VI INTERNATIONAL TELECOMMUNICATIONS SYMPOSIUM (ITS2006), SEPTEMBER 3-6, 2006, FORTALEZA-CE, BRAZIL Evaluation of Wireless Links in a Hybrid Network Structure for Mobile VoIP Services D. F. S. Santos, J. L. Nascimento, O. C. Silva Filho and A. Perkusich Abstract Wireless technologies and new Internet services are making available new kinds of solutions and driving new users needs. In order to guarantee the quality of services provided by these solutions, experimental analysis plays a key role. This article presents an evaluation of a solution to provide voice services for users with mobile devices based on the Internet. We also consider a scenario with different wireless network technologies, more specifically Bluetooth and Wi-Fi. The experimental results were obtained based on a functional mobile VoIP infrastructure, named M-VoIP. Based on the results it was possible to identify the major limitations of the M-VoIP solution. Index Terms VoIP, Bluetooth, Jitter, Wi-Fi I. INTRODUCTION The number of services available on the Internet is growing fast in the last few years. Besides aggregating new types of services, the availability of the traditional ones, such as television, radio and telephony is increasing as well. The VoIP (Voice over IP) is one of the most promising new generation services. Providers of VoIP services are rapidly spreading services everyday. On the other hand, wireless technologies such as Bluetooth and Wi-Fi are being embedded in different kind of mobile devices, such as cellular phones, personal digital assistants (PDAs) and internet tablets. These devices, together with wireless access points are making possible the deployment of wireless local area networks and personal area networks (piconets), both supporting the access to different services, such as VoIP. Following this trend, a system that integrates VoIP services with mobility support and multiple wireless networking technologies can be seen as an ideal solution. The M-VoIP project (Mobile Voice over IP) [1] is an effort to provide a complete working solution that makes use of both WPAN (Wireless Personal Area Networks) and WLAN (Wireless Local Area Networks) to enable VoIP connections. The M-VoIP goal is to provide voice communication as well as traditional telephony services to mobile users, using both Wi-Fi and Bluetooth technologies. However, dealing with voice over wireless data communication links is a challenging problem. WLAN technologies (such as Wi-Fi) can deal well with this kind of problem, but WPAN technologies (such as Bluetooth) imposes challenges related to their low range and throughput characteristics. Also, voice requirements related to delay, jitter and minimum bandwidth are critical for wireless links. Although technologies such as Bluetooth offers SCO (Synchronous Connection Oriented) The authors are with the Embedded Systems and Pervasive Computing Laboratory, Department of Electrical Engineering, Federal University of Campina Grande, C.P. 10105-58109-970 - Campina Grande - PB - Brazil, emails: (danilo, jluisn, olympio, perkusic)@dee.ufcg.edu.br voice links, they experience high levels of packet loss and the number of users or devices connected in the network is more constrained than in ACL (Asynchronous Connectionless Link) [2]. So, the use of a connectionless scheme is more suitable when considering scenarios where multiple users of TCP/IP traffic are present [3]. However, Bluetooth implementation of the link layer employ packets retransmission mechanisms for their connectionless schemes, and this can be prejudicial for real-time traffic such as VoIP [4]. In this paper we present analysis of the performance of the M-VoIP system and its Bluetooth and Wi-Fi links. The analysis is focused on the the delay and the jitter behavior during the transmission of voice packets taking into account different wireless network technologies. By the evaluation of experimental results it is shown that the retransmission feature of Bluetooth cause problems to transmission of voice packets that have real-time restrictions. The paper is organized as follows: In Section 2 basic concepts related to VoIP and Wireless technologies are presented. In Section 3 the M-VoIP system architecture is described. Section 4 describes the evaluation experiments. In Section 5 the results are analyzed. The last section presents the conclusion and final remarks. II. BACKGROUND In this section we present basic concepts related to the key technologies used in the M-VoIP system, namely VoIP and Bluetooth. A. VoIP Voice over Internet Protocol (VoIP) is a technology that allows digitalization, packing and transmission of sampled voice data over a packet switched network based on Internet protocol. The sampling process of the voice signal consider the frequency range that carries the most significant part of the information, between 0.3 and 3.4 KHz. According to the Nyquist s Sampling Theorem, this signal should be sampled with at least the double of the higher frequency present on it, so that there is no information loss the sampling frequency normally adopted is 8,000 Hz, in the case of voice. Each sample can then be quantized in a predetermined number of levels (levels = 2 Nobits ). Let us consider for example that the sample frequency is 8 HHz, and the digitalization process uses 8 bits. The resulting digital signal bandwidth is 64,000 bits/s with 8 bits per sample. This codification scheme is an ITU- T standard known as PCM (Pulse Code Modulation). There SBrT 298
are many different codification schemes that address different requirements such as low bandwidth or high voice quality. Another feature of VoIP is that it uses packet switching instead of circuit switching. This introduces many technical advantages such as: in the case of packet switching there is no dedicated circuits; information paths are shared with different users, thus increasing the system efficiency; bandwidth can be allocated on demand, among other features. However, the use of a packet switched network also rises some problems. Among then, we emphasize: packets can arrive out of order; packets can experience variable delays; it is difficult to provide QoS (Quality of Service) assurances; packet loss affects voice quality. In order to cope with some of the previously mentioned problems, more robust transport protocols can be used. One of these protocols is RTP Real Time Protocol - IETF RFC 3550. RTP is used over the UDP protocol User Datagram Protocol - IETF RFC 768. The RTP protocol includes functionalities for packet ordering over IP networks. Also, RTCP (Real Time Control Protocol) is used to exchange QoS parameters so that QoS can be improved. The variable delay problem (also known as delay jitter) is the main issue that affects the quality of voice communication. Such problem can be minimized through the use of dejitter buffers. The inter arrival Jitter is calculated continuously as each data packet i is received [5]. The way to do this is shown as follows: J = J 0 + ( D(i 1,i) J 0) (1) 16 where: J 0 is the previous jitter and D is known as the delay difference. It is important to point out that the 1/16 gain parameter defined in Equation 1 allows a good noise reduction ratio while maintaining a reasonable rate of convergence [6]. The delay difference can be calculated as follows: D(i, j) =(R j R i ) (S j S i ) (2) =(R j S j ) (R i S i ) where: R is the arrival time defined based on the RTP time stamps and S is the RTP time stamp. The most common delay sources are switching, routing, queuing and delay associated with the packet processing task. Due to so many sources of delay and to the real-time requirements of voice applications, the ITU standard G.114 [7] states that the end-to-end delay should be limited to 150 ms and packet loss must remain less than 5% in order to maintain quality of the voice signal. B. Bluetooth Bluetooth is an open specification for radio systems that provide short range wireless communication. Developed by Ericsson in 1994, it received support from Intel, IBM, Toshiba, Nokia, Lucent, Motorola and others. The Bluetooth Special Interest Group (SIG) founded by such companies is the body that supervises the development of Bluetooth standards and the licensing of the Bluetooth technologies [8]. Bluetooth uses spread spectrum to hop between 79 channels, 1 MHz apart from 2.402 to 2.480 GHz. This frequency band [9] can be used without any further authorization and can be shared, for example, with the Wi-Fi technology [10]. Bluetooth links can be ACL (Asynchronous Connectionless) for data communication and SCO (Synchronous Connection Oriented) for audio communication. The SCO link is a symmetric pointto-point connection. In this case there is a master node that can support up to 3 simultaneous connection to slaves. This link is recommended to be used in time-sensitive applications such as voice transmission. The ACL link can support both symmetric and asymmetric communication. In this case there is a master node that supports up to 7 slaves connected to it. ACL and SCO uses different types of packets that have different characteristics. The most important are: FEC (Forward Error Correction) used to correct errors in the transmission and to allow the reduction of the number of retransmissions; CRC (Cyclic Redundancy Check) used to detect errors in the transmission and to allow the receiver to ask the sender a retransmission of a packet. In tables I and II a summary of the packets accordingly to the type of the link is presented. TABLE I ACL LINK PACKET TYPES Type FEC CRC Symmetric rate (kb/s) DM1 2/3 yes 108.8 DM3 2/3 yes 387.2 DM5 2/3 yes 477.8 DH1 no yes 172.8 DH3 no yes 585.6 DH5 no yes 723.2 TABLE II SCO LINK PACKET TYPES Type FEC CRC Symmetric rate (kb/s) HV1 1/3 no 64.0 HV2 2/3 no 64.0 HV3 no no 64.0 Besides the FEC technique, another error correction technique used by Bluetooth technology is known as ARQ (Automatic repeat ReQuest). In this case packets retransmissions are used to assure error-free data transfer. Thus, packets of types DV (Data Voice), DM (Data- Medium rate) and DH (Data- High rate) are retransmitted until a positive acknowledge is received, or a timeout is exceeded [11]. The ARQ scheme protects only the payload of the data (the source of the CRC used to detect errors), the header and the voice payload (found in some types of packets) are not protected by it. The ARQ technique is used to bring reliability to the data link layer of the Bluetooth technology. It is well suited when such link is used to transport data. However, for applications such as VoIP, that uses this kind of link to transmit timesensitive data like voice, this scheme can be harmful. When SBrT 299
the link quality becomes bad, the approach of making multiple retransmissions increases packet delay. In the case of certain types of communications links, such as voice over IP, only a small amount of time delay is allowed. For such kind of links, there is a mechanism specified by Bluetooth technology known as flushing. This mechanism allows a payload, that is blocked in the transmission queue due to successive retransmissions, to be discarded. This mechanism is optional [11] although highly recommended. III. M-VOIP SYSTEM ARCHITECTURE In this section the M-VoIP system architecture is described [1]. As stated in the introduction, the M-VoIP system aims to provide a network architecture coverage with different technologies, more specifically Bluetooth and Wi-Fi. Gateways Bluetooth/Wi-Fi are used on the wireless network architecture, thus creating a mobile hybrid network that makes connectivity available for users through different network technologies within different wireless cells. Also, a Mobile IP [12] system was implanted to keep users on line when hand offs occurs between different networks. The infrastructure is connected to a VoIP PBX (Private Branch exchange) that controls the VoIP traffic in the network, and also provides interconnection to the PSTN (Public Switched Telephony Network) through a telephony gateway, as illustrated in the Figure 1. As can be observed in Figure 1, the solution can be divided in two different subsystems, namely: the wireless network, and the VoIP system. Observe that both can operate transparently of the underlying network technology. The VoIP server is composed by an Asterisk PBX [13], which is a convergent telecommunication platform. It was designed to allow the use of VoIP, offering support for connections with the PSTN. It offers various services from lower layers, including management of time division multiplexing (TDM) and packet-based telephony, to upper layers as well, including typical PBX applications such as Interactive Voice Response (IVR). Also, Asterisk is a open source platform compatible with the Linux OS. The Asterisk system was installed with a telephony board model TDM400P [14] providing access to the PSTN. Thus, a mobile user can make calls from its softphone installed on a handheld, to any telephone in the PSTN. As mentioned earlier, a Bluetooth/Wi-Fi gateway was deployment for the deployment of the wireless hybrid network This gateways provide device interconnection between different networks. They are also responsible by the configuration and attribution of IP address to the devices in the network. The communication process is transparent to the user. This is due to the fact that only a wireless connection is needed, either Wi-Fi or Bluetooth. To deploy the Bluetooth cell the Bluetooth Network Encapsulation Specification (BNEP) was used [15]. Therefore a Bluetooth piconet is created. The specification describes the procedures to send common networking protocols such as IP datagram s over Bluetooth links. The profile implemented by the Bluetooth peers is called PAN, and it is supported by the open source Bluetooth Bluez [16] implementation. In a PAN network the Bluetooth master device is called NAP (Network Access Point) and the slaves in the piconet are called PANUs (PAN Users). The Bluetooth cells are confined inside a Wi-Fi cell. The connection between them is made through Ethernet cables. This connection is established using Wi-Fi bridges, with only one of them accessing the external network. The gateways were implemented based on the Linux OS. IV. EXPERIMENTS In this section the description of the procedures carried out to evaluate the wireless links of the M-VoIP system are detailed. This evaluation is based on the measurement of jitter, delay and packet loss during VoIP calls between handhelds equipped with Wi-Fi and Bluetooth (Class 2) cards as wireless interfaces using the M-VoIP system. The voice codec used for these experiments was the ITU-T G.711, where the length of each packet is 216 bytes with RTP encapsulation. The bandwidth required by this type of codec is 64,000 bps. The length of voice frames are 20ms and each frame was encapsulated in one packet to avoid fragmentation. A. Evaluation Procedure To evaluate the M-VoIP system, scenarios considering users with and without mobility were defined. In the first set of experiments, delay, jitter and packet loss were evaluated. Also, the communication between different Bluetooth cells in the M-VoIP system were taken into account. For this last case, we considered Bluetooth nodes without mobility, and also considered the communication of each node in a different cell with different access points. In Figure 2 such evaluation scenario is illustrated. Observe that the Bluetooth nodes were kept, throughout the experiment, inside the coverage range of the Bluetooth link. The goal of the use of this scenario is to verify and analyze the influence of the overall M-VoIP system structure on voice communication. There are many factors that can influence the results obtained with this experiment. Such factor are related to the network entities, e.g. the access points, the gateways and the voice communication server, network technologies used (Wi-Fi, Bluetooth, LAN), etc. Thus, the results of the experiments can give a solid basis to analyze the functionality and efficiency of the proposed VoIP structure. For the second set of experiments the same metrics were obtained considering static Bluetooth nodes in the same Bluetooth cell. With this configuration, illustrated in Figure 3, the results are influenced by a more restricted set of entities and technologies allowing a deep analysis of them. The network path is composed only by Bluetooth and Ethernet links. The experiments were carried out using Ethernet links with very low load. So, the influence of such links will not be considered on the analysis. In both last cases, metrics were measured with distances of 5 m, 10 m, and 15 m away from the Bluetooth access points. For result analysis, the mean and standard deviation of the data were take account. SBrT 300
Fig. 1. M-VoIP system architecture Fig. 2. First scenario considered for the evaluation Finally, in the third set of experiments the mobility between Bluetooth nodes inside the cell coverage was considered, as illustrated in Figure 4. It was used only one access point to connect both Bluetooth nodes. The experiment was carried out as follows. After the connection has been established between the nodes and the access point, one VoIP call was placed between the nodes. Then both nodes were moved with constant speed away from the starting point that was next to the access point. After 60 s both nodes were at the maximum distance of the access point and then they went back to the access point, using the same 60 s. Based on this scenario it was possible to analyze the behavior of the Bluetooth link with moving nodes and located at the limit of the coverage range. A point to be stressed is that Fig. 3. Second scenario considered for the evaluation the environment where the experiments of this last scenario were carried out was a hall. Such experimental condition can be used to explain the fact that range, usually around 10 m, of the Class 2 Bluetooth device [17] was not a problem at all. For each one of the scenarios presented above, the users maintain a normal conversation flow during all the measurement process. SBrT 301
Fig. 5. Jitter obtained based on the first scenario Fig. 4. Third scenario considered for the evaluation B. Measurements The measurement process was performed using a portable computer running the Linux OS and using the network analyzer tool Ethereal [18] and the hcidump Bluez [16] feature. Therefore, it was possible to capture packets for both the Ethernet wireless network and the Bluetooth based network. It is important to point out that packets for the network layer as well as frames for the physical layer were captured. For the physical layer only Bluetooth frames were captured. Also, based on this network analyzer tool, the RTP packets was separated of the other packets in the network and analyzed as detailed follows. The tool have also provided measurements related to the delay, the jitter and the packet loss. C. Results The results was obtained with similar conditions that encompasses the load in the Wi-Fi link and in the VoIP server. This was done to make possible the comparison between the results. Based on the first set of experiments a small loss of RTP packets was detected, around 2%. The medium jitter was around 14 ms with a standard deviation around of 1.1 ms, as can be observed in the graphic shown in Figure 5. In the case of the second set of experiments the results were similar to the first case, that is, with RTP packet lost around 2% and a medium jitter of 11 ms with a standard deviation of 1 ms, as as can be observed in the graphic shown in Figure 6. Also, the medium delay in both previous cases was around of 19 ms. The difference appears with the standard deviation, in the first case it was around 15 ms and the second was around 13 ms. Differently from the first two scenarios, for the third set of experiments the results were much worse. The RTP packet lost increased to 20%, also the medium jitter increased to 39.7 Fig. 6. Jitter obtained based on the second scenario ms and delay to 25 ms, with standard deviation of 28.5 ms and 152.75 ms respectively. The general behavior of jitter and delay obtained can be observer in the graphic shown in Figure 7. Observe that, for this case, the delay in the network was very high causing intelligibly problems in VoIP conversations. D. Analysis The first and second scenarios presented similar results. In both cases neither the Wi-Fi network interface nor Bluetooth were stressed (i.e the load in both was low). The processing delay and jitter generated by the VoIP server and Wi-Fi link is the same to both cases as pointed out previously. There is a small difference on the results presented in Figures 5 and 6. The mean of the jitter obtained for the first scenario was a little bigger than that obtained for the second scenario. This can be explained only by the insertion of one more Bluetooth/Wi-Fi gateway. However, the results obtained to the these scenarios show that jitter and delay range was limited to low values. The third scenario presented the worst performance. The Bluetooth link was stressed up to the limit of its coverage range. As can be seen in the Figure 7, the most significant delay values are confined inside the interval between 20 and 100s. In such interval the conditions in the Bluetooth link were the worst because the distance between the nodes and the SBrT 302
In fact, the problem of the Bluetooth link is mainly due to the ARQ technique used that increases the associated delays and the jitter when considering bad operational channels conditions. One possibility to deal with such problem is to introduce changes in the ARQ technique. So the major limitation of the M-VoIP architecture is at the Bluetooth link, when this faces very bad channels conditions, and in the Bluetooth/Wi-Fi gateway. However, the impact of such limitations on the overall system is small due to the possibility to use devices enabled with Wi-Fi technology and the retricted scenario where the Blutooth link presents problems. ACKNOWLEDGMENTS The authors would like to thank Nokia do Brasil and Instituto Nokia de Tecnologia for the partial support to develop this work. Fig. 7. Jitter and delay obtained based on the third scenario access point was very high. In such conditions the packet loss increased to very high values (around 20%). The combination of such loss with the ARQ scheme of the Bluetooth link increases the delay and the variation. This can be seen through the Figure 7 where the graphic shows high delay values. The bad link conditions lead to high packet loss rates and this, in turn, increases delay. This can be explained as follows: The ARQ scheme forces each lost packet to be retransmitted until a positive acknowledge is receive or a timeout is exceed, i.e the packet is not discarted at first increasing the delay. The stochastic characteristic of the medium causes some packets to be received with random delays causing a high jitter. Thus, when the channel conditions are not good, the retransmission mechanism of the ARQ scheme results in an increase in the delay and the jitter. This situation can be observed in Figure 7. V. CONCLUSIONS This paper discussed experimental performance analysis results related to a mobile voice over IP system, named M- VoIP. Based on the results obtained, the influence of the M- VoIP system elements was characterized. As could be seen, the impact of the Bluetooth link layer on data communication with real-time requirements is high when the conditions of the channel are bad. Regarding the M-VoIP system elements, it was detected that the number of gateways represents small influence on jitter. REFERENCES [1] Danilo F. S. Santos, José L. do Nascimento, Olympio C. S. Filho and Angelo Perkusich, Deployment of a Wireless Hybrid and Mobile Network for VoIP Services Based on Open Source Software, Proceedings of the 7th International Forum on OpenSource Software - International Track Workshops (WSL06), Porto Alegre, Brazil, April 2006. [2] Y. Wu, T.D. Todd, and S. Shirani, SCO link sharing in bluetooth voice access networks, Journal of Parallel and Distributed Computing, Special Issue on Mobile Ad hoc Networking and Computing, 2003. [3] Amrit Prit Paul Singh Bilan, Streaming Audio Over Bluetooth ACL Links, Proceedings of the International Conference on Information Technology: Computers and Communications (ITCC03), 2003. [4] Ling-Jyh Chen, Rohit Kapoor, Kevin Lee, M. Y. Sanadidi, Mario Gerla, Audio Streaming over Bluetooth: An Adaptive ARQ Timeout Approach, Proceedings of the 24th International Conference on Distributed Computing Systems Workshops (ICDCSW04), 2004. [5] RTP: A Transport Protocol for Real-Time Applications, IETF Std., July 2003. [6] J. A. Cadzow, Foundations of digital signal processing and data analysis. Macmillan, 1987. [7] G.114 Recommendation, One-Way Transmission Time, ITU-T Std., June 1996. [8] (2004, October) Bluetooth. [Online]. Available: http://www.teleco.com.br/tutoriais/tutorialblue/defult.asp [9] J.C. Haartsen, The Bluetooth Radio System, IEEE Personal Communications, February 2000. [10] IEEE 802.11. [Online]. Available: http://grouper.ieee.org/groups/802/11/ [11] Specification of the Bluetooth system, SIG Std., December 1999. [12] C. E. Perkins, IP mobility support. RFC 2002, Internet Engineering Task Force, October 1996. [Online]. Available: http://www.ietf.org [13] (2005) Asterisk: The Open Source PBX. [Online]. Available: http://www.asterisk.org/ [14] (2005) Digium: Telephony for Business. [Online]. Available: http://www.digium.com/ [15] Bluetooth PAN Working Group, Bluetooth Network Encapsulation Protocol (BNEP) Specification v. 0.95a Draft, Bluetooth Special Interest Group Std., 2005. [16] (2005) Bluez - Official Linux Bluetooth Protocol Stack. [Online]. Available: http://www.bluez.org [17] M. D. Yacoub, Foundations of Mobile Radio Energineering. CRC Press, 1993. [18] (2006, March) Ethereal: A Network Protocol Analyzer. [Online]. Available: http://www.ethereal.com SBrT 303