QoS Mapping of VoIP Communication using Self-Organizing Neural Network



Similar documents
Assessing the quality of VoIP transmission affected by playout buffer scheme and encoding scheme

PERFORMANCE ANALYSIS OF VOIP TRAFFIC OVER INTEGRATING WIRELESS LAN AND WAN USING DIFFERENT CODECS

Traffic Characterization and Perceptual Quality Assessment for VoIP at Pakistan Internet Exchange-PIE. M. Amir Mehmood

New Models for Perceived Voice Quality Prediction and their Applications in Playout Buffer Optimization for VoIP Networks

Performance Evaluation of AODV, OLSR Routing Protocol in VOIP Over Ad Hoc

Performance Analysis Proposal

White Paper. PESQ: An Introduction. Prepared by: Psytechnics Limited. 23 Museum Street Ipswich, Suffolk United Kingdom IP1 1HN

Tech Note. Introduction. Definition of Call Quality. Contents. Voice Quality Measurement Understanding VoIP Performance. Title Series.

Evaluating Data Networks for Voice Readiness

ANALYSIS OF LONG DISTANCE 3-WAY CONFERENCE CALLING WITH VOIP

Monitoring VoIP Call Quality Using Improved Simplified E-model

NETWORK REQUIREMENTS FOR HIGH-SPEED REAL-TIME MULTIMEDIA DATA STREAMS

Speech Performance Solutions

Performance Analysis of Interleaving Scheme in Wideband VoIP System under Different Strategic Conditions

Simulation Based Analysis of VOIP over MANET

Performance Evaluation of VoIP Services using Different CODECs over a UMTS Network

VoIP. Wireless, P2P and New Enterprise Voice over IP

Analysis and Simulation of VoIP LAN vs. WAN WLAN vs. WWAN

MULTI-STREAM VOICE OVER IP USING PACKET PATH DIVERSITY

MultiDSLA. Measuring Network Performance. Malden Electronics Ltd

Methodology for Emulating Self Organizing Maps for Visualization of Large Datasets

ETSI TS V1.1.1 ( )

The Analysis and Simulation of VoIP

A Quality of Experience based Approach for Wireless Mesh Networks*

Perceived Speech Quality Prediction for Voice over IP-based Networks

Multiple Layer Traffic Engineering in NTT Network Service

Agilent Technologies Performing Pre-VoIP Network Assessments. Application Note 1402

Voice Quality Measurement in perfsonar

How to Measure Network Performance by Using NGNs

Task Dependency of User Perceived Utility in Autonomic VoIP Systems

Voice Quality Evaluation of a Call Using Fuzzy Logic

IP Telephony Terminal Solutions for Broadband Networks

Padma Charan Das Dept. of E.T.C. Berhampur, Odisha, India

12 Quality of Service (QoS)

QoS in VoIP. Rahul Singhai Parijat Garg

Troubleshooting Common Issues in VoIP

A New Adaptive Redundancy Control Algorithm For VoIP Applications

White Paper. Comparison between subjective listening quality and P.862 PESQ score. Prepared by: A.W. Rix Psytechnics Limited

Adaptive Coding and Packet Rates for TCP-Friendly VoIP Flows

QoS issues in Voice over IP

Adaptive Rate Voice over IP Quality Management Algorithm

Curso de Telefonía IP para el MTC. Sesión 2 Requerimientos principales. Mg. Antonio Ocampo Zúñiga

Analog-to-Digital Voice Encoding

NCTU-VT: a freeware for wireless VoIP performance measurement

VoIP 101. E911-Enhanced 911- Used for providing emergency service on cellular and internet voice calls.

Voice Transmission over Wireless Sensor Networks

Measurement of V2oIP over Wide Area Network between Countries Using Soft Phone and USB Phone

QoSIP: A QoS Aware IP Routing Protocol for Multimedia Data

Real Time Analysis of VoIP System under Pervasive Environment through Spectral Parameters

A Network Performance Application for Modeling, Simulation, and Characterization of Packet Network Behavior

ultra fast SOM using CUDA

THE ANALYSIS AND SIMULATION OF VOIP

The PESQ Algorithm as the Solution for Speech Quality Evaluation on 2.5G and 3G Networks. Technical Paper

IPv6 Broadband Access Network Systems

Statistical Measurement Approach for On-line Audio Quality Assessment

New Models for Perceived Voice Quality Prediction and their Applications in Playout Buffer Optimization for VoIP Networks

Voice Quality Planning for NGN including Mobile Networks

Network Sensing Network Monitoring and Diagnosis Technologies

Research and Development of IP and Optical Networking

PARAMETRIC SIMULATION OF IMPAIRMENTS CAUSED BY TELEPHONE AND VOICE OVER IP NETWORK TRANSMISSION

Visualization of Breast Cancer Data by SOM Component Planes

Network Virtualization Server for Adaptive Network Control

A Generic Algorithm for Midcall Audio Codec Switching

Visualization of large data sets using MDS combined with LVQ.

A Computational Framework for Exploratory Data Analysis

Fundamentals of VoIP Call Quality Monitoring & Troubleshooting. 2014, SolarWinds Worldwide, LLC. All rights reserved. Follow SolarWinds:

Chapter 4. VoIP Metric based Traffic Engineering to Support the Service Quality over the Internet (Inter-domain IP network)

Introduction and Comparison of Common Videoconferencing Audio Protocols I. Digital Audio Principles

Question: 3 When using Application Intelligence, Server Time may be defined as.

White Paper. ETSI Speech Quality Test Event Calling Testing Speech Quality of a VoIP Gateway

Gauging VoIP call quality from WLAN resource usage

Application Note. Pre-Deployment and Network Readiness Assessment Is Essential. Types of VoIP Performance Problems. Contents

Objective Speech Quality Measures for Internet Telephony

Visualization of Topology Representing Networks

Handover Management based on the Number of Retries for VoIP on WLANs

VoIP Analysis Fundamentals with Wireshark. Phill Shade (Forensic Engineer Merlion s Keep Consulting)

VoIP network planning guide

Basic principles of Voice over IP

Using Data Mining for Mobile Communication Clustering and Characterization

Application Note. Introduction. Definition of Call Quality. Contents. Voice Quality Measurement. Series. Overview

PART III. OPS-based wide area networks

IP-Telephony Quality of Service (QoS)

Application Notes. Introduction. Sources of delay. Contents. Impact of Delay in Voice over IP Services VoIP Performance Management.

Implementation of Video Voice over IP in Local Area Network Campus Environment

Introduction to Packet Voice Technologies and VoIP

Self Organizing Maps: Fundamentals

for guaranteed IP datagram routing

Index Terms Audio streams, inactive frames, steganography, Voice over Internet Protocol (VoIP), packet loss. I. Introduction

Research on Video Traffic Control Technology Based on SDN. Ziyan Lin

Simulation of SIP-Based VoIP for Mosul University Communication Network

A Study of Web Log Analysis Using Clustering Techniques

Classification of Engineering Consultancy Firms Using Self-Organizing Maps: A Scientific Approach

Load balancing in a heterogeneous computer system by self-organizing Kohonen network

VOICE QUALITY ESTIMATION IN WIRELESS NETWORKS

Accuracy Analysis on Call Quality Assessments in Voice over IP

JOURNAL OF TELECOMMUNICATIONS, VOLUME 2, ISSUE 2, MAY Simulink based VoIP Analysis. Hardeep Singh Dalhio, Jasvir Singh, M.

VoIP Conformance Labs

IP QoS Interoperability Issues

INTERACTIVE DATA EXPLORATION USING MDS MAPPING

Voice and Fax/Modem transmission in VoIP networks

Transcription:

QoS Mapping of VoIP Communication using Self-Organizing Neural Network Masao MASUGI NTT Network Service System Laboratories, NTT Corporation -9- Midori-cho, Musashino-shi, Tokyo 80-88, Japan E-mail: masugi.masao@lab.ntt.co.jp Abstract: This paper proposes a QoS mapping method of VoIP communications for real network environments. To totally take acount of the effects of several QoS-related parameters, we used a self-organizing neural network, which can map high-dimensional data into simple geometric relationships on a low-dimensional display. For training the self-organizing neural network, we measured sevaral QoS-related parameters such as PSQM+, end-to-end delay, and packet loss rate for three VoIP systems. Evaluation results confirmed that our method can effectively evaluate the total QoS level composed of several QoS-related factors. Keyword: VoIP, QoS mapping, PSQM, Self-organization I. Introduction Voice over IP (VoIP) enables voice data integration over IP networks, reducing the network transmission cost for IP protocol users. Due to the shared nature of current network structures, however, it is difficult to guarantee the quality level of voice speech. The quality of service (QoS) of VoIP depends on several factors such as network delay, packet loss rate, and the kinds of codec, so that it is naturally categorized as a best effort service. In order to perform appropriate evaluations of voice speech quality degradation, there have been various studies and proposals for packet-based voice communications []- []. For example, PSQM (Perceptual Speech Quality Measurement) and PSQM+, which can generate a perceptual distortion level for each voice frame, have been adopted as ITU-T Recommendation P.86 for assessing voice speech quality [],[6]. In addition, PESQ (Perceptual Evaluation of Speech Quality) has been also standardized as Recommendation P.86 in ITU-T [7], [8]. However, especially for interactive communications, the QoS level of packet-based voice network systems depends on many factors such as end-to-end delay, so estimating the speech quality level only is not enough for executing the end-to-end evaluations. To account for compound factors in estimating the QoS level of the end-to-end network system, the E-model, which can deal with several QoS-related factors including the voice speech quality and end-to-end delay, has also been proposed [9]. The E-model-based method is effective for evaluating the QoS level of VoIP communications, because it can deal with several QoS-related factors including the speech quality level and end-to-end delay. However, this method outputs only the calculated QoS value, and the position of each factor in the result and the correlation of results among other QoS levels cannot be efficiently displayed. Therefore, a QoS level mapping technique that can show multilateral aspects of the QoS evaluation is required. In addition, when some of the components for its calculation cannot be obtained, it does not necessarily provide appropriate outputs, meaning that its applicability is limited in evaluating the QoS level in real network environments. Incidentally, a self-organization-based mapping model [0], [] is an effective tool that can clarify the relative relationships in high-dimensional input data. Based on this method, nonlinear statistical relationships in highdimensional data can be converted into a two-dimensional space, while preserving the metric and topological relationships of the input data. As a result, this mapping model can be used to evaluate and categorize the relative relationships of high-dimensional input data. This paper describes a QoS mapping of VoIP communications for objective evaluations in real environments. To totally take into account the effects of several QoS-related factors, we used a self-organizing neural network, which can map high-dimensional data into simple geometric relationships on a low-dimensional display. Section two shows the basic flows of the QoS mapping and evaluations for VoIP communications. Section three presents case studies using several parameters such as PSQM+ and end-to-end delay for three VoIP systems. Section four summarizes the results of this paper and mentions on further studies for this subejects. II. QoS evaluation of VoIP communication using self-organizing mapping A. Flow of self-organization [0],[] The self-organizing algorithm can calculate multidimensional parameters so that they optimally denote the domain, in which the relationships of primary data are preserved topologically. In this paper, a two-dimensional map is employed to model the QoS level of VoIP communications. The basic training process of the self-organization model is defined as

m i (t+) = m i (t) + h c(x),i (x(t) m i (t)) () where m is the weight vector, x is the input vector, h c(x),i is the neighborhood function, i is the node number of an output layer, and t is the regression step index. The concept of this self-organizing map is shown in Fig., where the n-dimensional vector x is projected onto the output layer. In this process, the input vector x is compared with all m i, and the subscript c(x) is defined by the Euclidean condition x m c = min{ x m i }, () i where m c is the winner that best matches x. Here, the initial value of m i is set to random value, and the Gaussian type neighborhood function can be given by h c(x),i =α(t) exp{ r i r c / σ (t) }, () where 0 < α(t) < is the learning rate parameter, r i R and r c R are the vertical locations on the grid, and σ(t) corresponds to the width of the neighborhood function. Also, assuming that T is the total training number, α(t) and σ(t) can be defined as α(t) = α ( t / T) () σ(t) = σ(t) + {σ σ(t)}( t / T). () The procedure of this training process can be described as follows; (a) Initialize m i to a random value, (b) Input x(t), one at a time, (c) Calculate eq.(), and find m c, (d) Calculate eq.() using eqs.() - (), (e) Repeat from (b). Output layer neighborhood radius is set to a small value to fine-tune the map. B. Application to QoS mapping of VoIP communications The concept of QoS evaluation of VoIP network communications is shown in Fig., where the input data for training the self-organizing map can be given by the target VoIP communication condition. In this training, two categories of variables are input into the self-organizing map: one is the QoS-related parameters and the other is the system identifiers that indicate the basic system performance level and properties. As for the QoS-related parameters, we focus on two aspects for both the voice speech level and the delay impairment, and select PSQM+ and end-to-end delay as input variables to the self-organizing map. Incidentally, the packet loss is a major source of speech impairment in VoIP communications, and is a very important parameter that shows the degree of loss patterns on the voice speech quality. Therefore, to supplement QoS evaluations, the packet loss rate is also employed as an input variable to the self-organizing map. In this paper, in terms of the QoSrelated input parameters to the self-organization map, three parameters --- PSQM+, end-to-end delay, and packet loss rate --- were measured in a real environment, and they were used to evaluate the total QoS level among other conditions. Level Tel Time VoIP GW Measurement of data t (Delay) IP network Level VoIP GW Time Tel Data for the training m ij Input :(variable_(),variable_(),,variable_n()) Input :(variable_(),variable_(),,variable_n()) Input N:(variable_(N),variable_(N),,variable_n(N)) x x x n Input data Fig. Configuration of the self-organizing map. Self-organizing map Training Fig. Concept of QoS evaluation procedure for VoIP communication (GW: gateway). To raise the training efficiency, the algorithm is performed in two phases. In the first phase, a relatively large initial neighborhood radius is used to tune the map approximately. Then, in the second phase, the initial On the other hand, as for the system identifiers that correspond to the basic system specifications, we use the PSQM+ for no background load conditions between the originating and terminating VoIP gateways, which can be

regarded as the standard PSQM+ for each system. The QoS level of VoIP communications depends on network system factors such as the kind of codec, and inputting the standard PSQM+ helps to map target data when some of the input variables are lost. To obtain the standard PSQM+ as an input variable, we directly connect the originating and terminating gateways via a switch and measure it for no background data conditions. By inputting the measured QoS-related parameters and system identifiers to the self-organizing map, we can map the QoS level of VoIP communications on a twodimensional space. III. Case study A. Setup for training data The test-bed for measuring QoS-related parameters is shown in Fig., where both a LAN (00 Base-T) and the Internet are used as transmission networks. In this measurement setup, three types of VoIP gateways (GWs) were used, and we define the connection patterns as : GW- to-gw as System, GW-to-GW as System, and GW- to-gw (PC-to-PC using VoIP communication software) as System. As for the transmission network, the LAN was used for Systems and, both the LAN and Internet were used for System, and the bandwidth of access lines between the Internet and VoIP gateways was 8kbps. GW GW GW Source signal Originating side R SW Voice measuring device Packet loss generator LAN Internet GW GW GW When measuring QoS-related parameters of PSQM+ and the end-to-end delay for these systems, voice sample data [], based on ITU-T Recommendation P.800 [], was transmitted. After the voice sample data was sent from the originating gateway to the terminating one, the source and transmitted voice signals were compared, and then the values of PSQM+ were calculated by a voice measuring device. In this calculation, the time-averaged results of all PSQM+ for each frame of 6 ms were calculated across the voice sample files (five files in Japanese language, each s long). The end-to-end delays were also measured by SW Transmitted signal Terminating side Fig. Test-bed for measuring QoS related parameters (GW: gateway, SW: switch, R: router). R comparing the leading parts of transmitted and received voice signals. Here, the mesurement condition for three VoIP systems and the voice sample data used is shown in Table. Incidentally, for the measurements in the LAN, two conditions were employed: for one, there was no other background data in the network and for the other, there was. In the case when there was no other background load in the LAN, the packet loss based on the Poisson distribution was generated at the transmission network part by a packet loss generator. Furthermore, we also confirmed that the processing time caused by the switch was less than ms for these VoIP systems. Item System setup Voice file Table Measurement setup Condition (a) Voice packet sending period - System : 0 ms - System : 0 ms - System : 0 ms (b) Type of codec G.79, G.7 - Language : Japanese - Time length : s B. Training with measured data Examples of measured PSQM+ and end-to-end delay for no background load in the LAN are shown in Fig., where measured values of five voice files were averaged for each system. Here, the packet loss rates were set in the ranges of 0 to 9%, and both PSQM+ and end-to-end delay were measured for three systems. Fig.(a) shows that PSQM+ tended to increase with the packet loss rate for both G.7 and G.79, and the order of voice quality level was Systems,, and. Also, Fig.(b) shows that the end-to-end delays tended to increase in the order of Systems,, and, and the processing speed using G.7 was almost the same as that for G.79. Figure shows examples of measured PSQM+ and endto-end delay when there was background data in the transmission network (both the LAN and the Internet). The values of packet loss rates were not measured in these cases, meaning that the trainings were performed without this variable. From Fig., we confirmed that the QoS levels for both PSQM+ and end-to-end delay were degraded when there was other background data in the transmission network. Based on measured data in Figs. and, the training was performed to evaluate the QoS level of VoIP communications for three systems. In this process, a rectangle was used as the map topology type, and the map size was set to 8 8. Here, we set the total training number to 000.

PSQM+ 8 Also, to raise the training efficiency, we performed the training in two phases []: in the first phase with training number of 000, the initial value of learning rate parameter α was set to 0., and in the second phase it was set to 0.0. In addition, for the neighborhood radius, initial value σ and final value σ(t) in the first phase were set to and, while the values in the second phase were and respectively. End-to-end delay (ms) PSQM+ 0.8 0 Packet loss rate (%) 0 System = : G.7, : G.79 System = : G.7, : G.79 System = : G.7, : G.79 (a) Relationship between packet loss rate and measured PSQM+ 0 00 0 00 0 0 0 6 9 0 6 9 0 6 9 0 6 9 0 6 9 0 6 9 G.7 G.79 G.7 G.79 G.7 G.79 System System System Packet loss rate (%) (b) End-to-end delay of each system Fig. Examples of measured PSQM+ and end-to-end delay (LAN, no background data). 8 0 00 00 00 00 700 End-to-end delay (ms) System = System = System = : G.7, : G.79 : G.7, : G.79 : G.7, : G.79 : via the LAN, : via the Internet Fig. Examples of measured PSQM+ and endto-end delay when there was background data. C. Evaluation Figure 6 shows a visualization result projected onto twodimensional domain, in which the axes are set as x and y. The results for Fig. 6 are as follows; - The longer the distance from the left-bottom points became, the more PSQM+ tended to increase, meaning that the voice quality level tended to decrease with increases in x and y. - The left part corresponds to the domain where the end-toend delay level was relatively low, and the right upper part is equivalent to the domain where end-to-end delay was highest. - The packet loss rate tended to increase along the y-axis, so we estimate that the voice quality level tended to degrade along the y-axis. y (9) (9) (9) (9) (9) (9) () () x () () : System [G.7], : System [G.79] : System [G.7], : System [G.79] : System [G.7], : System [G.79] ( ) : Numerical value represents packet loss rate : via the LAN which has background load : via the Internet Fig. 6 Visualization result of the training data. As shown in Fig. 6, the shorter the distance from the left bottom became, the higher the total QoS level of VoIP communications tended to become. For lecture-type () ()

communications, in which interactive conversation scene is not necessary, however, the priority of the end-to-end delay factor can be sometimes low, so the mapping position may be allowed to shift to the right side in two-dimensional space. Our method can deal with multi-dimensional QoSrelated parameters and project the results onto a twodimensional space, so we can effectively evaluate the positioning of QoS level for each condition composed of several variables. IV. Conclusion This paper proposed a QoS mapping method of VoIP network communications for real network environments. To totally take account of the effects of several QoS-related factors, we used the self-organizing neural network, which can map high-dimensional data into simple geometric relationships on a low-dimensional display. By employing a self-organizing training scheme, we showed that our method can combine multi-dimensional QoS-related parameters and project the results onto a two-dimensional space, so that we can effectively evaluate the positioning of QoS level for each condition composed of several variables. In this training, two categories of variables were input into the self-organization map: the QoS-related parameters and the system identifiers that indicate the system performance level and properties. In terms of the QoSrelated input parameters to the self-organizing map, we measured three parameters --- PSQM+, end-to-end delay, and packet loss rate --- in the test-bed, and used them to evaluate the total QoS level. On the other hand, as system identifiers that correspond to the basic system specifications, we used the PSQM+ for no background load between the originating and terminating VoIP gateways, which can be regarded as a standard PSQM+ for each system. Evaluation results confirmed that our method effectively evaluates total the QoS level composed of several QoS-related factors. We also confirmed that this method can deal with input data that lacks one of its variables, and can help to estimate its unknown QoS level. Future study will include QoS evaluations of input data using other QoS-related variables, further evaluations of precision and applicability in real environments, and application to other multimedia communications. References [] C. J. Weinstein, Experience with speech communication in packet networks, IEEE J. Selected Areas. Commun. vol.sac-, no.6, pp.96-980, Dec.98. [] S. Voran, Objective estimation of perceived speech quality part : Development of measuring normalized block technique, IEEE Trans. Speech and Audio Processing, vol. 7, no., pp. 7-8, July 999. [] S. B. ZahirAzami, A. Yongacoglu, L. Orozco- Barbosa, and T. Aboulnasr, Evaluating the effects of buffer management on voice transmission over packet switching networks, Proc. of ICC 00(Helsinki, Finland), pp. 78-7, June 00. [] M. Masuda and K. Ori, Network performance metrics in estimating the speech quality of VoIP, proc. APSITT 00 (Kathumandu, Nepal and Atami, Japan), pp. - 7, Nov. 00. [] ITU-T Recommendation P.86, Subjective quality measurement of telephony-band (00-00 Hz) speech codecs, Aug. 996. [6] ITU-T Contribution COM -0-E, Improvement of the P.86 perceptual speech quality measure, KPN research, Netherlands, Dec. 997. [7] ITU-T Recommendation P.86, Perceptual evalua- tion of speech quality (PESQ), an objective method for endend speech quality assessment of narrow-band telephone networks and speech codecs, Feb.00. [8] A. Takahashi, Performance evaluation of objective speech quality measure, ITU-T Recommendation P.86 PESQ, The 6th Int. workshop on communcation quality & reliability (CQR00, Okinawa), pp.9-, May 00. [9] ITU-T Recommendation G.07, The E-model, a computational model for use in transmission planning, May 000. [0] T. Kohonen, Self-organization and association memory, Proc. IEEE, vol. 78, no.9, pp. 6-80, 990. [] T. Kohonen, Self-organizing maps, Springer, Berlin, Heidelberg (Second extended edition), chapter, 997. [] NTT-AT, Multi-lingual speech database for telephonmetry 99, NTT-AT CD-ROM, 99. [] ITU-T Recommendation P.800, Methods for subjective determination of transmission quality, Aug. 996. Acknowledgements I would like to thank Mr. Hajime Sugawara, Tsuyoshi Takenaga, and Hiroyuki Oouchi of NTT Network Service Laboratories for their contributions to this work.