Experimental Evaluation of the Impact of Network Frequency Synchronization on GSM Quality of Service During Handover Stefano Bregni*, Senior Member, IEEE, Lucia Barbieri** * Politecnico di Milano, Dept. of Electronics and Information, P.zza L. Da Vinci, Milano, ITALY Tel.: +--. Fax: +--. E-mail: bregni@elet.polimi.it ** Omnitel Vodafone - Network Quality Headquarters, Torre Professional Via Mozzoni, Milano, ITALY Tel.: +--. E-mail: lucia.barbieri@omnitelvodafone.it Abstract Poor accuracy of frequency references used in GSM Base Transceiver Stations () can lead to dropped calls, slow handover between cells and even co-channel interference. In this paper, we report some results of an experimental trial carried out in the Omnitel-Vodafone test plant in Milano (Italy). In authors' knowledge, this is the first paper confirming with experimental data that lack of synchronization affects GSM handover performance, leading to some degradation of quality of service. The speech quality of calls between two GSM handsets, undergoing handover between two s synchronized with variable frequency offset, was measured in terms of Mean Opinion Score () and. For values of fractional frequency offset f/f n = and -, all completed calls featured optimal quality in spite of handover. On the contrary, for f/f n = - and -, performance degradation was experienced: about % of completed calls exhibited unpleasant values and high speech-clipping percentages. Such experimental results confirm the importance of GSM network synchronization and warn GSM operators about leaving clocks to operate in free-run: frequency accuracy in the order of - to - is typical for quartz oscillators commonly used in this equipment. Index Terms Clocks, GSM, handover, mobile communication, synchronization. N I. INTRODUCTION etwork synchronization is any distribution of time and frequency over a network of clocks, even spread over a wide area [] []. The goal is to align the time and frequency scales of all clocks, by using the communications capacity of links among them. A synchronization network is the facility implementing network synchronization: its basic elements are nodes (autonomous and slave clocks) and links (e.g., Mbit/s transmission facilities). Network synchronization has gained increasing importance in telecommunications throughout the last thirty years, especially since transmission and switching turned digital []. Actually, the quality of most services offered by network operators to their customers is affected by network synchronization performance. Since the Seventies, digital switching equipment has been requiring synchronization to avoid slips at input elastic stores. Ten or twenty years later, network synchronization became a thorny matter for telecommunications operators with the deployment of SDH (Synchronous Digital Hierarchy)/SONET networks, which pose new and more complex requirements on the stability of synchronization systems. More recently, it has been recognized the importance of network synchronization also for cellular mobile telephone networks, including those according to GSM (Global System for Mobility), GPRS (Global Packet Radio Services) and UMTS (Universal Mobile Telecommunications Services) standards. Therefore, Omnitel-Vodafone and other mobile telephone operators have decided the deployment of network synchronization facilities, in order to deliver advanced digital services with superior dependability and quality. In particular, poor accuracy and stability of frequency references used in Base Transceiver Stations () can lead to dropped calls, slow handover between cells and even cochannel interference. Rather surprisingly, although it is widely claimed that the availability of good frequency references in GSM networks is beneficial under several aspects, in literature it is very difficult to find reports of experimental trials aiming at evaluating the impact on quality of service of poor GSM network synchronization. Among the very few documents available on this issue, a major supplier of network synchronization systems [] reported that, according to a trial conducted in a not specified GSM network, the deployment of Global Positioning System (GPS) timing systems and associated precision clocks had a dramatic impact on handover success rates. In this trial, the inter-cell handover failure rate was reported to be reduced from about % to less than % right after network synchronization facilities were deployed. In this paper, we report some results of an experimental trial carried out in the Omnitel-Vodafone test plant located at company headquarters in Milano, Italy. The aim of the trial was at investigating experimentally the impact of frequency offset between two GSM on the quality of service, as perceived by the user, when handover is taking place. In authors' knowledge, this is the first paper confirming with experimental data that lack of synchronization affects GSM handover performance, leading to some degradation of quality of service. In particular, in Sec. II, the basics on synchronization of cellular mobile telephone networks are highlighted, by summarizing their synchronization standard requirements and technical solutions. Then, in Sec. III, the experimental set-up used in this trial and the procedure followed are described; the call quality measures adopted are also defined. Finally, Sec.
IV reports the most significant measurement results and Sec. V draws some conclusions stemming from the experiments carried out in this work. II. SYNCHRONIZATION OF CELLULAR MOBILE TELEPHONE NETWORKS The generic architecture of a cellular mobile telephone network is shown in Fig.. The cell site equipment () allows routing of calls from the to the network infrastructure, which comprises Base Station Controllers (BSC), Mobile Switching Centers (MSC) and then on to the Public Switched Telephone Network (PSTN). Current standards include the European GSM/DCS (/ MHz) [] and the North American cellular systems (/ MHz)., BSC and MSC nodes are interconnected via (e.g. on E/T, PDH or SDH systems). to PSTN MSC BSC trunk lines Fig. : System architecture of a cellular mobile telephone network. As with any other network element, s, BSCs and MSCs need to be synchronized to ensure slip-free interconnection at. Moreover, should be synchronized to ensure frequency stability on the on-air wireless channels. ETSI specifies a fractional frequency accuracy f max /f n better than - (i.e.,. ppm) for GSM systems []. In North America, for Time-Division-Multiple-Access (TDMA) and Code-Division-Multiple-Access (CDMA) systems, frequency accuracy is specified better than - and -, respectively. In addition to this frequency synchronization, CDMA systems require also precise time synchronization, while GSM and North-American TDMA do not. The fractional frequency accuracy is defined as the maximum frequency error f max compared to the nominal value f n, measured over a very long time interval (e.g., the equipment life time). It is usually expressed by the adimensional ratio f max /f n and is often measured in - units [µhz/hz], in the engineering practice called also [parts per million] abbreviated as [ppm]. Note that [ppm] is not an International System (SI) unit, and therefore it is not used by time and frequency metrologists. Finally, newer services such as locating GSM mobile handsets, GPRS and third-generation wireless systems (UMTS) require even more stringent frequency and time stability, although not yet specified in detail. Owing to both requirements, CDMA systems mostly use GPS receivers to time synchronize their elements, as well as to produce a stable frequency reference for on-air signals. On the other hand, GSM and TDMA carriers require stable frequencies but do not require precise time synchronization. In order to fulfil such frequency requirements, different technical solutions have been adopted, which include the deployment in each BSC of precision autonomous clocks (compliant with ITU-T Recs. G. [] or G./type-II [] specifications) or of slave Stand-Alone Synchronization Equipment (SASE) clocks (compliant with G. []), synchronized by a synchronization network. What's more, among, BSC and MSC nodes are usually E/T circuits, transported over PDH/SDH systems, but they may migrate to some other transport technology that does not feature a good timing transparency, such as Asynchronous Transfer Mode (ATM) virtual connections, radio links, etc. In this case, the availability of good external synchronization facilities will be even more needed. In GSM, handover is the procedure initiated by the network to force the user handset (Mobile Station, MS) to change time slot and frequency (Traffic CHannel, TCH) from one cell to a neighboring one, based on several quality parameters measured continuously by and MS. This procedure, once initiated, must be completed in a very short time (about ms) and appears to be quite sensitive to stability and accuracy of frequency references used in s of involved cells. During handover, the broadcasts a series of instructions to the MS. Under timing impairments, the may be forced to send several corrective instructions, adjusting repeatedly frequency and time slot parameters. Hence, inaccuracies in the frequency references used in s can lead to dropped calls and slower handover between cells. In the next sections, we report some results of an experimental trial carried out in the Omnitel-Vodafone test plant (Milano, Italy), with the aim at investigating the practical impact of frequency offset between two GSM s on the quality of service when handover is taking place. III. EXPERIMENTAL SET-UP The experimental set-up used in this trial is depicted in Fig.. Two s (of different types A and B), supplied by a major GSM equipment manufacturer and widely deployed in the Omnitel-Vodafone network, were synchronized by signals generated by a frequency synthesizer and a BSC, no other way involved in the test. The frequency offset f between the two reference signals was varied by setting the synthesizer and was also measured by a digital time counter used as frequency meter. Two radio cells, controlled by A and B, were thus covering the plant site, where two handsets were able to
connect to either stations, depending on transmit power. For each frequency offset, a few test calls were established between the pair of handsets, transmitting a pre-recorded speech sample in each call. At the beginning of each call, both handsets were under A. Then, handover of both handsets to B was forced by attenuating sharply the transmit power from A (step power variation). frequency synthesizer BSC. MHz. Mbit/s f meter A B cell A cell B Fig. : Experimental set-up to measure quality of calls between two GSM handsets undergoing handover between two base stations synchronized with given frequency offset. Handset management and quality of service measurement were performed by means of two PCs running a specific software (NetQual by SwissQual []). To characterize speech quality of calls, the following parameters were measured in this trial. Mean Opinion Score (), a measure of listening quality with scores ranging from to (namely, : bad; : poor; : fair; : good; : excellent), according to the scale defined in ITU-T Rec. P. []. The, therefore, should be intended as the average of many individual opinions on speech quality. In our results, moreover, we defined = to indicate dropped call. Speech clipping (or front-end clipping), to indicate the loss of speech segments. It may occur, for example, when voice activity detection is used or during uncontrolled slips. In the former case, clipping is the phenomenon that cuts off a short segment of speech in the time the transmitter takes to detect presence of speech. NetQual measures clipping by segmenting the speech in -ms frames and then counting clipped frames, by comparison of the received signal to the reference sample. Clipping values are reported as average percentages (i.e., the number of clipped frames over the number of active speech frames) per speech sample (the call length in our experiment). Speech quality is a major factor determining customer satisfaction. Although large-scale auditory tests should be carried out to assess it, by definition, it is most commonly preferred to use instruments capable of measuring some speech quality parameters and producing results that correlate as closely as possible with subjectively acquired results. IV. MEASUREMENT RESULTS By tuning the synthesizer, four values of frequency offset between A and B were set: f/f n = ppm, f/f n =. ppm, f/f n = ppm and f/f n = ppm. For each value of frequency offset, a series of calls of length s were established between the pair of handsets, both connected initially to A, transmitting a pre-recorded test speech sample in every call. During each call, handover of both MSs to B was forced as explained in the previous section. Moreover, and were being measured and recorded for each call. The histograms in Fig. show the distribution of the values measured in the four series of test calls. First, we note that, given the limited number of test calls, it is not possible to recognize any significant variation of the number of dropped calls (those reported with =). Moreover, for f/f n = ppm and f/f n =. ppm, we note that all completed calls featured optimal quality, with in the range. to., which are typical values for GSM Full-Rate communication in ideal conditions. On the other hand, for f/f n = ppm and f/f n = ppm, it is clearly noticeable a performance degradation, probably due to slower MS handover: about % of completed calls exhibited unpleasant values around. f/f n= ppm,,,,,,,,,,,,,,,,,,,,,,,,,, f/f n=. ppm,,,,,,,,,,,,,,,,,,,,,,,,,, f/f n= ppm,,,,,,,,,,,,,,,,,,,,,,,,,, f/f n= ppm,,,,,,,,,,,,,,,,,,,,,,,,,, Fig. : Distribution of the values measured in series of test calls undergoing handover between A and B, synchronized with frequency offset f/f n = ppm,. ppm, ppm and ppm.
% % % % % % % % f/f n= ppm % % % % % % % % % % % % % min max mean f/f n=. ppm % % % % % % % % % % % % % % % % % % % % % f/f n= ppm % % % % % % % % % % % % % % % % % % % % % f/f n= ppm % % % % % % % % % % % % % % % % % % % % % Fig. : Distribution of the speech-clipping percentages measured in series of test calls undergoing handover between A and B, synchronized with frequency offset f/f n = ppm,. ppm, ppm and ppm (excluding dropped calls). The histograms in Fig., on the other hand, show the distribution of the speech-clipping percentages measured in the same four series of test calls, excluding the dropped ones. Similarly to results, we note that for f/f n = ppm and f/f n =. ppm all completed calls featured best quality, with speech-clipping percentages not above few points per cent. On the contrary, for f/f n = ppm and f/f n = ppm, slower MS handover caused to raise: % of completed calls, those with low values, exhibited substantially higher speech-clipping percentages, most in the range % to %. Finally, to further highlight the performance degradation experienced with increasing frequency offset between A and B, the bar graph in Fig. plots minimum, maximum and mean of the values measured in the same four series of test calls, excluding the dropped ones. While the maximum measured in calls keeps close to for all frequency offsets, the minimum and mean values decrease significantly for f/f n = ppm and f/f n = ppm. Similarly, the graph in Fig. plots minimum, maximum and mean of the speech-clipping percentages. Analogous considerations as done for Fig. can be made also for this graph. ppm. ppm ppm ppm f/f n Fig. : Minimum, maximum and mean of the values measured in series of test calls undergoing handover between A and B, synchronized with frequency offset f/f n = ppm,. ppm, ppm and ppm (excluding dropped calls). % % % % % ppm. ppm ppm ppm f/f n min max mean Fig. : Minimum, maximum and mean of the speech-clipping percentages measured in series of test calls undergoing handover between A and B, synchronized with frequency offset f/f n = ppm,. ppm, ppm and ppm (excluding dropped calls). V. CONCLUSIONS In this paper, we reported some results of an experimental trial carried out in the Omnitel-Vodafone test plant located at company headquarters in Milano, Italy. In authors' knowledge, this is the first paper confirming with experimental data that lack of synchronization affects GSM handover performance, leading to degradation of quality of service. The speech quality of series of calls between two GSM handsets, undergoing handover between two s synchronized with frequency offset f/f n = ppm, f/f n =. ppm, f/f n = ppm and f/f n = ppm, was measured in terms of and. Due to the limited number of test calls, it was not possible to recognize any significant variation of the number of dropped calls with the frequency offset. Nevertheless, all completed calls undergoing handover with f/f n = ppm and f/f n =. ppm featured optimal quality, with in the range. to. and speech-clipping percentages not above few points per cent. On the contrary, for f/f n = ppm and f/f n = ppm, it was clearly noticeable a performance degradation, probably due to slower MS handover: about % of completed calls exhibited
unpleasant values around and high speech-clipping percentages, most in the range % to %. In conclusion, such experimental results confirm that strict control on frequency references should be always enforced in GSM network synchronization. On the one hand, the - frequency accuracy requirement specified by ETSI [] for GSM systems can be met only by deploying precision clocks, such as those compliant with ITU-T Recs. G. [] or G. [] specifications. On the other, the experimental findings reported in this paper warn GSM network operators about leaving clocks to operate in free-run: frequency accuracy in the order of - - is typical for quartz oscillators as those commonly used in this equipment. Actually, even setting up synchronization chains made of a few s, in most cases equipped with poor-stability clocks, may lead to handover performance degradation. ACKNOWLEDGEMENTS The Authors wish to thank warmly Marco Merlonghi and Marco Costa, formerly with Omnitel-Vodafone, for their friendly support and cooperation during the experimental work. Moreover, a special note of thanks is due also to Michele Ruzzi, responsible of Network Quality Dept. at Omnitel-Vodafone headquarters. REFERENCES [] W. C. Lindsey, F. Ghazvinian, W. C. Hagmann, K. Dessouky, "Network Synchronization", Proceedings of the IEEE, vol., no., Oct., pp. -. [] P. Kartaschoff, "Synchronization in Digital Communications Networks", Proceedings of the IEEE, vol., no., July, pp. -. [] S. Bregni, Synchronization of Digital Telecommunications Networks. Chichester, UK: John Wiley & Sons,. [] S. Bregni, "A Historical Perspective on Network Synchronization", IEEE Communications Magazine, vol., no., June. [] Symmetricom, Application Note "Mobile Switch Center (MSC) and Base Station Controller (BSC) Synchronization", January. Available at URL: http://www.symmetricom.com. [] P. Mouley, M. B. Pautet, The GSM System for Mobile Communications. Lassay-les-Château: Europa Media Duplication S. A.,. [] ETSI EN Digital Cellular Telecommunication System (Phase +); Radio Subsystem Synchronization. [] ITU-T Rec. G. Timing Characteristics of Primary Reference Clocks, Geneva, Sept.. [] ITU-T Rec. G. Timing Requirements of Slave Clocks Suitable for Use as Node Clocks in Synchronization Networks, Geneva, June. [] SwissQual Ltd., NetQual User Manual, ver..,. [] ITU-T Rec. P. Methods for Subjective Determination of Transmission Quality, Geneva, Aug..