0 1 12 3 0 0.575

0 11 5 0 0 0.542

0 0 0 6 10 0.5

0 0 1 10 5 0.458