2G/3G INTER-RAT HANDOVER PERFORMANCE ANALYSIS

Transcription

1 2G/3G INTER-RAT HANDOVER PERFORMANCE ANALYSIS A. Mohammed, H. Kamal2, S. AbdelWahab2 'Department ofsignal Processing, Blekinge Institute of Technology, Ronneby, Sweden 2Alcatel-Lucent, Egypt Keywords: GSM, UMTS, Handover Performance, Handover Algorithms. Abstract One of the most interesting UMTS networks features is their integration with the 2G networks that provides seamless Endto-End services. Current widely deployed 2G and 3G networks provide InterRAT (Inter Radio Access Technology) mechanisms enabling interoperability between them in a manner that is almost transparent to the subscriber. These enable maximum benefit from 3G services while ensuring wireless coverage continuity in geographically extended networks, where 3G coverage halls exist. These mechanisms include InterRAT idle mode reselection, InterRAT dedicated mode reselection for PS (Packet Switching) services, and InterRAT handover for voice calls. The aim of this paper is to evaluate the InterRAT hand-over (HO) performance between 2G and 3G networks. Analysis of the performance would be presented in the two HO directions. We propose an analytical model where the introduction of the 3G cell's load as one of the HO initiation parameters is considered. In addition, an Inter-RAT HO ping-pong defensive mechanism is proposed. 1 Introduction The purpose of the inter-rat HO (handover) procedure is to keep the service provided to the UE (User Equipment) while moving away from the coverage are of one RAT (GSM or UMTS) to another (UMTS or GSM). Before the handover process is completely achieved, different phases should take place; HO triggering conditions, measurements done by the mobile UE on the neighbor cells, and selection of the best candidate cell that fulfills the handover criteria condition. The handover initiation could be based on several criteria; whether signal level, and UE distance away from the BS, or even based on the desired service. Over facilitating the HO procedures might cause the mobile terminals to handover their calls from one cell to another and return back to the first cell after a very short period causing the increase of the signaling messages on the network, which is known by the Ping-Pong effect. In the opposite sense, handovers might be designed in a way not to be performed easily, which would increase the call drop rate due to signal level degradation received by the UE. A tradeoff between fast and easy HO and between delayed HO exists. In this paper we propose an analytical model where the signal quality in the UL (uplink) is also considered as one of the handover initiation parameters for the 3G-2G HO direction. Real data from live networks is also used to complement the analytical results. The two handover directions; 3G-to-2G inter-rat HO and 2G-to-3G inter-rat HO will be considered. 1) 3G-to-2G Inter-RAT HO direction. Due to the channel fading conditions faced by the UE, received signal sees fluctuations that might cause unnecessary handovers, even for non-moving mobiles. In order to reduce these unnecessary handovers, a proper design for the handover initiation conditions is required. Unlike in 2G technology, in 3G networks, the traffic is considerable, as it has a direct effect on the network QoS. The noise level increases with the increase of active users in the cell. In this paper we propose the consideration of the 3G-cell traffic, which is related to the noise level seen at the Node_B, for the 3G-to-2G HO triggering conditions, beside the signal quality evolution done by the UE, which is specified by 3GPP standard. 3G-cell traffic could be evaluated by the Node B, based on the noise level received from each UE in the UL direction. 3G-cell capacity is usually limited by the UL interference level generated by the increased number of users messaging the Node B at the same time and on the same frequency. Beside the signal quality measured by the UE and expressed by the CPICH Ec/Jo, corresponding BER value could be obtained from individual link level simulations, the cell load is also added to the HO triggering conditions. The UE signal quality reflects only the interference level received by the UE and not the interference level received by the Node-B. After the completion of the handover triggering conditions, the UE goes into CM (Compressed Mode) status where it could make measurements on the 2G neighbor cells. The UE sends the measurements results to the network to decide which cell is the best for HO after the fulfillment of the HO criteria. The HO criterion for the Inter-RAT case is the (event 3a) criteria as specified by 3GPP standard. The criteria states that, "The estimated quality of the currently used 3G frequency is below a certain threshold and the estimated quality of the 2G system is above a certain threshold".

2 This introduction of the 3G-cell traffic is expected to reduce the noise generated in the 3G cell and hence, the enhancement of the 3G system performance is anticipated. Different from [1], the signal quality (BER) is considered as one of the HO initiation conditions in our analysis beside the UL noise level seen by the Node B. In the next section our proposed analytical model will be presented. 2) 2G-to-3G Inter-RAT HO direction. In the GSM to UMTS direction, it's desirable to have all Mobile Station (MS) terminals that are UMTS ready to be served in the UMTS service layer. Nowadays operators deploy 3G services in two strategies. The first strategy is to extend the coverage of the deployed GSM network, and the other is to co-locate UMTS sites side-by-side of the existing GSM sites, which aims at introducing new 3G services as a new service layer. Decision to handover the UE/MS from the serving GSM cell to one possible target 3G cell must ensure acceptable radio conditions in the target 3G cell. Radio conditions of the target 3G cells is defined by two main quantities [3] CPICH Ec/Jo and CPICH RSCP, which are the received energy per chip of the primary common pilot channel of neighbor 3G cell and the Received Signal Code Power of the primary common pilot channel, respectively. CPICH RSCP reflects 3G radio propagation conditions at the UE/MS location. CPICH Ec/Jo reflects 3G cell signal quality at the UE/MS location, as well as load situation in target 3G cell [4], recalling that noise level in 3G is affected by the number of UE/MS served by the cell. Joint optimization of GSM to UMTS handover and UMTS to GSM handover is necessary in order to minimize call drop probability encountered in both radio access technology and to minimize probability of unnecessary handovers. These unnecessary handovers result in unnecessary increase in signaling load, and is considered the main cause of the undesired ping-pong phenomena. In this paper we propose a defensive mechanism to reduce the effect of ping-pong phenomena between GSM and UMTS without a great effort in optimizing handover detection thresholds. Such defensive mechanism could be implemented in either BSS, or UTRAN. The rest of this paper is organized as follows. In Section 2 we demonstrate handover process in both GSM to UMTS direction and visa versa. We underpin the model by performance metrics proposition that can be used for the purpose of handover process optimization. Section 3 includes numerical results extracted from field measurements. Section 4 proposes the ping-pong defensive mechanism. Finally, Section 5 concludes the paper. 2 Analytical Model The HO process passes by phases before it is completely performed. In this section we present the HO initiation and execution analytical model that will be used in this paper. Two different models will be used for the two different HO directions: 3G-to-2G HO and 2G-to-3G HO. 1) 3G-to-2G Inter-RATHO direction. The mobile UE performs measurements on the 2G neighbor cells, once the triggering conditions are fulfilled. These triggering conditions take into consideration the signal quality at the mobile UE denoted by Qused, presented by the measured CPICH Ec/Jo value, and also the cell load presented by the UL interference level seen by the Node B denoted by ULin-. The best cell among the neighbor cells that fulfills the HO criteria (event 3a) is to be selected for HO execution. Event 3a criteria could be denoted mathematically as: QUsed< ThUsed -H3a and Qtarget +CfOtarget > Thtarget +±H3a where Qused is the quality of the 3G cell signal measured by the UE and presented by CPICH Ec/Jo, Qtarget is the quality of the target 2G cell presented by the received signal level in dbm, CIO is the Cell Individual Offset of the target 2G cell., H3a is the hysteresis margin for event 3a, Thused is the threshold value for the received CPICH Ec/Jo, Thtarget is the 2G cell received signal level threshold for event 3a. The measured CPICH Ec/Jo reflects the signal quality received by the mobile UE. The common pilot channel (CPICH) is the channel that has to be detected by all the mobile UEs everywhere in the cell. If the UE is not able to detect the CPICH of certain (x) cell, this UE is said to be: "not covered' by this (x) cell. Hence CPICH defines the coverage area for the cell and it is transmitted by 10% of the total Node_B power. We propose the division of the CPICH Ec/Jo range into three regions defined by two thresholds Th1 and Th2; a) HO is needed region, where CPICH Ec/Jo < Th1. b) No HO is needed region, where CPICH Ec/Jo > Th2. and c) HO might be needed region, where Th1 < CPICH Ec/Jo < Th2. In the third region the UE receives somehow a degraded signal quality due to the quite increased interference level. At this point another metric needs to be confirmed before activating the CM and triggering the measurements on the neighbor 2G cells. This metric is the UL interference level (ULint) which presents the 3G cell load. Figure 1 shows the three regions on the CPICH Ec/ Jo axis. HITs nee4d Th Th2 CPICH-EcIo HO might be needed Figure 1: CPICH Ec/Jo regions. No HOis 7needed

3 The handover procedures of initiation and execution for 3Gto-2G HO direction are explained in Figure 2. and also indicates whether these tasks shall be performed when RXLEV (Received Signal level from GSM cell in dbm) of the serving cell is below or above the threshold. Qsearch C can take the value from two different sets. It controls either if the UE/MS search for 3G cells if signal level below threshold (0-7): -98, -94,..., -74 dbm, oo (always) or above threshold (8-15): -78, -74,..., -54 dbm, oo (never) [2]. Appropriate setting of Qsearch C reduces number of unnecessary measurement reporting and hence affects handover probability. Moreover, setting of this parameter directly reflects operator UMTS deployment strategy. Setting Qsearch C such that UMTS measurement is triggered only if GSM level is below a certain threshold is a preferable scenario in case UMTS cells are deployed in areas where GSM coverage holes exist. In this case, it becomes reasonable to set Qsearch C in 0 to 6, as handover becomes necessary only the UE/MS experienced degradation in GSM signal level. In case UMTS sites are colocated with GSM sites, it is reasonable to trigger UMTS measurements if GSM level is above a certain threshold. This is because in this case the GSM signal strength will be almost always higher than UMTS signal strength. Hence it makes no sense just to wait until the GSM signal strength is degraded, as 3G signal won't be good anyway. Recommended setting is "Always search for 3G neighbors". yes HO execution Figure 2: 3G-to-2G Handover initiation and execution analytical model. From the previous analytical model, conditions for a HO from 3G cell to 2G cell to be occurred are; {[Qused < Th1] or [(Th1 < Qused < Th2) and (ULint > Ith)]} and {Qused < Thused -H3a} and {Qtarget + ClOtarget > Thtarget + H3a} and {Qtarget = max [Qt, t =1,.N]} (1) Where ULint is the UL interference level seen by the NodeB, and Ith is the interference threshold value. 2) 2G-to-3G Inter-RAT HO direction For GSM to UMTS handover the UE/ME has to measure the UMTS neighbour cells. UMTS cells that should be measured by the UE/MS are identified and sent to the mobile on Signaling Associated Control Channel (SACCH).The network controls the measurements of UMTS cells by the parameter Qsearch_C sent on SACCH. Qsearch_C defines a threshold The UE/MS measures the neighbor UMTS cells indicated by the network. FDD REP_QUANT parameter controls which measurement quantity (RSCP or Ec/Jo) the UE/MS shall measure. In the rest of this paper we assume that the UE/MS measures CPICH Ec/Jo. In case Transmit diversity is applied on primary CPICH the received energy per chip (Ec) from each antenna shall be separately measured and summed together into a total received energy per chip on the Primary CPICH, before calculating the Ec/Jo. The UE/MS reports best UMTS cells, which is part of the neighbor cell list. The number of reported UMTS cells is controlled by the GSM network according to the value of the parameters FDD_MULTIRAT REPORTING in case of UMTS FDD and by TDD MULTIRAT REPORTING in case of UMTS TDD. The network keeps a bookkeeping of the reported UMTS cells and evaluates if handover is needed to be performed towards a target cell. Cells are ranked for handover condition evaluation according to the reported value of Ec/Jo, so that best cells are considered first. In order to compensate fast fluctuations in radio environment the network performs a sliding window averaging algorithm on the reported measurement quantity (CPICH Ec/Jo). TEC/IO is the handover detection threshold. Handover to a UMTS cell is performed if the condition in Equation (2) is fulfilled. Average (Ec /IO )> TE /Io (2)

4 3 Results In this section we would present how the probability that a connection executes a handover from 3G cell to 2G cell has been calculated. The calculations are based on both the analytical model presented previously and also on the measurements results from real live networks. A. UMTS to GSM Handover 1) Handover Probability Figure 3 shows the CDF (Cumulative Distribution Function) for the received CPICH Ec/Jo, based on measurements in a typical urban outdoor mobile environment. Normally the CDF should change for different environments and also depending on the number of installed base stations. CDF 1 7lul:::: i±wi(eb N,)*R*v where ilj, is the load factor, W is the chip rate, R is the bit rate of the service, Eb /NO is the required energy per bit over the noise power spectral density to establish a connection with a service bit rate R, and activity factor v. Figure 5 shows the load factor for three different services (bit rates), Speech AMR (Adaptive Multi Rate) 12.2 kbps, PS64 kbps, and PS384 kbps, respectively. It is clear from the figure that the load increases with the required bit rate for the different services. As the load increases, also the noise level in the cell increases. They are directly related by the following equation: NoiseRise= 1/1 - '71u (4) (3) 25% Load Factor 20% 15% 10% 5% CPICH EcIIO (db) 0% Se rviice Rate (bps) Figure 3: CDF for outdoor measurements on CPICH Ec/Jo. Figure 4 shows the CDF for the obtained Qu,sed (received signal level) in a typical outdoor rural 2G mobile environment. CDF G Rx Level (dbm) Figure 4: CDF for outdoor measurements on the 2G received signal level. In [6], it has been shown that the uplink load factor caused by one connection in the WCDMA system depends on the service the connection is establishing with the network. Figure 5: Load factor for different services per one connection. Figure 6 shows the PDF of the UL interference in different 3G cells deployed in a rural radio propagation environment. We can notice that the cells are facing low UL interference level. These curves are the output of averaging process over several days. Actually in nowadays 3G networks the traffic is low especially for PS services. One of these curves has been used to calculate the HO probability in case the load exceeds a certain threshold. PDF UL interference (dbm) 0.8 -Cell Cell Cell 3 Cell Cell Cell Cell Cell Figure 6: The PDF of the UL interference level received at the Node B.

5 For loaded cells these curves are expected to shift to the right. In addition, as the UL interference level increases this might cause call drops due to: 1) PRACH (Packet Random Access Channel) failure 2) RRC (Radio Resources Control) connection failure. One of the major interests in deploying 3G networks is to reduce the interference. Using the previous analytical model, the cumulative distribution functions obtained from the live network measurements, beside the CDF of the chosen cell with the UL interference values: the probability that the handover conditions are satisfied is calculated. Figure 7 shows the probability curves as a function of the threshold (Th1), and for different values for the hysteresis margin H3a that varies from 0 to 4. The other parameters have been set as shown in Table 1. Parameter Value CIO 0 Thused -12 db ThTarget -95 dbm Th2-5 db ULint -100 dbm Table 1: Parameters used for 3G-2G HO probability calculations. - H3a=0 H3a= 1 H3a=2 H3a=3 H3a=4 P_2G3G Thl (db) Figure 7: Probability of 3G to 2G HO curves We can notice that, when the "HO might be needed" region becomes smaller, whether by increasing Th1 or by decreasing Th2, the probability that a handover occurs from the 3G cell to the 2G cell increases. This could be explained, as the HO triggering conditions becomes harder, the HO probability decreases. 2) Call Drop Probability The call drop rate is an important metric used to evaluate the QoS in cellular networks. Call drop could be caused by uplink or downlink break [5]. The Downlink break is our point of interest in this paper. The link is assumed to break when the n received CPICH Ec/Jo (CPICH energy per chip over the total noise received in the band) is less than -19 db at the UE level. The CPICH Ec/Jo is directly related to the CPICH Eb/No by the processing gain of the established service by the user. This minimum CPICH Ec/Jo value is then depending on the service; Table 2 shows the minimum CPICH Eb/No values, corresponding to -19 db CPICH Ec/Jo value, for different services. Service Speech AMR 12.2 kbps CS 64 kbps PS 128 kbps PS 384 kbps Minimum CPICH Eb/NO -44 db db db -29 db Table 2: Minimum CPICH Eb/No values corresponding to minimum CPICH Ec/Jo. In order to calculate the CDP (Call Drop Probability) due to unsuccessful HO to a 2G cell, the following model has been assumed for a call drop to occur; a- The received CPICH Ec/Jo by the UE is below a certain threshold, and b- The HO to a 2G cell is unsuccessful. This model could be expressed by the following equation; CDP = P{(Ec Io) < ThCd } * {1- P(3G _2G)} where Th,d is the threshold for minimum received CPICH Ec/Jo value before the call drops, and P(3G 2G) is the probability that a HO occurs from 3G cell to 2G cell. CDP -.H3a =01 Th = Call Drop Ec/lo threshold (db) Figure 8: Call drop probability due to unsuccessful UMTS to GSM Handover. For considerable values for the DL interference the CDP increases. The DL interference is shown in the previous equation in the term: P{(Ecl Io) < ThCd }

6 Using the same parameters mentioned in Table (1), with H3a set to zero, and Th1 to -OdB, the CDP has been calculated and the results are shown in Figure 98. Although in nowadays live 3G networks, the traffic (load) is low, but in case one PS 128 kbps user performs HO to a 2G cell; that would decrease the cell load by about 15-16% and hence the UL interference level would be decreased by 0.75dB. For a PS 384 kbps user a decrease of 1.5dB could be achieved. B. GSM to UMTS Handover 1) Call Interruption Time A measurement campaign was carried out to assess call interruption time due to handover. The test was designed such that several handover types are experienced by the UE/MS. During the test the UE/MS experienced 20 handovers from GSM to UMTS, 14 handover from UMTS to GSM, and 46 Intra-GSM handovers. The call interruption time due to handover is computed as the time difference the UE/MS receives "Handover Command" message from the old cell and sends "Handover Complete" message to the new cell. This does not take into account the Handover decision time, as during that period the UE/MS is still in call. Figure 9 illustrates the call interruption time encountered due to Inter-RAT Handover and compared to Intra-GSM Handover. Similar performance was encountered for GSM to UMTS handover and for UMTS to GSM Handover. This is explained by the fact that signaling path is the same for the Inter-RAT Handover case. On the average the User experiences 200 milliseconds of call interruption, which should not affect voice quality compared to Intra-RAT handover ' E Q $ i, G) G-2G HO 2G-3G HO Handover Direction 3G-2G HO * Average * Maximum r StdDiv Figure 9: Call Interruption time in milliseconds for the three cases of GSM-GSM HO, GSM-UMTS HO, and UMTS-GSM HO. The Intra-GSM Handovers in Figure 9 includes both Inter and Intra BSC Handovers. The maximum call interruption period in case of Intra-GSM Handover was around 330 milliseconds, which is a typical value in case of Inter-BSC Handover. 2) Handover Probability GSM to UMTS handover probability indicates the degree by which the GSM network is off loaded and UMTS service is preferred. Figure 10 shows evolution of GSM to UMTS Handover probability with different values of TEC/IO. The effect of setting Qsearch C is also shown. For simplicity, the effect of averaging sliding window is not taken into account. However this should not affect the results because its rule is to compensate fast fluctuations in measured value, and not affecting the Handover detection process. The Handover probability is computed numerically as: JP(Ec Jlo > TECI )x P(RxLev < Q), Q(belo-w) HO lp(ec J o> TECIO)x P(RxLev > Q), Q(above)J (5) Where Ec/Io is the CPICH Ec/Io measured by the UE/MS in GSM dedicated mode, TEC/IO is the Handover detection threshold based upon CPICH Ec/Jo value, RxLev is the GSM signal level in dbm, and Q is the Qsearch_C parameter as specified in [2][2]. Qsearch C limits probability of GSM to UMTS handover, as stated in previous section, and it should reflect Operator strategy of employing UMTS technology. In case UMTS is employed for coverage extension, it is reasonable to trigger UMTS search in case GSM RXLEV is below Qsearch C value. When UMTS is co-located with GSM, it is reasonable to trigger UMTS search in case GSM RXLEV is above Qsearch_C value. 3) Call Drop Probability Hardening Handover decision by increasing handover detection threshold is a good choice to reduce the chance of unnecessary handovers. Reducing unnecessary handovers has a direct impact on reducing Call drop in UMTS side as the probability of underestimating radio conditions in UMTS vanishes. Also it has an influence on reducing ping-pong effect. t._ n 0 co n 0IL , u.1 0- lln"d.k$~~l ~ ~,~'b lb 'l0i,z TEcIIo Alwayes Search UMTS 0...a...Q=-2(below) Q=-78(below) Q=-74(below) Q=-78(above) 0Q=-74(above) Q=-70(above) Figure 10: GSM to UMTS Handover probability. Effect of setting Qsearch_C on Handover probability is shown.

7 A trade off between hardening and easing handover decision is required. As Handover decision gets harder, Probability of call drop increases as a result of signal strength degradation in the serving cell. Figure 11 demonstrates increase of call drop probability in correspondence to increasing TEC/Io- Call drop probability is simplified as [1]: PDrop = P(RxLev < RxLevMin)x [1- PHO] (6) where PDrop is the probability of call drop due to GSM to UMTS Handover, RxLevMin is the minimum level in dbm of accessing a GSM cell specified in [2] as RXLEV ACCESS-MIN, and PHO is the GSM to UMTS handover probability in Equation (5) GSM UMTS HO Prob Call Drop Prob o c G L TEcIlo Figure 11: Call drop probability due to delayed GSM to UMTS Handover versus GSM to UMTS Handover Probability. In practice Call drop probability experienced by the UE/MS is much less than the values shown in Figure 11; as the figure shows only call drop that might be encountered if GSM to UMTS was not probably triggered. Typically this is not the case, as the UE/MS has a chance to handover to another GSM cell if degradation is encountered in the serving cell. Call drop model that capture all GSM system dynamics should take into account the following Call drop events: 1) Call drop due to radio interface failure. 2) Call drop during handover execution, which include, Intra-cell, Inter-cell, Inter-BSC, and Inter-RAT Handovers. 3) Call drop due to System failures, which include, Hardware failures, Software failures, and Transmission subsystem failures. 4) Call drop due to explicit preemption. As it's common to have a grade of service assigned to individual subscribers, The GSM network can explicitly pre-empt calls in case of congestion so that it gives priority to a higher grade subscribers Inter-RAT Handover Ping-Pong Defensive Mechanism Proposal GSM to UMTS handover as described in Section 2 is considered a better cell handover, in which the network will always keep trying to send the UE/MS to the UMTS cell as long as the radio conditions are good. This raises the risk of ping-pong effect, as the UE/MS oscillate back and forth between the GSM and UMTS cells in case of variations in radio conditions, and sensitive reactivity of the handover triggering thresholds. One possible way to minimize such undesired effect is to harden or prevent handover decision from a serving GSM cell to a target UMTS cell, if the GSM access request indicates handover from UTRAN. We modify handover detection in Equation (2) to be: Average (Ec Io) > TEClIO + Hp (7) Where Hp is a penalty in db added to the TEC/IO threshold, if the call is handover from UMTS. Handover to UMTS must be enabled again by neglecting Hp value after a certain time. We define T1p as time during which handover decision is based on (4). After expiry of TIp timer the handover decision must be based on (2) again to enable GSM to UMTS handover again. A similar defensive mechanism could be employed in UMTS to prevent handover to GSM if call access request is indicating handover from GSM. 5 Conclusions 2G/3G InterRAT handover is an important telecommunication feature in nowadays heterogeneous wireless networks. We provided a brief description of this feature in the two directions, as well as analysis for its performance through probability calculations and measurements from live networks. An analytical model, and suggested algorithms that should bring improvement to the performance of these telecommunication procedures have been proposed. Tuning of different algorithms parameters was elaborated and tested in a commercial network. 6 Acknowledgements Many people have been involved in the preparations and execution of the tests, and have provided support led to this paper. The authors would like to acknowledge the contributions of their colleagues, in particular Ervin Farkas at Alcatel-Lucent Romania, Florent Colin and Didier Esclamadon at Alcatel-Lucent France, Hesham Sabry at Alcatel-Lucent Egypt, and Idris Yusof at Alcatel-Lucent Indonesia. "The views presented in this work are of the authors, and does not represent Alcatel-Lucent position regarding the 2G/3G HO".

8 References [1] W. Zhao, R. Tafazolli, B. G. Evans, "Intemetwork Handover Performance Analysis in a GSM-Satellite Integrated Mobile Communication System", IEEE Journal on Selected Areas in Communications, 15, (1997). [2] 3GPP TS "Radio subsystem link control". [3] 3GPP TS "Physical layer - Measurements (FDD)". [4] 3GPP TS "Radio Resource Control (RRC) protocol specification". [5] C. Brunner, A. Garavaglia, M. Mittal, M. Narang, J. V. Bautista. "Inter-System Handover Parameter Optimization", QUALCOMM Incorporated, (2006). [6] H. Holma and A. Toskala, "WCDMA FOR UMTS". Third Edition, (2004).