Wireless Personal Communications 24: 205–218, 2003. © 2003 Kluwer Academic Publishers. Printed in the Netherlands.

Performance Characterization of Wireless Multimedia through Heterogeneous Terrestrial-Satellite Network
ANDREA CONTI 1, ANDREA GIORGETTI 1 and FRANCESCO TARANTOLA 2 1 DEIS, IEIIT-BO/CNR, CNIT, University of Bologna, v. le Risorgimento 2, 40136 Bologna, Italy

E-mail: {aconti,agiorgetti}@deis.unibo.it 2 University of Pavia, via Ferrata 1, 27100 Pavia, Italy

E-mail: [email protected]

Abstract. In this work the transmission of multimedia services through heterogeneous integrated terrestrialsatellite networks is analyzed by means of measurement on a real test-bed including a video on demand (VoD) server, an IEEE802.11 wireless local area network (WLAN) integrated with a wired LAN, and a geostationary Ka-band satellite link in service at the Telecommunication Laboratory at IEIIT-BO/CNR, University of Bologna, Bologna, Italy. At first the WLAN radio link is characterized in terms of the mean received power level for different terminal locations within the Lab. Secondary, MPEG-1 video streams are transmitted with different datarate from the VoD server to many wireless stations and the traffic, for both the single source and the aggregated traffic, is characterized. An integrated terrestrial-satellite LAN is simulated as an example of utilization of source characterization. Moreover, the quality of service in terms of the average dropped MPEG frames-to-transmitted frames ratio is measured for different channel conditions, data-rate, and the number of wireless terminals requiring the service. Keywords: wireless LAN, heterogeneous integrated network, satellite, multimedia, traffic characterization.

1. Introduction In the last years the request of multimedia services has been rapidly increasing and satellite networks appear to be attractive to cover and extend the area of service. Moreover, in comparison with traditional wide area networks, a characteristic of satellite communication systems is their ability in broadcasting and multicasting flows of multimedia information with costs that do not depend on the number and the position of terrestrial terminals. In this context of integrated service packet network, video sources are expected to play a major role. This work addresses the transmission of multimedia services through heterogeneous terrestrial wired and wireless local area networks integrated with a satellite link. By means of the test-bed depicted in Figure 1, on-field measurements of the quality of service (QoS) and traffic characterizations are carried out. An IEEE802.11 WLAN [1] requires and receives video streams at different data-rates from a VoD server that is connected by a wired LAN or by a 2 Mbps satellite link through ITALSAT II, made available by CNIT.1 The satellite link provides a country-wise coverage in the single spot-beam on the Ka-band (20–30 GHz). Many
This work has been performed within the CABIS MURST project “Code Division Multiple Access for Broadband Mobile Terrestrial-Satellite Integrated System” and the CNIT/ASI project “Integration of Multimedia Services on Heterogeneous Satellite Network”. 1 Consorzio Nazionale Interuniversitario per le Telecomunicazioni, research unit of Bologna, Italy.

206 Andrea Conti et al.

Figure 1. Test-bed at the Telecommunications Laboratory in Bologna, Italy.

wireless users are potentially linked to the IEEE802.11 Access Point (see Figure 1) requiring an MPEG stream from the VoD server. The QoS, in terms of average dropped MPEG frames-to-transmitted frames ratio (DFTFR), is evaluated depending on channel conditions, the number of active stations, and the type of required service, e.g., as MPEG-1 video stream. As far as the analysis of traffic is concerned, both a single source and the aggregated traffic characterizations are carried out to exploit their self-similarity [2]. Moreover, the cumulative distribution function of the inter-arrival time is analyzed. The previously mentioned points are the main results of the paper; moreover, an example of how to use measurements to simulate an integrated terrestrial-satellite IP network is given. In this regard, source models are traditionally implemented in the form of Markov processes as they provide a flexible framework that can be customized to fit different system properties [3]. They are also attractive since the statistical behavior of the system can be directly derived from the model. Their drawback is the definition of the state diagram and of its transition probabilities to properly describe the target system. This often leads to large and complex structures with several parameters to be tuned. Moreover, it has been shown that packet traffic exhibits long range dependence (LRD) that cannot be captured by Markov chains as they can reproduce the autocorrelation function only in the initial, fast decaying steps [4]. Since LRD is related to some physical characteristic of traffic [5], several studies have been performed to develop models that are intrinsically self-similar such as those based on chaotic attractors [6]. In this work we use a characterization based on Hidden Markov Models (HMM) [7] that have been found suitable to mimic the behavior of chaotic maps [8] and preserve a mathematical tractability. As far as the QoS evaluation is concerned, we measured the average DFTFR, for varying signal-to-noise ratio of the radio link between the access point and a wireless terminal, as a function of the number of wireless stations. Since self-similarity in variable bit-rate (VBR)

Performance Characterization of Wireless Multimedia 207 traffic can cause degradation of QoS [9], an accurate characterization of the VBR MPEG traffic, is today crucial for many important applications such as QoS control, admission control policies, network resource allocation, and buffer design [10, 11]. The paper is organized as follows: in Section 2 the network configuration is described, in Section 3 the statistical characterization of the traffic generated by the VoD server is presented and an example of utilization to simulate an integrated terrestrial-satellite IP network is given. Section 4 shows the measured QoS for MPEG-1 video streams transmission.

2. Network Configuration and WLAN Coverage The architecture of the real test-bed used in this work is depicted in Figure 1. As explained before, many wireless users are potentially linked to the IEEE802.11 Access Point (AP) requiring an MPEG-1 stream from the VoD server. The wireless AP is connected to the VoD server trough a wired LAN, a router, and a 2 Mbps satellite link. A traffic monitor in the wired LAN measures the traffic. In order to statistically characterize the only traffic from the VoD server, other transmissions are not permitted. The satellite link uses ITALSAT II that provides a country-wise coverage in the single spot-beam on Ka-band (at 30 GHz for the link from earth to satellite and 20 GHz from satellite to earth). As far as the satellite coverage is concerned, following the ITU-R specifications [12, 13], and the characteristics of our equipment, we have obtained a link-budget requirement, essentially bounded from the up-link (from earth station to satellite), of: Eb · Br = 64.7 dBHz, N0

(1)

where Br is the bit-rate and Eb /N0 is the signal-to-noise ratio (energy per information bit-tonoise power spectral density ratio). The effective radiated power (ERP) at the transmitter is 51.4 dBW, and a margin on the attenuation due to the rain of 7.5 dB is taken into account for an outage probability of 1%. Experiments that will be presented in the following have been realized with good weather conditions, hence without problems related to the coverage of the satellite link. First of all a measure of the mean, over fast fading, received power has been carried out in order to evaluate the coverage at the physical layer of the WLAN. In Figure 2, a plant of the Laboratory, with the position of the AP and a measure of the mean received power level in different positions are shown. It is possible to see that all the useful area of the Laboratory appears covered with a power level sufficient to permit the maximum data-rate (2 Mbps) for the radio transfer [1]. At this point it is important to underline that for indoor environment, the signal-to-noise ratio depends also on people movement, hence we have repeated many times the measurements presented in the following and averaged the results. Starting from this consideration, we will call SNR the mean signal-to-noise ratio, where the average is over different values experimented during the tests for a fixed position of the wireless terminal in a time interval of about 22 minutes (the duration of the video transmission). Hence, this average is also over fast fading.

208 Andrea Conti et al.

Figure 2. (Top) Plant of the Telecommunications Laboratory at IEIIT-BO/CNR, Univ. of Bologna, Bologna, Italy. (Bottom) WLAN coverage: measurements of the mean received power (values in dBm).

Performance Characterization of Wireless Multimedia 209 3. Traffic Characterization In this section we present the characterization, by measurements, of both the single source traffic and the aggregated traffic for the connection between the wireless access point and the wired LAN, when many wireless terminals require the service. Moreover, the simulation of an integrated IP network is made to evaluate the inter-arrival time cumulative distribution function and the percentage of packets loss at the satellite and at the AP. 3.1. C HARACTERIZATION BY M EASUREMENTS We analyze the delivery of multimedia contents by means of the UDP/IP protocol that, at the contrary of the TCP/IP, does not require retransmission. As far as the geostationary satellite link is concerned, in which the satellite acts as a non regenerative transponder, it is known that for UDP/IP transmissions it does not affect the statistics of the traffic only adding a systematic delay over all packets (about 256 ms). Moreover, due to the available bandwidth, the bottleneck of the entire network is the WLAN segment. To analyze the traffic characteristics, it is important to consider that recent internet applications, such as web surfing, VBR video and voice over IP, generate traffic with self-similar properties that are different from conventional traffic [2, 5]. However, it has also been reported that self-similar characteristics such as long-range dependence (LRD) play a significant role on the performance and on the dimensioning of queues and buffers for packet network [14, 15]. Moreover, some related works show that self-similarity in VBR traffic can cause a degradation of quality [9]. For this reason, an accurate characterization of the VBR MPEG traffic is crucial for many important applications such as QoS control, admission control policies, network resource allocation, and buffer design [10, 11]. It is also known [4, 16] that VBR encoded video transmission such as H.261, MPEG-1, MPEG-2, generates a self-similar traffic. In this context, every source can be described by means of a self-similar model able to capture long-range dependence characteristics of the multimedia traffic. Here, we have employed a VoD system that delivers VBR MPEG-1 video content, with a duration of about 22 minutes, at different average bit-rates: 500 Kbps, 1 Mbps and 1.2 Mbps. To evaluate the arrival time of the packets in the test configuration of Figure 1, we have used the tcpdump unix command. In order to accurately characterize the traffic and avoid the loss of arrived packets that are significant for the statistics, in general attention must be paid to the accuracy of the measured arrival instants. For the MPEG-1 traffic considered in this work, the tcpdump command appears sufficiently accurate for the considered data-rate because the timing jitter of the packets arrival time is two order of magnitude less than the considered time scale. In Figure 3 the cumulative distribution function of the inter-arrival times is shown for different bit-rates. Regarding the inter-arrival times we have measured a mean value of 19.2 ms, 10.5 ms and 7.9 ms and a standard deviation of 12.2 ms, 13.1 ms and 13.8 ms for MPEG-1 streams at 500 Kbps, 1 Mbps and 1.2 Mbps, respectively. As far as the traffic characterization is concerned, the first step consists in the measure of the number of arrived bytes per time slot, with time slot of 10 ms and a measure duration of about 22 minutes. Hence, we make a vector X which element i represents the number of arrived byte in the time-slot i. The proof of the self-similarity both of the single source and the aggregated traffic is carried out by a method that is well known in literature [2] and starts from the variance-time plot, i.e., the decaying of the variance of the process (that is the packet arrivals process) versus the considered time scale.

210 Andrea Conti et al.

Figure 3. C.d.f. of the inter-arrival times.

Now, let X be a wide-sense stationary stochastic process with mean m and variance σ 2 , it is exactly second-order self-similar if, for all m = 1, 2, the following relationship is verified [17]: Var(X (m) ) = σ 2 m−β .

(2)

Generally, the self-similarity is characterized by means of the Hurst parameter that is defined as H = 1 − β/2. It characterizes the degree of self-similarity of the process, i.e., its long-range dependence. The process is called asymptotically second-order self-similar if Equation (2) is valid only for m sufficiently large. The so-called variance-time plot is obtained by plotting Log10 (Var(X (m) )) versus Log10 (m), where X (m) is the aggregate process over m time slots. In Figure 4 the variance-time plot and the corresponding best-fit line are reported in double log-scale, with m in the range from 2 to 16384 for one terminal experimenting a SNR close to 30 dB. It is possible to verify that the variance approaches a linear behavior and from its slope the Hurst parameter can be determined [5]; errors due to the arbitrary choice of the threshold for the best-fit lines that asymptotically fits the curves affect this evaluation. To proof the self-similarity we have fitted the resulting points for m greater than 32 with a simple least square line (continuous line). We have verified that for SNR greater than 10 dB (that provides the continuous radio connectivity) the Hurst parameter practically does not depend on SNR, being H = 0.98 for 1.2 Mbps and 1 Mbps streams and H = 0.97 for the remaining 500 Kbps stream. This means that, for all these three cases the traffic is strongly self-similar, so the use of a Poissonian distributed model appears not sufficiently accurate. Moreover, the average number of received packets per second is m = 126 packet/s, m = 95 packet/s and m = 52 packet/s for 1.2 Mbps, 1 Mbps and 500 Kbps, respectively.

Performance Characterization of Wireless Multimedia 211

Figure 4. Single source traffic characterization.

As far as the aggregated traffic (i.e., the total amount of streams) is concerned, it has been measured at the wireless access point by means of the same traffic monitor used for the single source. We have verified that also the aggregated traffic is strongly self-similar with Hurst parameter that in every considered cases is close to 0.97. Since the value for the single source approaches one without dependence on channel conditions (for SNR greater than 10 dB), the same occurs also for the aggregated traffic. 3.2. S IMULATION OF AN I NTEGRATED N ETWORK We use a characterization based on HMM that can be described as a 5-ple, λ = (S, V , A, B, ), where S is the set of the states, V is the set of observable values, A is the state transition probabilities matrix (the element ai,j represents the transition probability from state Si to state Sj ), B is the observable probabilities matrix (the element bi,j represents the probability of emitting symbol vj from state Si ), and  is the initial state probability vector. A larger number of states, in general, increases the ability to emulate the statistical properties of the real system, although this increase becomes marginal after a number of states depending on the system complexity. On the other side, increasing the number of states clearly leads to an increasing in terms of computational complexity, due to the processing time required for convergence of the regeneration process to the best transition probabilities. Model parametrization is commonly performed using algorithms such as the method of moments or gradient or the Baum Welch (BW) procedure, [18], used in this work. Given a set of observed samples O = O1 O2 . . . Oτ , the BW algorithm defines an iterative procedure that determines the parameters of a HMM by maximizing a likelihood function Pr(O). It is a very robust algorithm in that it always converges, but there is no guarantee that the convergency point

212 Andrea Conti et al.

Figure 5. Complementary cumulative distribution of the inter-arrival time for a single user.

is a global maximum: the final parameters may consequently be not the optimal ones in an absolute sense. Some initial study has been performed on the HMM to select a proper number of states. Another key parameter is the matrix of symbols B as it greatly affects optimization time. After these experiments, the best compromise has been found and a system with 10 states and 50 bins as possible bit rates. We will show the simulation results related to the transmission of MPEG-2 packets (max length of 1024 bytes) at 590 Kbps with Hurst parameter H = 0.81 from a VoD server through a multiplexer (queue of 25 packets) that combines more video sources, the satellite network, and the AP modelled as a queue of 25 packets. The UDP-RTP packets from our video server are sent to the AP of the wireless LAN imposing a limitation to 2 Mbps on the satellite link and 1.2 Mbps at the AP. In Figures 5 and 6 the complementary cumulative distribution of the inter-arrival time is shown when one or three users, respectively, require the service. The results for the real MPEG-2 sequence and that one generated by means of HMM are plotted. It is possible to note that HMM is sufficiently accurate for this kind of simulations. Starting from sources characterization it is possible to evaluate the network performance. As an example, the percentage of lost packets at the satellite transmitted and at the AP are reported in Table 1 when one, two or three users, respectively, require the service (total number of transmitted packets equal to 110812).

Performance Characterization of Wireless Multimedia 213

Figure 6. Complementary cumulative distribution of the inter-arrival time for three independent users. Table 1. Percentage of lost-packets.

Lost at SAT (%) Lost at AP (%) Total received (%)

One user

Two users

Three users

0 10−3 97

10−2 10−1 83

8 · 10−2 2, 2 · 10−1 65

4. Quality of Service Measurements As a figure of the video QoS we consider the average DFTFR, i.e., the average ratio (over different repetitions on a fixed position of the wireless terminals) between the number of MPEG-1 frames dropped during the transmission of the video stream and the total number of transmitted frames. We consider the real test-bed of Figure 1 and, to measure the average DFTFR we have considered the wireless terminals in different spatial positions with a fixed SNR. Moreover, for each position we have repeated the video stream transmission in order to obtain the QoS averaged over different propagation conditions. In Figure 7, the QoS measurements over the real test-bed discussed before, are shown as a function of the SNR (dB) for the three classes of the considered video streams when a wireless terminal requires the service. The drops are due to the channel conditions for SNR values less than or about 10 dB, whereas they are principally due to the limitations on radio bandwidth (2 Mbps) when the channel is sufficiently good, that means for SNR greater than 15–20 dB. The same considerations could be made when two and three terminals require the service, as considered in Figures 8 and

214 Andrea Conti et al.

Figure 7. QoS vs. SNR for one terminal at different data-rates.

9, respectively. Here the limitations on the available bandwidth per terminal have a stronger effect on the average DFTFR. In Figure 10 a transmission at 500 Kbps is considered, and the effects on the QoS of the channel impairment and the number of stations are reported. In this figure it is possible to observe the strongly decaying of the perceived QoS by the user due to the lower signal-to-noise values and the increasing number of terminals that fills up the available bandwidth. 5. Conclusions In this paper wireless transmission of multimedia services through heterogeneous wireless and wired integrated terrestrial-satellite networks has been analyzed. By means of a real testbed, both a single source and an aggregate traffic characterization, for MPEG-1 video streams transmission, has been carried out by means of measurements. The traffic appears self-similar with an Hurst parameter that does not strongly depend on the channel conditions (e.g., signalto-noise ratio). This characterization can be used to simulate an extended network; at this regard the simulation of an integrated terrestrial-satellite IP network is given. Moreover, the average dropped MPEG frames-to-transmitted frames ratio, as figure for the quality of service, has been carried out by means of measurements for a different number of wireless terminals requiring the service. It results strongly dependent on channel conditions and on the requested bandwidth with respect to the available bandwidth.

Performance Characterization of Wireless Multimedia 215

Figure 8. QoS vs. SNR for two terminals at different data-rates.

Figure 9. QoS vs. SNR for three terminals at different data-rates.

216 Andrea Conti et al.

Figure 10. QoS vs. Number of terminals for 500 Kbps and 1.2 Mbps MPEG-1 streams at different SNR.

Acknowledgements The authors would like to thank Professors Oreste Andrisano, Marco Chiani, Eugenio Costamagna, Lorenzo Favalli and Dr. Davide Dardari for helpful suggestions. This work has been performed within the CABIS MURST project “Code Division Multiple Access for Broadband Mobile Terrestrial-Satellite Integrated System” and the CNIT/ASI project “Integration of Multimedia Services on Heterogeneous Satellite Network”. References 1. 2. 3. 4. 5.

6.

7. 8.

IEEE 802.11 Standard for WLAN MAC and PHY Specifications, IEEE Std 820.11-1997. W.E. Leland, M.S. Taqqu, W. Willinger and D.V. Wilson,“On the Self-Similar Nature of Ethernet Traffic (extended version)”, IEEE/ACM Trans. on Networking, Vol. 2, No. 1, pp. 1–15, Jan. 1994. D. Heyman and T. Lakshman, “Source Models for VBR Broadcast-Video Traffic”, IEEE/ACM Trans. on Networking, Vol. 4, No. 1, pp. 40–48, Feb. 1996. J. Beran, R. Sherman, M.S. Taqqu and W. Willinger, “Long-Range Dependence in Variable-Bit-Rate Video Traffic”, IEEE Trans. on Communications, Vol. 43, Nos. 2–4, pp. 1566–1578, 1995. M.S. Taqqu, W. Willinger, D.V. Wilson and R. Sherman, “Self-Similarity through High-Variability: Statistical Analysis of Ethernet LAN Traffic at the Source Level”, IEEE/ACM Trans. on Networking, Vol. 5, No. 1, pp. 71–86, Feb. 1997. E. Costamagna, L. Favalli, P. Gamba and G. Iacovoni, “A Simple Model for VBR Video Traffic Based on Chaotic Maps: Validation through Evaluation of ATM Multiplexer QoS Parameters”, in Proc. IEEE Int. Conf. on Commun. 1998, Atlanta (GA), June 1998, pp. 568–572. L.R Rabiner and B.H. Juang, “An Introduction to Hidden Markov Models”, IEEE ASSP Mag., Vol. 3, pp. 416, Jan. 1986. D.H. Kil and F.B. Shin, Pattern Recognition and Prediction with Application to Signal Characterization, American Institute of Physics Press: Woodbury, New York, 1996.

Performance Characterization of Wireless Multimedia 217 9. 10. 11. 12. 13. 14.

15. 16. 17. 18.

T. Kato and Y. Ji and S. Asano, “A Study on the Self-Similarity of MPEG2 Video Traffic and the Effect for Transmission Quality”, in Proc. of 10th International Packet Video Workshop, Italy, 2000. M. Krunz and S. Tripathi, “Exploiting the Temporal Structure of MPEG Video for the Reduction of Bandwith Requirements”, in IEEE Proc. of INFOCOM ’97, Italy, 1997, pp. 67–74. M. Krunz,“Bandwidth Allocation Strategies for Transporting Variable-bit-Rate Video Traffic”, IEEE Comm. Magazine, Vol. 37, No. 1, pp. 40–46, Jan. 1999. ITU, Reccomendation ITU-R P.676-3, ITU. ITU, Reccomendation ITU-R P.838, ITU. O. Andrisano, D. Dardari and G. Mazzini, “Wireless Multimedia Assessment with Traffic, Access and Transmission Protocols in Actual Environment at Millimeter Waves”, in Proc. Of IEEE Multimedia Comm. Workshop 1998, Italy, September 1998. M.K. Shahin, A. Giovanardi and G. Mazzini, “Simulation of CSMA WLAN Systems with Hidden Terminal and Advanced Capture Effects”, in ICACC’99, Athens, Greece, July 1999. D. Heyman and T. Lakshman, “’What Are the Implications of Long-range Dependence of VBR Video Traffic in ATM Traffic Engineering?”, IEEE/ACM Trans. on Networking, Vol. 4, No. 3, pp. 301–317, June 1996. B. Tsybakov and N.D. Georganas, “Self-Similar Processes in Communications Networks”, IEEE Trans. on Information Theory, Vol. 44, No. 5, pp. 1713–1725, Sept. 1998. G. Lindgren and U. Hoist, “Recursive Estimation of Parameters in Markov-Modulated Poisson Processes”, IEEE Trans. on Communications, Vol. 43, No. 11, pp. 2812–2819, Nov. 1995.

Andrea Conti was born in Bologna, Italy, on December 20, 1972. He received the Dr.Ing. degree (with honors) in telecommunications engineering and the Ph.D. degree in electronic engineering and computer science, both from the University of Bologna, Bologna, Italy, in 1997 and 2001, respectively. In 1999, he joined the Consorzio Nazionale Interuniversitario per le Telecomunicazioni (CNIT) at the Research Unit of Bologna, working in the design of a DSP-based CDMA satellite modem within the ASI (Italian Space Agency)-CNIT project “Integration of Multimedia Services on Heterogeneous Satellite Networks”. In the Summer of 2001, he joined the Wireless Section of AT&T Labs-Research, NJ, U.S.A., working on the performance of digital telecommunications systems with diversity reception. His research interests include mobile radio resource management, frequency hopping, coding in faded MIMO channels, nonlinear effects in CDMA, and digital signal processing.

218 Andrea Conti et al.

Andrea Giorgetti was born in Cesena, Italy, in 1974. He received the Laurea degree in electronic engineering (with honors) from the University of Bologna, Italy, in July 1999. In 2000, he joined the Department of Electronics, Computer Science and Systems of the University of Bologna, and he is currently working toward his Ph.D. His research interests are concerned with wireless networks, traffic modeling, and performance evaluation of digital communication systems. He is a student member of IEEE.

Francesco Tarantola was born in Broni on September 23, 1976. In March 2001 he received the Laurea degree in electronics engineering at the University of Pavia. Since November 2001 he is attending a Ph.D. course in electronics and computer science engineering at the University of Pavia. Currently his fields of interest include channel and traffic modeling with chaotic and Markov models, video compression (MPEG 2 and 4, H263 and H26L), high-speed communications and wireless networks. He is a member of IEEE.