You are on page 1of 5

OPTIMIZING THE GSM/EDGE AIR INTERFACE FOR MULTIPLE SERVICES WITH DYNAMIC FREQUENCY AND CHANNEL ASSIGNMENT

Matti Salmenkaita, Jose Gimenez and Pablo Tapia
Nokia Networks Malaga, Spain matti.salmenkaita@nokia.com

Mariano Fernandez-Navarro
Departamento de Ingeniería de Comunicaciones E.T.S.I. Telecomunicación. Univ. of Málaga Málaga, Spain Mariano@ic.uma.es combined with the diverse CIR requirements of different services in the air interface calls for a radio resource management solution that can flexibly adapt to the mix of services so that the spectral efficiency is always maximized. Dynamic channel allocation (DCA) methods can be used to achieve these goals. The DCA schemes can be categorized as traffic adaptive (TA-DCA) or interference adaptive (IADCA) [1]. The traffic adaptive DCA schemes tend to be geared towards minimizing the call blocking probability and they don't incorporate real time measurement based interference control. Examples of TA-DCA methods can be found in [2,3,4]. The interference adaptive DCA algorithms utilize real time interference measurements and are therefore best suited to maximize the performance in a mixed services environment. Examples of IA-DCA methods are presented in [4,6,7]. This paper presents the application of a new interference adaptive DCA method for GSM networks called Dynamic Frequency and Channel Assignment (DFCA) in mixed services environments. Previous studies of DFCA in mainly single service environments can be found in [8,9]. The DFCA functionality is described in the next section. This is followed by a description of the simulation environment and then the simulation results obtained by network level simulations are presented. Finally some conclusions are drawn. II. DYNAMIC FREQUENCY AND CHANNEL ASSIGNMENT DFCA is a channel assignment scheme that individually selects the used radio channel for each connection based on the prediction of the achievable carrier-to-interference ratio (CIR) in each channel candidate. Therefore, the used radio channel can be chosen depending on the CIR requirements of the used channel mode. This makes it possible to provide less robust channel modes such as EFR with sufficiently high CIR while taking full advantage of the more robust channel modes such as AMR FR by allowing the usage of radio channels with lower CIR. As a result, the differences in the robustness of the used channel modes can be hidden from the end user.

Abstract— The variety of services offered in cellular networks is rapidly increasing. However, optimizing the GSM/EDGE air interface for a multitude of services remains a big challenge. The Dynamic Frequency and Channel Assignment (DFCA) provides powerful tools to manage the highly variable interference tolerance properties of different services in the air interface. In this paper the application of the DFCA concept in GERAN network with a mix of services is studied. The technical implementation of DFCA with multiple service support is also presented. The concepts are validated with extensive simulations. Simulations involve multiple scenarios with varying service mixes including both circuit switched and packet switched services. The results show significant capacity gains in both single service and multiple service scenarios. Keywords— AMR, DCA, DFCA, EDGE, EGPRS, GERAN, GSM, Radio resource management, Synhronization

I. INTRODUCTION Since their introduction, the cellular networks have experienced rapid traffic growth. In many cases the bandwidth available to operators has not increased at a similar pace forcing the operators to deploy solutions to enhance the spectral efficiency of their networks. Such solutions include micro cell layers, underlay-overlay and frequency hopping schemes. Until recently, the services provided in the GSM networks have had quite uniform requirements for the carrier-tointerference ratio (CIR) in order to deliver acceptable quality of service. However, the recent introductions of the adaptive multiple rate (AMR) speech coding and the 8-PSK modulated EGPRS data service bring along a diversification of the radio channel interference tolerance. The low bit rate AMR speech codecs are very robust against interference tolerating as low as 6dB CIR while still providing reasonable speech quality whereas an EGPRS channel reaches the maximum throughput at a CIR of about 30dB. The demand for ever higher spectral efficiency is expected to continue as the wireless data services gain popularity. This

CIR estimation The estimation of CIR is based on the downlink received signal level measurements of the serving cell and the neighbouring cells. The CIR is then compared to the CIR target of the interfered connection. During DFCA operation the occupied (E)GPRS time slots are taken into consideration whenever CIR estimations are calculated. but the CIR is only realized when the same MA list and MAIO are being used in a neighbouring cell in the same time slot. This common radio channel is pre-assigned to each transceiver unit maximizing the re-use distance. . These measurements are continuously performed and reported to the network by the mobile stations (MS) in active mode. Adjacent channel interference is taken into account by applying a fixed 18dB CIR offset. Whenever a new (E)GPRS time slot is taken into use the system examines if there are any circuit switched connections that would either cause interference or be interfered. The optimum DFCA operation requires that a base station will have to be able to use any DFCA MA list and MAIO in any of the time slots. The frequency band dedicated to DFCA operation is divided into one or more mobile allocation (MA) lists. In such case the conflicting circuit switched connection is moved to another radio channel. the chosen approach keeps the (E)GPRS traffic in consecutive time slots that all use a common radio channel within each transceiver unit. Furthermore.1 can be formed. The method described above provides means to determine a potential co-channel or adjacent channel CIR towards each of the reported neighbouring cells. MAIO and time slot combinations as shown in Fig. For the neighbouring cells the broadcast control channel frequencies are measured. Offset in the frame number means similar offset (modulo MA length) in the MAIO of the interfering connection. The CIR statistics can also be used to determine an estimate of the uplink CIR. Instead.The DFCA can also be applied to the packet switched (E)GPRS traffic. the transmission power reduction applied in the interfering connection must also be considered in order to obtain an accurate CIR estimate. The same principle also works the other way around. Often there are more than six such cells. DFCA frequency hopping The air interface synchronisation makes it possible to control the frame numbers in each of the cells. The CIR estimate towards such unreported cells can be obtained by collecting long term neighbouring cell RXLEV measurement statistics from all the mobile stations in a cell. From these statistics a CIR with a given outage probability can be determined for all the neighbouring cells. The main functionality of the DFCA system is listed below. This makes it possible to take all the interference considerations to the time slot level. Synchronization DFCA is based on time slot alignment provided by network level synchronisation. in case of the best effort traffic the connection specific independent radio channel selection is not feasible as shared channels controlled by the packet scheduling function can provide superior performance. This is a standard GSM functionality primarily used to provide path loss information for handover decisions. This allows utilization of cyclic frequency hopping while still keeping the interference relations fixed in the network. B. A frequency in the non-hopping case corresponds to a combination of an MA list and a mobile allocation index offset (MAIO) in the cyclic frequency hopping case. This means that the base station will have to support time slot level synthesized frequency hopping and wide band combining. This way the interference from the circuit switched connections affecting the (E)GPRS time slots can be controlled. The CIR difference indicates the predicted channel CIR compared to the CIR target and it is calculated as follows: CIRdiff = CIR − CIRtarget For each channel the CIR difference of the new connection as well as the CIR difference of the interfered ongoing connections is considered. Two connections using the same MA list and MAIO are co-channel interferers to each other if the cells have the same frame number. By processing this information a matrix showing the minimum CIR difference for all the possible MA list. In most cases this is sufficient for handover evaluation purposes but for interference estimation the potential CIR has to be determined for all the surrounding cells that may be significant interference sources. The time slot alignment ensures that the GSM air interface time slots are co-incident throughout the network. MAIOs and transmission power reductions used by the active connections in the potentially interfering cells. The potential CIR towards each of the reported neighbouring cells can then be calculated as follows: CIR n = RXLEV servcell + Prservcell − RXLEV neigbcelln where RXLEV is the GSM parameter corresponding to the measured signal level expressed in dBm and Pr is the current power control power reduction in the serving cell. In order to determine the real CIR the network must examine the time slots. Out of all these the minimum CIR difference is then determined. MA lists. This way the network is able to have a CIR estimate for all the neighbouring cells: a more accurate measured one for the directly reported cells and a statistical one for the rest of the neighbours. However. C. In GSM the RXLEV values of a maximum of six neighbouring cells can be reported. A.

SIMULATION ENVIRONMENT The simulation results presented in this study have been obtained using a dynamic GERAN network simulator called SMART. Discontinuous transmission and power control can be optionally applied. DTX) against a DFCA network. and the rest for AMR. in order to test the behaviour of the channel allocation mechanism under different conditions. Example CIR matrix D.5 (BCCH blocked) 1500 m 3 km/h 12 RF-FH. which was decided to be an acceptable quality for speech. The graphs include the performance of EFR and AMR-FR (also HR codec in the case of DFCA). This simulator models a wireless network with 75 cells in which the mobiles appear following a random process and move along the scenario with a certain trajectory. the gain figures presented here will be increased under irregular scenario conditions. The call length is variable. a 70% capacity gain can be achieved in the case of EFR. III.. SIMULATION RESULTS Several simulation cases have been performed. These curves show that. 5. 0 2. decr.90-FR. decr. The 2 curves on the lower load side are for EFR. MAIO 1. and to 30% with the half rate one. 1 2. MAIN SIMULATION PARAMETERS Simulated traffic Speech Mean call length Number of terminals EFR. The channel ranking based on the CIR difference also preserves the relative differences in the end user CIR between user of different channel modes having different CIR targets. Thus. Table 1 summarizes the most important parameters. a network with more than 2% of bad quality samples is considered unacceptable. PC. 4) . decr. PC ON EFR: (inc. following a negative exponential distribution with mean value of 120s. The bad quality samples are those that present a frame erasure rate (FER) higher than 4. (E)GPRS traffic can also be simulated. 3. Channel selection The channel with the highest CIR difference is preferred in the channel selection. 2) AMR-FR: (inc. This target is GENERAL CONFIGURATION Number of cells Number of traffic TRXs/cell Intersite distance Mobile speed Number of frequencies Other RXQUAL PC Thresholds (RXLEV disabled) 75 3. Finally. The performance is presented in terms of percentage of bad quality samples (BQS) versus effective frequency load (EFL). 2) AMR-HR: (inc. AMR5. The simulations can be performed using different kind of services and different type of traffic at the same time in order to study the interaction between them. FTP (EGPRS) 120s (Exponentially distributed) Up to 2% blocking rate IV. DTX.9-HR. 3. as DFCA dynamically adapts the interference situation in the system while the reference network relies on random 1/1 frequency hopping that randomizes the interference but cannot automatically adapt to the irregularities. 1 1. 2 presents the performance of a reference network (typical GSM 1/1 random frequency hopping. as there is also a model of RLC/MAC protocols for packet switched traffic. Note that in order to have an optimum performance. 3 2. In case of AMR. with different services and traffic types. 0 1.MA_id.2%. 2 2. when DFCA is activated. The simulator has realistic mobility models in which intracell or intercell handovers can be triggered due to level or quality reasons. the CIR targets must be setup correctly in the case of DFCA. AMR5. TABLE I. Both uplink and downlink are simulated and the propagation is based on Okumura-Hata model with TU channel profile. the gains get reduced down to a 12% with the FR codec. These simulations have been run using only one type of speech codec: either EFR or AMR. All of them have been executed over a regular grid scenario. although DFCA achieves better relative performance in more irregular environment as shown in [9]. Isolated EFR/AMR speech traffic Fig. 2 1. This ensures that the CIR is maximized across the network. 3 Time Slot 0 1 2 3 6 15 14 9 -1 13 -1 5 -1 1 7 -7 16 21 19 21 15 4 22 7 10 8 23 16 13 -6 -2 -7 11 12 21 -4 PROPAGATION 4 5 6 7 14 6 20 21 14 0 -3 11 10 21 7 12 25 18 3 10 8 -8 20 17 7 -8 -9 8 9 18 -9 7 0 -2 24 16 Band Slow fading correlation distance Slow fading std Fast fading 900 MHz 50 m 6 dB according to TU channel ANTENNA Sectorization Beam width Gain 120 º 65º 13 dB TRAFFIC PARAMETERS Figure 1. A.

5 2 1. which indicates that it is able to extract the gains from AMR share even when the percentage of these mobiles is relatively low. Isolated EFR and AMR performance different for each type of codec: 20dB for EFR. the chart presented in Fig.. where the maximum network capacity is shown depending on the share of different codec types. where the reference network EFR users are highly degraded when compared to the AMR users in the same network. and also why the performance is better with AMR-HR than with AMR-FR. Mixed EFR/AMR performance (50% share) DL PC offset (P0) . In the reference case. based on its CIR driven allocation: those users who request higher link quality (higher CIR target) will be allocated to channels with low interference. and thus big gains are achieved only in the case of 100% AMR penetration. when the level of interference in the network is lower. Mixed speech and EGPRS traffic This section presents simulation results of a mixed scenario consisting of EFR speech and EGPRS traffic. This is observed in Fig. C. whereas in DFCA case the quality for both is more or less the same. Mixed EFR/AMR speech traffic The following simulations consider the case when there are different types of mobiles in the network that use different codecs. where the performance of a network with 50% users of EFR and 50% users of AMR5. The table II presents the EGPRS environment used in this analysis TABLE II. Network capacity with mixed EFR/AMR (reference vs DFCA) While a normal channel allocation method doesn't take into account the specific needs of each user.EFR/AMR Performance for REF/DFCA Isolated services 16 4 3. EGPRS SETTINGS On 5 dB 6 dB Effective Frequency Load (%) Ref EFR Ref AMR-FR DFCA EFR DFCA AMR-HR Incremental Redundacy DFCA CIR target for EGPRS Figure 3. In this kind of scenario. a file with a fixed size is transferred to the MS). 12dB for AMR-FR and 20dB for AMR-HR. the gains from AMR codecs get diluted in terms of network capacity increase because the EFR users will be unsatisfied when the global interference level increases. This has been selected as a good model for new upcoming services such as MMS messaging (i.e. 3. Looking at the capacity figures.5 0 0% Bad quality samples (%) Mixed EFR-AMR Performance DFCA vs Ref Case Erl/MHz/Cell (AMR) 14 12 10 8 6 4 2 0 0 2 4 Erl/MHz/Cell (EFR) 10% 20% 30% 40% Effective Frequency Load (%) Ref EFR DFCA AMR-HR DFCA EFR DFCA AMR-FR Ref AMR-FR 6 8 Reference DFCA Figure 2.5 1 0. An illustration to this is shown in Fig.5 3 2. it can be observed that DFCA presents a much more linear behaviour than the reference case. Only DL traffic is simulated.90 is displayed. DFCA does and it is able to get a homogeneous quality distribution among all users. which is a fair assumption according to the asymmetric behaviour of this traffic model. 3. the algorithm has more flexibility to choose within the available channels. As a particularity of the allocation algorithm. DFCA provides powerful means to equalize the quality perceived by all the users. 4 can be drafted. while the users with strongest link level protection will be allocated to channels with more interference. This explains why the gains get reduced from EFR to AMR cases. The packet switched service studied here is the FTP service. If we consider that the capacity of a network is determined by the quality of the worst performing channel type (EFR). B. Performance of REF/DFCA with 50% AMR share Bad quality samples (%) 12 10 8 6 4 2 0 10% 15% 20% 25% 30% Figure 4.

on Communications.5 4. "A performance bound on dynamic channel allocation in cellular systems: Equal load.e. vol. T. S. Wiley. Borgonovo. If the same load situation is compared. 2001 spring. In a network with partial AMR capable terminal penetration DFCA is able to extract the maximum benefit of the more robust AMR channel coding while providing equally good speech quality to users with legacy EFR terminals. This way a capacity gain of about 75% is achieved with partial AMR terminal penetration. Data performance indicator is the user throughput. Cheng. J. Gimenez. Zuliani. H. Molinaro. 14-23. The ability to select the radio channel individually to each connection based on the CIR criteria enables equalization of the end user experienced quality irrespective of the used channel coding. 4. 2600-2607. and a user is considered to be satisfied if this measure is higher than 70 Kbps. Miljanic. i.5 2. August 1994. This throughput is presented in terms of the average throughput. Conf. September/October 2001. Chuang. Tapia. pp. "Performance evolution of distributed measurements-based dynamic channel assignment in local wireless communications. 698-710. 178-185. M. vol 43. pp. These requirements are specified by a certain FER limit for the speech connections and with some minimum average throughput requirement for PS users. Sestini.Number of slots per user (DL) FTP file size (average) 3 120 KBytes The results are shown in terms of network capacity for different service proportions. 25292533. no. pp. 4. no. Foschini. L. no." IEEE International Conference on Communications.5 6.e. "The geometric dynamic channel allocation as a practical strategy in mobile networks with bursty user mobility. Jordan. pp. Halonen. 2002. "Dynamic channel allocation schemes in mobile radio systems with frequency hopping. GPRS and EDGE performance.5 8." IEEE Veh. Cimini.J. Z. May 1996."Strategies for handover and dynamic channel allocation in micro-cellular mobile radio systems. A. pp. Panzer. 333-344. vol.D. Tech.162.EFR Speech performance DFCA vs. where for the same speech load more PS traffic can be carried by the network (more than double PS capacity for medium to high CS loads)." IEEE Journal on Selected Areas in Communications. This agreement is partially supported by the Program to promote technical research (Programa de Fomento de la Investigación Técnica. on Vehicular Technology. F. Priscoli. pp. 42. DFCA is providing considerably higher TP and much better speech performance. Romero. A." IEEE Veh. no.. Indoor and Mobile Radio Communications. J. P. 1998. Capone. M. Melero." 12th IEEE International Symposium on Personal." IEEE Trans. The data traffic is measured in Kbps/Mhz/cell. [2] [3] Kbps/Mhz/cell (FTP EGPRS) 100 80 60 40 20 0 0. how much packet switched and circuit switched traffic may be carried by the network while some minimum quality requirements are met. REFERENCES [1] R. "Call blocking performance of distributed algorithms for dynamic channel allocation in microcells. Reference case 120 V. vol 3. on Vehicular Technology. A. F. A. G. Grilli. Beck. vol 2. ACKNOWLEDGMENT This work has been performed as part of the co-operation agreement between Nokia and the University of Malaga.2. 14. May 1994. "A practical DCA implementation for GSM networks: Dynamic Frequency and Channel Assignment. i. Verdone. Zanella. Salmenkaita.1." IEEE Trans. F. CONCLUSIONS The DFCA functionality is shown to yield significant capacity gains in a GSM network with multiple services. pp. February 1995. Khan. L. PROFIT) of the Spanish Ministry of Science and Technology. vol.5 [4] [5] [6] Erl/Mhz/cell (EFR speech) Ref erence DFCA [7] Figure 5. R. which indicates the network spectrum efficiency.. Mixed EGPRS. 1989. Network capacity with mixed EFR/EGPRS (reference vs DFCA) [8] [9] . network capacity. 8. which considers the data throughput in LLC level without accounting for the impact of higher level protocols. F. Baiocchi. The speech traffic is measured in terms of Erlangs/Mhz/cell. "The impact of signal strength measures on the efficiency of dynamic channel allocation techniques. When applied to a network with a mix of speech and EGPRS services DFCA can provide significant data capacity gain with medium to high circuit switched traffic loads. J. Tech. pp. 1400-1404. vol 44.. The last figure show a clear capacity gain provided by the DFCA solution over the reference case.J. A. Speech performance is measured and benchmarked as explained in section A. GSM. J." IEEE Tans. Conf. 157. Chih-Lin I.