UMTS to WLAN Handover based on A Priori Knowledge of the Networks

UMTS to WLAN based on A Priori Knowledge of the Networks Mylène Pischella, Franck Lebeugle, Sana Ben Jamaa FRANCE TELECOM Division R&D 38 rue du Général Leclerc -92794 Issy les Moulineaux - FRANCE mylene.pischella@francetelecom.com Abstract Mobile terminals measurements capabilities are strong limitations that may impede seamless from UMTS to WLAN. In this article, we propose an inter-system method that does not require in-line measurements on WLAN. It identifies the most probable WLAN Access Point for, from coverage point of view, depending on UMTS signal measurements and on a probability database. This method can be implemented in the RNC, or in a B3G common controller. Simulation results in a WLAN hot spot deployment show that based on a priori knowledge of the networks increases success rate, compared to classical blind. Its performance can be improved by optimizing the probability database. Besides, the proposed method enables to forecast 's success, and adapt accordingly. based on a priori knowledge of the networks can consequently be integrated into several B3G algorithms, to decrease delay and complexity. Index Terms Inter-system, UMTS, WLAN, blind. I. INTRODUCTION UMTS and Wireless LAN networks are complementary solutions to provide broad coverage with low data rates, and high data rates in hot spots. Inter-working between both networks has been studied in the standard [1] and in several articles [2][3]. In Beyond third Generation (B3G) context, Joint Radio Resource Management will be used in order for the mobile terminal to be always best connected. To achieve seamless from UMTS to WLAN [4][5], information on WLAN load and coverage is needed. Network architecture requirements have been specified for B3G to enable load information exchange [6]. Signal strength information can be obtained by the mobile terminal through measurements, but it depends on the mobile terminal's capabilities. Inter-system measurements may be quite costly, even if the mobile terminal is equipped with a dual receiver. For from UMTS to WLAN, nothing is forecast in 3GPP standard for informing the mobile terminal of the characteristics of the WLAN Access Points it has to measure, and the mobile terminal may have to scan all possible frequencies. 2009 1-4244-0355-3/06/$20.00 (c) 2006 IEEE Inter-system based on a priori knowledge of the networks [7] can be used in order to bypass mobile terminal's measurements limitations. It is a blind method, that computes the identity of the most probable target cell or Access Point towards which the mobile terminal could perform an inter-system. The candidate target cells or Access Points are sorted by descending probability of being best-suited for, from signal strength point of view. The probability is obtained by combining measurements on the primary network's cells or Access Points with the probability database values. No measurement on the target network is required. In the present article, the efficiency of inter-system based on a priori knowledge of the networks for from UMTS to WLAN in a WLAN hot-spot deployment scenario is assessed. The article is organized as follows. Section II describes intersystem based on a priori knowledge of the networks, and reminds the performance results obtained in previous article for from UMTS to GSM [7]. Section III presents different options for the implementation of the method into UMTS and WLAN networks. Section IV presents simulation results in order to assess the method's relevance. The influence of the probability database's building method is evaluated, and an improvement of the method is proposed. Section V concludes this paper. II. DESCRIPTION OF THE METHOD Inter-system based on a priori knowledge of the networks [7] uses prior information on the networks and signal strength measurements on the current network, in order to perform blind s. The probability of performing a from each cell (or Access Point) of the primary network to each cell (or Access Point) of the target network is stored in a probability database. When inter-system becomes necessary, probability database information is combined with the signal strength values of the cells (or Access Points) that the mobile terminal has been measuring in normal operations on its current network. The obtained combined probability of each target cell (or Access Point) gives the probability for this cell or (Access Point) to be best suited for inter-system, from signal strength point of view, at the mobile terminal's location. The

target cell (or Access Point) with highest combined probability can be chosen for direct blind. For from UMTS to WLAN, the probability database contains, for each UMTS cell, the probability of performing a successful, from coverage point of view, from this UMTS cell to each of the WLAN Access Points: P UMTS->WLAN (i,j). It can be initialized either with a planning tool modeling the studied area, or with measurements on field. Then, the probability database can be updated in-line, by reinforcement learning [8], in order to improve the method's accuracy. The influence of the probability database's building method on the efficiency of based on a priori knowledge of the networks is studied in section IV.B. For each mobile terminal, UMTS to WLAN based on a priori knowledge of the networks provides the identity of the target WLAN Access Point, depending on the CPICH Ec/Io values received from the different UMTS cells. Let M 1, M 2,, M N be the measured signal strength of the N best UMTS cells, among which M best is the highest signal strength. M i values are the CPICH Ec/Io of the active set cells and of the monitored set cells, which have been reported to the RNC in a MEASUREMENT REPORT message [9]. For each WLAN Access Point j, the combined probability (CHO_Probability (j)) is computed, based on these measurements and on the probability database: M i CHO_Probability (j) = PUMTS > WLAN ( i, j) i= { 1, N} M best In this equation, PUMTS > WLAN ( i, j) refers to the probability of performing a successful, from coverage point of view, from this UMTS cell i to WLAN Access Point j, stored in the probability database. Then, WLAN Access Points are sorted by descending combined probability. The Access Point with highest combined probability is the target Access Point for blind. Besides, combined probability values can also be used for deciding whether blind can be performed or not. A threshold can be set on the combined probability of the best target Access Point, above which blind is highly likely to succeed. The distance between the combined probability of the best target Access Point and of the second best target Access Point can also be used to discard the cases with too high uncertainty on the target Access Point. In [7], this method has been applied to from UMTS to GSM in a border-of-coverage UMTS deployment scenario, where some UMTS cells were not co-located with GSM cells. GSM coverage was continuous. The method has been compared with blind using a one-to-one mapping table between each UMTS cell and a GSM cell. Simulation results with a dynamic simulation tool have shown that the proposed method enabled to decrease blind failure rate, and to increase the probability that the target GSM cell has the best RxLEV at the location. Contrary to the GSM case, in the present study, WLAN Access Points are deployed in non-contiguous hot spots areas, and have far lower coverage than UMTS cells. However, the simulation results of section IV will show that the method still leads to important performance increase, compared to one-to-one mapping blind. III. DIFFERENT OPTIONS FOR IMPLEMENTATION A. Direct implementation into UTRAN RNC based on a priori knowledge of the networks can be used for from UMTS to WLAN, even without interworking between the two networks. Indeed, the RNC can be responsible for storing the probability database, and for taking blind decisions. However, this implementation only allows simple algorithms, in which decision is only based on the WLAN coverage expectation obtained with probability database. As a consequence, may fail because of WLAN overload. Moreover, the RNC cannot be informed of the success of a, because it is not linked to target WLAN Access Point. Thus, it will not be possible to update the probability database. B. Integration into Common Radio Resource Management 3GPP has defined several levels of inter-working between UMTS and WLAN [1][10][11]. For Scenario 4 (Service Continuity), and especially Scenario 5 (Seamless services), it is likely that an inter-working unit will be required. Several coupling schemes between UMTS and WLAN have been proposed in the literature [2][3]. Joint Radio Resource Management algorithms between UMTS and WLAN will be located into a cooperation entity, which may be similar to the Common RRM defined for RRM between UTRAN and GERAN [12][13]. This cooperation entity will be responsible for gathering information on the different networks, on coverage, load, and users' requested Quality of Service. It will then take RRM decisions, among which decision, based on multi-criteria algorithms. The probability database could be integrated into the cooperation entity, and be responsible for providing signal strength estimations on the target network. The general flowchart for integration of the proposed method into multi-criteria algorithm is depicted on Fig. 1. Only the updating of highly dynamic criteria values, load and coverage, is represented. Service information, for each mobile terminal and each network, varies at a larger scale, and it is therefore assumed that the cooperation entity is aware of them. Load information is directly provided by the RNC and the Access Points, either periodically or on event. Signal strength information on UMTS cells corresponds to the measurements performed by the mobile terminal for its active set maintenance. Internal UMTS management requires measurements on the active set and on the monitored set cells, that are sent to the RNC either periodically or on event. The 2010

RNC can then forward this information to the UMTS-WLAN cooperation entity. The identity of the most probable target WLAN Access Point for is then computed thanks to UMTS signal strength values and to the probability database. The combined probability of target WLAN Access Point can also be provided to the multi-criteria algorithm, in order to take into account this estimation of the success probability in the decision. Then, the multicriteria UMTS to WLAN algorithm is responsible for deciding whether is required for this mobile terminal or not. No assumption is made on this algorithm, with the exception that actual signal strength values on WLAN should not be required. UMTS network is a hexagonal network with tri-sectored sites. Radio Resource Management is only modeled on one site (3 cells), however the closest 12 cells of the first interfering ring may be included in a mobile terminal's active set. 8 WLAN Access Points are spread in the 3 central UMTS cells, as shown on Fig. 2. AP 3 Cell 0 AP 0 AP 7 Consequently, the proposed method can serve as an input for multi-criteria algorithms. Thanks to the time and complexity gained from avoiding measurements on WLAN, a more accurate reaction to radio conditions modifications can be expected. The proposed method will also permit to prepare in advance and therefore enable easy context transfer between UMTS and WLAN, which is required in order to achieve seamless mobility. Cell 0 AP 1 AP 4 AP 2 AP 5 Cell 2 AP 6 WLAN AP WLAN Access Point load UMTS RNC UMTS cells load UMTS RNC UMTS cells signal strength received in Measurement Report message from the mobile terminal UMTS-WLAN Cooperation entity probability databases WLAN AP And combined probability AP com- -putation Multicriteria UMTS to WLAN algorithm decision towards WLAN AP Fig. 1: Integration of the method into multi-criteria algorithm IV. PERFORMANCE OF HANDOVER BASED ON A PRIORI KNOWLEDGE OF THE NETWORKS A. Simulation assumptions We use an internal dynamic UMTS Release 99 and WLAN 2.11b simulation tool, developed by France Telecom and based on OPNET. It models all layers, from the physical layer to the application layer. It implements UMTS layer 2 and 3 procedures including macrodiversity, admission control, control plane delays, RRC state management. WLAN CSMA/CA protocol and roaming are also modeled. The Fig. 2: Hot spot deployment We suppose that additional information is available, that indicates, at any location, if the mobile terminal is under WLAN coverage at this location. This could be achieved by detection of a WLAN beacon, without further decoding, if the mobile terminal is equipped with a dual receiver. WLAN Access Points could also broadcast beacon information on the UMTS frequency. Consequently, the performance of UMTS to WLAN based on a priori knowledge of the networks is only evaluated in locations where is likely to take place, e.g. in locations with WLAN coverage. B. Optimization of the probability database The performance of based on a priori knowledge of the networks depends on the accuracy of the probability database. An optimal means for building the probability database would be to obtain on-field statistics of inter-system s based on mobile terminals' measurements. This method has been used for from UMTS to GSM in [7], and proved to be quite efficient. However, for from UMTS to WLAN, such information is not available. As a consequence, our simulation tool, which computes propagation losses according to Okumura-Hata model [14] in a UMTS macro-cell, dense urban environment, with a spatially-correlated shadowing, is used to design different probability databases. The accuracy of each probability database is then evaluated by computing the success rate obtained with our method ( based on a priori knowledge of the networks), and using this database. probability databases are built thanks to a probe mobile terminal, which moves within the modeled area and checks probability criterion at every shadowing 2011

value update (e.g., every 20 meters). The different criteria are given hereafter: Best CPICH Ec/Io criterion: For each UMTS cell i and each WLAN Access Point j, P UMTS->WLAN (i,j) is increased every time that the probe mobile terminal is under the coverage area of Access Point j, and receives the best CPICH Ec/Io from UMTS cell i, whatever its value. Sufficient CPICH Ec/Io criterion: For each UMTS cell i and each WLAN Access Point j, P UMTS- >WLAN(i,j) is increased every time that the probe mobile terminal is under the coverage area of Access Point j, and receives a CPICH Ec/Io from UMTS cell i which is more than a threshold T. This criterion has been tested for T varying from -12 db to -20 db [15]. Besides, success is compared with a reference blind method in which each UMTS cell is assigned only one target WLAN cell. The target WLAN Access Point in this one-to-one mapping is the one with highest probability, according to the probability database obtained with best CPICH Ec/Io criterion. is successful if the mobile terminal is in the coverage area of the chosen WLAN Access Point, at the location. Fig. 3 shows that with the one-to-one mapping blind, success rate is limited to %. It is equal to 74% with the probability database, built on Best CPICH Ec/Io criterion. Sufficient CPICH Ec/Io criterion leads to higher success rate, if the CPICH Ec/Io threshold is more than -18 db. A maximum of 82% is reached with CPICH Ec/Io threshold equal to -15 db. With this threshold, only the UMTS cells that are close enough from the mobile terminal are considered in the probability calculation. 85 For use on field, the criterion for building probability database should consequently be adapted to local deployment's characteristics. Each local area with similar characteristics should possess its probability database. Additional optimization could be considered, for instance by using different probability databases for the same local area, depending on the hour of the day and on the corresponding expected traffic load. Moreover, the probability database could be updated thanks to reinforcement learning. statistics obtained on field, once the method has been implemented with the initial probability database, could be used to automatically correct the probability database. C. Improvement of efficiency In the previous section, the target WLAN Access Point for was the Access Point with highest combined probability CHO_Probability, whatever the value of the combined probability. In order to improve the efficiency of based on a priori knowledge of the networks, we propose to use the characteristics of the combined probability values. A is likely to succeed if: The combined probability of the best Access Point is more than a given threshold (called absolute confidence value), And the distance between the combined probability of the best Access Point and of the second best Access Point is more than a given threshold (called relative confidence value). If, for a mobile terminal, both criteria are fulfilled, then can be triggered directly with high success probability. Absolute and relative confidence values shall be set in order to obtain a high success probability, but should also lead to a sufficient number of triggered s. 100 90 success rate (%) 75 65 one-to-one mapping Best CPICH Ec/Io criterion Sufficient CPICH Ec/Io criterion % 50 40 success rate of triggered blind probability of triggering a blind 55-20 -18-16 -14-12 CPICH Ec/Io Threshold (db) 30 18 20 22 24 26 28 30 32 34 36 38 absolute confidence value (%), relative confidence value = 0% Fig. 3: Comparison of the different methods for building the probability database Sufficient CPICH Ec/Io criterion is able to smooth down shadowing variations, and is therefore more efficient than Best CPICH Ec/Io criterion. Fig. 4: Influence of absolute confidence value on the probability to trigger a blind and on the success rate of the triggered s. Fig. 4 represents the influence of absolute confidence value, with relative confidence value set to 0%. The probability of triggering a blind rapidly decreases, whereas the success rate's increase is not very important. 2012

In order to improve success without discarding too many users, we study the influence of relative confidence value, with the limit absolute confidence value for obtaining 90% of triggered s: 24 % (see Fig. 5). % 100 90 50 40 0 2 4 6 8 10 12 Relative confidence value (%), absolute confidence value =24% success rate of triggered blind probability of triggering a blind Fig. 5: Influence of relative confidence value on the probability to trigger a blind and on the success rate of the triggered s. A good compromise can be achieved with an absolute confidence value of 24% and a relative confidence value of 4%. With this parameter's setting, success rate is equal to 87%, and the probability of triggering a is equal to 76%. The previous results have shown that the combined probabilities can be used in order to forecast 's performance, and consequently to improve it by restricting blind triggering to the most likely to succeed. This information could be inserted into complex multi-criteria algorithms, as seen in section III.B. Moreover, if mobile terminals are capable of performing WLAN measurements, absolute and relative confidence values could be used to discriminate the cases when blind can be performed, from the cases when measurements are necessary. The setting of absolute and relative confidence values should then depend on operator's strategy. Indeed, measurements are costly in terms of signaling and delay, but success following measurements is almost certain. A trade-off should consequently be found between the signaling and delay gains brought by blind, and the remaining failure probability. V. CONCLUSION In this article, inter-system based on a priori knowledge of the networks has been adapted for from UMTS to WLAN. The method uses a probability database and UMTS cells signal strength to identify the WLAN Access Point that maximizes the probability of performing a successful, from coverage point of view. It can either be directly inserted into the RNC, or be integrated into UMTS-WLAN cooperation entity, depending on UMTS-WLAN implementation's choices. In the second case, the proposed method can serve as an input for multi-criteria inter-system algorithms, in order to avoid signal strength measurements on WLAN. The performance of based on a priori knowledge of the networks has been compared with one-to-one mapping blind. success increases of up to 22%, depending on the probability database building method. Additional improvement can certainly be obtained by using reinforcement learning based on field statistics, in order to update the probability database. The characteristics of combined probabilities have also been used in order to restrict triggering to the most favorable cases. Without restriction, 82 % of success rate can be achieved. Optimal restriction leads to a success rate of 87%, for a triggering probability of 76%. Consequently, based on a priori knowledge of the networks leads to satisfying results for from UMTS to WLAN in a WLAN hot spot deployment. Further work could consist in optimizing the probability database, via reinforcement learning, and in evaluating the performance of multi-criteria s using the probability database. REFERENCES [1] 3GPP TR 22.934, Feasibility study on 3GPP system to Wireless Local Area Network (WLAN) interworking, Release 6, v6.2.0, 09/2003 [2] N.Vulic et al., "Architectural options for the WLAN integration at the UMTS radio access level", VTC 2004-Spring, May 2004, p. 3009-3013, vol. 5. [3] J-Y Song et al., "Hybrid Coupling Scheme for UMTS and Wireless LAN Interworking", VTC 2003-Fall, Oct. 2003, p. 2247-2251, vol. 4. [4] N. Sattari et all, "Seamless between WLAN and UMTS", VTC 2004-Spring, May 2004, p. 3035-3038 vol. 5. [5] H. Bing et al., "Performance analysis of Vertical in a UMTS- WLAN integrated network", PIMRC 2003, Sept. 2003, p.187-191 vol.1. [6] J. Luo et. al, "Investigation of Radio Resource Scheduling in WLAN coupled with 3G cellular network", IEEE Communications Magazine, June 2003, p.108-115. [7] M. Pischella, "Efficient inter-system based on a priori knowledge of the networks", ASWN 05, June 2005. [8] Richard S. Sutton and Andrew G. Barto, "Reinforcement Learning: An Introduction", The MIT Press, Cambridge, Massachusetts, 1998. [9] 3GPP TS 25.331, "Radio Resource Control (RRC) protocol specification", Release 99, v3.21.00, 12/2004. [10] 3GPP TS 22.234, "Requirements on 3GPP system to WLAN interworking", Release 6, v6.3.0, 06/2005. [11] 3GPP TS 23.234, "3GPP system to WLAN interworking: system description ", Release 6,, v6.5.0, 06/2005. [12] 3GPP TR 25.881, Improvement of RRM across RNS and RNS/BSS, Release 5, v5.0.0, 12/2001 [13] 3GPP TR 25.891, Improvement of RRM across RNS and RNS/BSS (post-rel-5), Release 6, v0.3.0 (draft version), 02/2003 [14] J. Laiho et al., "Radio Network planning and optimization for UMTS", John Wiley & Sons, 2002., pp. 103-104. [15] 3GPP TR 25.133, "Requirement for support of radio resource management (FDD)", Release 99, v3.21.00, 06/2005. 2013