VoIP performance on Local Area Networks Anton Kos, Aleš Vugrinec and Sašo Tomaži č Abstract—This paper investigates the possibilities of real-time voice transmission over local area networks networks that provide no guarantees of quality of service. The network of choice is 10 Mbit/s Ethernet and voice coder of choice is G.723.1. Both are widely used. At the beginning we define the requirements for real-time voice transmission and give the basic properties of G.723.1 coder. Delay is identified as the key parameter to real-time voice transmission. Then we identify the most common simplifications that are introduced on the way to theoretical results. We present results of a bimodal delay analysis that is tailored to the transmission of VoIP frames. We identify the upper bounds of network utilisation under which the delay is within acceptable range. Next we present the measurement results for different levels of network utilisation and compare them with simulation results that matched the measured network. Results of both are very similar and demonstrate that simulations are a good tool to predict VoIP behavior on the network. Despite the fact, that theoretical results were not directly comparable with measurements and simulations, they still give good guidelines about network operation under different conditions. All three methods thus complement each other. Index Terms—VoIP, real-time voice transmission, VoIP simulation, VoIP measurements, VoIP performance Ethernet ---------------------------------  --------------------------------- 1 INTRODUCTION HE purpose of this paper is to investigate the possibility of voice transmission of sufficient quality over Ethernet networks with the help of theoretical analysis, simulations and measurements on a real system. As elements of the system we chose standards and technologies most used for this purpose. The Ethernet network considered is IEEE 802.3 10BaseT, and the voice coder ITU G.723.1. We chose 10BaseT for various reasons. During the implementation of simulations and measurements, this technology was dominant in terms of the number of connections. In addition, it also represents an extreme case in terms of delay, since more modern versions of Ethernet, such as 100BaseT and others, are better in this respect. We chose G.723.1 because this is the only voice coder available in all considered VoIP devices. First we evaluate the capabilities of Ethernet networks and recognise their advantages and limitations. Here we pay most attention to analysis of delays under different network loads and different proportions of short frames to long frames, since this parameter is the key to real-time voice transmission. Next, we discuss the results of measurements on a real system, which we implemented in the laboratory, and compare them to simulations of the same measuring system. The analysis is not limited to general analysis of Ethernet frame delays, but we also provide an analysis of voice frame delays at different levels of network utilisation. 2 VOICE TRANSMISSION REQUIREMENTS Real-time voice transmission is a challenging task for networks, since they must meet the following requirements, some of which are intertwined: • sufficient available bit rate, • sufficiently low delays, • echo cancellation, • bandwidth reservation, and • low bit-error rate. Some of these requirements can be defined with absolute values, while others depend on the coding methods used and the properties of the transmission network. For real-time transmission, simultaneous compliance with the bit-rate, bandwidth reservation and particularly delay criteria is essential. Let us look in more detail at the delay criterion, which is the most important in terms of voice transmission over Ethernet networks 1 . Delay must be within certain limits, otherwise the quality of voice and interactivity of communication declines significantly. The upper limits of acceptable delay for voice transmission under recommendation ITU G.114 [9] are: • delay of up to 150 ms is suitable for most user and voice applications, • delay of between 150 and 400 ms is acceptable, provided that the network operator and the user are aware of the impact of increased delay on the quality of voice, • delays above 400 ms are unacceptable for most user and voice applications. 3 BASIC PROPERTIES OF VOICE CODER The G.723.1 voice coder is part of the wider set of standards H.323, which are intended for real-time data transmission of multimedia applications over local networks that provide no guarantees of QoS. Transmission of coded speech uses protocol stack RTP/UDP/IP, which adds a further 40 octets of overhead to a speech frame 24 octets long. For G.723.1, a typical value of delay due to speech processing is 67.5 ms. This consists of the time required to generate a speech signal belonging to one speech frame (30 ms), look-ahead in the 1 Frame loss is a very important measure that strongly influences the quality of reconstructed speech. In Ethernet networks, frame losses arise primarily due to excessive delays, and so we will not discuss them in detail here (for details, see [6]). ———————————————— • Anton Kos is with the Faculty of Electrical Engineering, University of Ljubljana, Slovenia. E-mail: anton.kos@fe.uni-lj.si. • Aleš Vugrinec is with the Ministry of Defense, Slovenia. E-mail: ales.vugrinec@gov.si. • Sašo Tomažič is with the Faculty of Electrical Engineering, University of Ljubljana, Slovenia. E-mail: saso.tomazic@fe.uni-lj.si T