1

Inter-System Handover Parameter Optimization Christopher Brunner1, Andrea Garavaglia2 , Mukesh Mittal1, Mohit Narang1, Jose Vargas Bautista1 1

2 QUALCOMM Incorporate QUALCOMM GmbH 5775 Morehouse Drive Nordostpark 89 San Diego - CA 92121 - USA 90411 Nuremberg - Germany {chris, andrea, mmittal, mnarang, josev}@qualcomm.com

Abstract — With WCDMA networks being deployed in Europe and throughout the world, one of the big challenges is to make cell reselection and handover between WCDMA and GSM work smoothly. In this paper, we highlight different strategies for inter-system handover, discuss the impact of intra-frequency handover parameters on inter-system handover performance, and study inter-system handover parameter settings by processing field measurement data collected in different networks. Optimization trade-offs are illustrated with examples and a recommended set of parameters that led to significant reductions in call drop rates in commercial WCDMA networks is provided. Index Terms—inter-system handover, parameter optimization, simulation, WCDMA, UMTS

I. INTRODUCTION Initial deployment of 3G WCDMA is focusing mainly on city centers and business districts where higher subscriber density and concentration of demand for services beyond voice allow an early return on investment. As a result, WCDMA coverage is available only in islands as opposed to ubiquitous GSM coverage. In addition, imperfections within WCDMA islands result in coverage holes and limited indoor coverage which is an on-going issue as load is increasing and HSDPA is being deployed on the same carrier. This drives the need for seamless transition between 3G and 2G networks, i.e., seamless inter-system handover for circuit switched (CS) services and seamless inter-system cell changes for packet switched (PS) services are required. CS and PS calls may follow different strategies for WCDMA to GSM (W-to-G) handover. Circuit switched (e.g. voice) calls need to be handed over from WCDMA to GSM reliably without link interruptions. Hence, the user equipment (UE) is switched into compressed mode (CM) where gaps in transmission and reception are created to tune away and measure other frequencies 1 . GSM to WCDMA handover for CS services may not be enabled in initial network deployments to minimize call drops for voice services. For a

1

Approaches avoiding CM have disadvantages: A second RX chain would remove the need for CM but negatively impact UE cost. In blind W-to-G HO, one GSM cell is selected based on mapping active set combinations to GSM cells. This GSM cell may not be strong enough to avoid a call drop.

PS call, a blind handover can be configured to take place shortly before the connection would drop. Alternatively, PS calls can be allowed to drop in WCDMA at the coverage boundary and to transition to the GPRS network automatically via inter-system idle mode cell reselection. This is expected to be acceptable for background services (web-browsing and ftp are less sensitive to interruptions) and would avoid the throughput degradation observed in CM while on a PS call in poor WCDMA coverage. Field tests also indicate that throughput in REL-99 is higher at the cell edge than GPRS throughput. In PS, the G-to-W transition takes place via cell reselection and should always be enabled to maximize throughput rates. In case of multi-RAB calls (UE in call on both CS and PS domain), HO strategy should follow CS needs, voice being the most important service. The sequel of this paper discusses inter-system HO parameters for CS as it is challenging to time CM and inter-system HO to maximize coverage and minimize call drops. Previous publications have discussed intra-frequency and inter-system 2 cell reselection parameter optimization, cf. [3], [4], [5]. The need for CM triggering based on CPICH Ec/No and RSCP measurement quantities is motivated in [1] and [2]. Section II discusses the W-to-G inter-system handover mechanism for CS. Section III proposes intra-frequency HO parameters to reduce call drops during CM. Tradeoffs and recommendations for inter-system handover parameter settings are discussed based on simulations driven by field measurements of the RF channel. II. INTER-SYSTEM HANDOVER MECHANISMS A. CM and Inter-System HO Triggering To trigger CM, a measurement report message (MRM) 2d or 6a is sent to the RNC. Measurement reporting is controlled by a set of rules defined in the layer 3 RRC protocol [10]. If the channel recovers before the GSM measurements are complete, the UE sends MRM 2f or MRM 6b to the network which then

2

Idle mode coverage in WCDMA is controlled by inter-system cell reselection parameter settings and should be as large as the largest connected mode service coverage while maintaining excellent call setup performance in the coverage area, cf. [1].

2 deactivates CM 3 . Event 3a (The estimated quality of the currently used UTRAN frequency is below a certain threshold and the estimated quality of the other system’s frequency is above a certain threshold) is the most commonly used event to trigger inter-system HO. The event 2d triggering condition [10] is fulfilled if

Q ≤T2d − H2d / 2

(1)

CPICH RSCP are used to track a weak uplink. For one, the uplink is coverage limited and the downlink capacity limited [1] and RSCP tracks the coverage. For two, the RSCP metric is used in radio network planning tools to capture coverage, so it is consistent to use this metric for CM triggering and intersystem HO as well. Given that the downlink is capacity limited, events 2d and 2f based on CPICH Ec/No can be used to trigger CM and inter-system HO.

holds for the duration set by the event 2d time-to-trigger parameter, where T denotes the threshold parameter, H the hysteresis parameter for event 2d, and Q is defined as follows:

UE

Network MCM: Configure Events 1E/1F/2D/2F/6A/6B

1 MRM: Measurement Report (Event 1F/2D/6A)

2

⎞ ⎛ Q = W ⋅ 10 ⋅ Log ⎜⎜ ∑ M i ⎟⎟ + (1 − W ) ⋅ 10 ⋅ LogM Best (2) ⎠ ⎝ i =1

Configure (& Activate) CM*

NA

3 MCM: Configure Events 3A/3C & include inter-RAT cell info (& Activate CM) MRM: Measurement Report (Event 3A/3C)

The event 2d triggered state condition is cancelled [10] if

Q>T2d + H2d / 2

Handover From UTRAN Command

(3)

holds at any measurement instance, allowing for another measurement report to be sent once (1) is fulfilled again. Here, the parameter W denotes the weighting between the best server in the active set and all cells in the active set. Moreover, Mi is the measurement of the i-th cell. The measurements M have been filtered with a single-pole IIR filter. Different time constants can be configured. The measurement quantity can either be pilot Ec/No or RSCP [8]. Two events can be configured to capture both [10]. Event 2f is opposite to event 2d. Measurement reports are sent if Q exceeds pre-configured thresholds. Otherwise, the same set of parameters is defined for event 2f [10]. Event 6a is triggered if the filtered UE transmit power exceeds a pre-defined threshold for a preconfigured duration (time-totrigger). The UE transmit power is filtered with a single-pole IIR filter for which different time constants can be configured. Event 6b is the opposite of event 6a. MRMs are sent if the UE transmit power drops below a pre-configured threshold [10]. Triggering conditions for event 3a are equivalent to event 2d except that in addition, the GSM cell has to be identified (BSIC identification and reconfirmation) and GSM has to be better than a pre-defined threshold [10]. Event 3a is triggered as soon as the first GSM cell is re-confirmed, cf. Section B. The inter-system HO decision or granting algorithm (Handover UTRAN Command) is driven by MRMs (as is CM activation and de-activation). Figure 1 shows a call flow between the UE and the UTRAN/GERAN during CS intersystem handover from WCDMA to GSM. Events 6a and 6b are ideal to track a weak uplink (and trigger CM), because they accurately reflect fast fading and noise rise in the uplink. In general, however, events 2d and 2f based on 3 Instead of using events 2d and 2f to activate and de-activate CM, 1f and 1e can be used, respectively. The events differ since 2d and 2f apply to the active set where as 1f and 1e apply to individual cells (in the active set) [10].

4 5

Perform GSM Handover Sequence Handover Complete (to GERAN)

6 Figure 1: W-to-G Handover Call Flow Diagram for CS B. Measurements in Compressed Mode For CS services, CM is based on the spreading factor reduction method (“SF/2”). The spreading factor is halved during the compressed radio frames and more transmit power is used to keep quality targets [7]. To prevent hard blocking, a second scrambling code tree can be used during CM at the expense of increased downlink interference. Measuring a GSM/GPRS cell may involve up to three Transmission Gap Pattern Sequences (TGPS): ™ GSM Carrier RSSI Measurements ™ GSM Initial BSIC Identification ™ GSM BSIC Re-confirmation The duration of the RSSI measurements depends on the Transmission Gap Length (TGL), the frequency of the gaps, and the size of the GSM Monitored Set List. RSSI measurements are completed before starting BSIC identification and re-confirmation. In a gap of TGL=7 slots, the UE should obtain 6 GSM carrier RSSI measurements. To meet the measurement accuracy requirements, each GSM carrier is measured three times [9]. If the transmission gap pattern length (TGPL) for RSSI measurements corresponds to 8, an RSSI measurement gap becomes available every 8*10ms. Here, the RSSI measurement duration corresponds to:

TRSSI (N) = 8⋅10ms⋅ N⋅ 3/ 6

(4)

The number of GSM neighbors is denoted by N. So for 32 and 8 GSM neighbors, the measurement duration equals 1.28s and 0.32s, respectively.

3 Next, the UE identifies the BSIC of up to 8 GSM cells. The UE shall use all the gaps in the pattern with purpose “GSM Initial BSIC Identification” to identify the BSIC of a single cell before attempting to decode the BSIC of the next cell. The UE shall update the timing of up to 8 identified GSM cells using the pattern with purpose “GSM BSIC reconfirmation”. The BSIC of a cell is considered verified if the BSIC is identified and re-confirmed. Although BSIC verification is a lengthy process, it allows the UE to acquire the timing of the GSM cells and increases the success rate of inter-system HO. The time between starting CM measurements and triggering an Event 3a (triggering conditions are fulfilled) can be computed as follows:

T(N) = TRSSI (N) +TID +TRE

(5)

Because an event 3a can be sent after the first BSIC reconfirmation has taken place which can take place before the BSIC identification of other cells have been completed, only the RSSI measurement duration depends on N. In the best case, identification and reconfirmation can be completed in one cycle (of 80ms if the TGPL corresponds to 8), in the worst case, several cycles are required. Note that multiple cycles are required if the information to identify and reconfirm the BSIC is not being transmitted during the WCDMA gap.

Figure 2: Potential Impact of Active Set Updates in CM C. Impact of Active Set Updates in CM At call setup time, the network sends an MCM to initially configure inter-system measurements. Inter-system measurements need to be re-configured if an active set update takes place. This is done by sending an MCM to modify the GSM neighbor list [10]. If an MRM 3a is sent before the MCM (to modify) is received at the UE, the target GSM cell index within the MRM 3a is based on the old GSM NL list [10]. In REL-99, the network may issue a “HandoverfromUTRAN” command to the wrong GSM cell since the network is already using the new GSM NL list, causing the call to likely drop, cf. Figure 2. This mismatch

should only occur if event 3a is received at the network after the MCM (to modify) has been transmitted by the network but has been transmitted by the UE before the MCM (to modify) has been received at the UE. If bad channel conditions which are not unusual at the coverage boundary delay RLC retransmissions, the mismatch becomes more probable. In REL5 and onwards, an optional field (“Inter-RAT cell info indication”) has been added to the MRM and the cell info list containing the GSM NL. The use of this field in MRM and cell info list allows the network to identify a mismatch. III. SIMULATIONS, TRADEOFFS, AND RECOMMENDATIONS A. Intra-Frequency Handover Parameter Settings To minimize call drops caused by ASET updates in CM, intrafrequency HO parameters can be set to reduce ASET updates. Events 1a (add cell to the active set), 1b (drop cell), and 1c (swap into and out of the active set) are used in intrafrequency handover. Triggering conditions are explained in [10]. Four primary metrics are relevant in intra-frequency HO: ™ Mean measurement reporting rate: Related to mean active set update rate. A reduction leads to fewer active set updates, i.e., fewer call drops during inter-system HO due to the mismatch of GSM cells previously outlined. ™ Mean active set size: if reduced, less risk of hard blocking ™ Call quality: Captured by the tail of the active set combined pilot (CPICH) Ec/Io distribution. The lower 5%-tile relates to the transmit power exceeding its maximum threshold (potentially causing call drops). ™ Air interface capacity: Downlink transmit power summed over the active set cells is normalized with respect to the transmit power that would be required if the active set size were one and is then inverted 4 . To illustrate the impact of different system parameter sets, these metrics are determined for sets listed in Table 1 based on a drive route which covers different scenarios (pilot pollution, quickly changing best servers, good RF). The metrics have been computed by passing RF measurements into a handoff emulator, cf. [1], [5]. In addition, the following settings were used: filter coefficient with a time constant of 458ms (K=3), deactivation threshold of 2, replacement activation threshold of 3, reporting amounts for event 1a and 1c of infinity, reporting intervals of 1s for 1a and 1c, and time to trigger (TTT) for 1c of 0.1s. In Figure 3, call quality is plotted against capacity. For the same TTT 1b parameter, call quality and capacity trade off. If the TTT 1b is increased, the capacity is negatively impacted if the call quality stays the same. Weak cells remain in the active 4 In the results shown below, a macro-diversity gain of zero (pure line-ofsight scenario) is assumed. Therefore, macro-diversity gains are zero and the capacity is lost, not gained (indicated by a normalized capacity below 1). Transmit power increases if the path-loss differences between UE and active set sites increases. Transmit power would decrease with increasing macrodiversity gain.

4 set longer, contribute less to the received CPICH Ec/No combined across the ASET, and interfere more with other users. Note as well that the performance of sets 1 and 3 as well as 2 and 5 is similar. In both cases, the parameter W has been lowered to 0 and the reporting ranges, R, have been reduced. TABLE 1: PARAMETER SETS set #

W

R, 1a [dB]

R, 1b [dB]

H, 1c [dB]

TTT, 1a [s]

TTT, 1b [s]

1

1

5.0

7.0

3

0

0.1

2

1

4.0

6.0

3

0

0.1

3

0

4.0

6.0

3

0

0.1

4

0.5

4.0

6.0

3

0

0.1

5

0

3.0

5.0

3

0

0.1

6

0

2.5

4.0

2

0.1

0.64

7

0.5

2.5

4.0

2

0.1

0.64

8

0

3.0

4.5

2

0.1

0.64

9

0

2.5

4.0

2

0

0.64

10

0

2.5

4.0

1

0.1

0.64

11

0

2.5

4.0

2

0.1

0.32

12

0

3.0

4.5

3

0.0

0.64

UE from unnecessary CM activations and fast fades, hence improving call retention and WCDMA coverage. The downside is a small hit on capacity because weak cells that remain in the active set contribute marginally to the user of interest but increase interference levels seen by other uesrs, cf. Figure 3. Since signal strength drops logarithmically over distance, pilot pollution is more likely to occur at the WCDMA network boundary (and in outdoor coverage holes within WCDMA) than at outdoor-to-indoor coverage boundaries. To further minimize the risk of call drops during CM, appropriate site choice and RF configuration help. For instance, a road cutting through the WCDMA boundary should be covered by a dominant cell.

W=1

W=0 TTT 1b = 0.1s

Figure 4 depicts mean MRM rate versus mean active set size. Contrary to Figure 3, the metrics for parameter sets 1 and 3 as well as 2 and 5 differ strongly. The MRM rate is significantly higher for sets 1 and 2 which are characterized by W=1 at the benefit of a slightly reduced mean active set size. Moreover, a large TTT 1b significantly reduces the MRM rate.

TTT 1b = 0.64s

Figure 4: Mean MRM rate versus mean active set size TTT 1b = 0.1s

TTT 1b = 0.64s

B. CM and Inter-System HO Parameter Settings Parameters for CM activation and inter-system HO from WCDMA to GSM for CS services should be set such that: ™ ™ ™ ™ ™

Call drops (see appendix for model) are minimized. WCDMA coverage is maximized. Signaling to activate/de-activate CM is minimized. Hard blocking related to CM (SF/2 mode) is minimized. Loss in air interface capacity related to CM in minimized.

In the simulations, WCDMA coverage is captured by additional time spent on the drive route. Since CM related hard blocking and loss in air interface capacity due to CM is related to time in CM, the latter is logged in the simulations. Figure 3: Call quality (tail of the active set Ec/Io distribution) versus capacity Hence, our recommendation is to choose a large TTT 1b and W=0 to minimize the active set update rate (in CM). In addition, cells fluctuate in pilot polluted areas. If the TTT 1b is small, they are dropped and cannot be added again immediately 5 . So a larger TTT 1b will also help protect the 5

The delay between sending MRMs and receiving ASET update messages corresponds to 300-600ms.

The following tradeoff is fundamental for inter-system HO: ™ To maximize coverage, the CM threshold is set higher than the inter-system HO threshold. ™ To minimize time in CM, the CM threshold is set lower than the inter-system HO threshold leading to immediate inter-system HO after completing the CM measurements. To better illustrate this tradeoff, simulation results are shown for different parameter sets, cf. Table 2, applied to a WCDMA RF channel measured in the field (crossing the WCDMA

5 network boundary). Parameters that remain constant across the sets are W=0, TTT 2d = 0.32s, and TTT 2f = 1.28s. CM activation delay was set to 1.2s, inter-system HO delay (between sending an MRM 3a and carrying out inter-system HO) to 1.2s, and the CM measurement duration to 5s. For the sake of simplicity, a constant minimum CM measurement duration has been assumed instead of more accurately modeling the varying CM measurement durations according to (5). If only one measurement quantity can be configured (a limitation still seen in infrastructure implementations), the CPICH Ec/No should be selected because it better represents both up- and downlink channels. The CPICH RSCP metric mainly captures coverage, i.e., the uplink, and is less versatile. TABLE 2: PARAMETER SETS set #

T, 2d [dB]

T, 2f [dB]

H, 2d [dB]

H, 2f [dB]

T, 3a [s]

TTT, 3a [s]

A

-11

-9

0

0

-13

0.1

B

-11

-10

2

2

-9

0.0

updates reduces the probability of dropped calls due to the GSM neighbor mismatch issue outlined in Section II.C. To better describe the impact of inter-system handover thresholds (especially the event 2d threshold), the next plots show call drop ratio and coverage as a function of RSCP and Ec/No threshold settings. Here, we assume that the infrastructure supports the configuration of both measurement quantities. The other parameters are chosen as above. To obtain these results, RF measurements captured during drive testing leaving WCDMA coverage were passed on to an emulator. The loading conditions in the network were low. The call drop ratio is computed by passing multiple series of RF measurements on to the emulator and applying the call drop model outlined in the appendix. In all call drops observed below, the uplink break first which is expected given the low loading conditions. 100 EcNo EcNo EcNo EcNo

90 80

Th 2d Th 2d Th 2d Th 2d

= = = =

-9 dB -11 dB -13 dB -15 dB

call drop ratio [%]

70 60 50 40 30 20 10 0 -120

-115

-110

-105

-100

RSCP threshold Event 2d [dBm]

Figure 7: Call drop ratio as a function of Event 2d RSCP and Ec/Io thresholds and a minimum CM measurement duration of 5s (no load, coverage boundary) Figure 5: Inter-System HO Event Triggering with Set A A dditional W CDMA Distance compared to E2d, E3a = -97 dBm, -93 dBm [m/call]

(Scenario = IRA THO FGR dge - GSM T = 5, Th3a - Th2d = 4 [dB]) V

E

M

1600 E cIo Th2d = -9 [dB] E cIo Th2d = -11 [dB ] E cIo Th2d = -13 [dB ] E cIo Th2d = -15 [dB ] 1500

Additional WCDMA Distance [m]

1400

1300

1200

1100

1000

900 -118

Figure 6: Inter-System HO Event Triggering with Set B The comparison between Figure 5 and Figure 6 shows a significant reduction of unnecessary CM activations. In addition to reduced signaling, the probability of hard blocking is reduced and air interface capacity is increased. Most importantly, a significant reduction of number of active set

-116

-114

-112

-110 -108 -106 RSCP Threshold for starting CM (event 2d) [dBm]

-104

-102

-100

-98

Figure 8: Gain in coverage measured in distance along the drive route as a function of Event 2d RSCP and Ec/No thresholds for a minimum CM measurement duration of 5s (no load, coverage boundary) Figure 7 and Figure 8 indicate that the best parameter setting for this scenario would be an Ec/No threshold of -15 dB and an RSCP threshold of -109 dBm. The next two plots are based on a load of 60%. Here, to achieve a call drop ratio of zero,

6 the Ec/Io threshold can be set to -15dB and the RSCP threshold to -113 dBm. The RSCP threshold can be relaxed because now the downlink breaks first (no link imbalance). 100 EcNo EcNo EcNo EcNo

90 80

Th 2d Th 2d Th 2d Th 2d

= -9 dB = -11 dB = -13 dB = -15 dB

What is the impact of ASET size on call drop probability? Since macro-diversity gain is small on the uplink, a larger ASET does not significantly reduce the probability of dropped calls if due to insufficient UE TX power. A larger ASET size reduces the call drop probability if the cause is insufficient DCH power per link on the downlink 6 . Since downlink channel estimation takes place separately for each cell, a larger ASET size does not prevent a dropped call caused by CPICH channel estimation failure.

call drop ratio [%]

70 60 50 40 30 20 10 0 -120

-115

-110

-105

-100

RSCP threshold Event 2d [dBm]

Figure 9: Call drop ratio as a function of Event 2d RSCP and Ec/Io thresholds and a minimum CM measurement duration of 5s (load, coverage boundary) Additional WCDMA Distance compared to E2d, E3a = -97 dBm, -93 dBm [m/call]

(Scenario = IRATHOV FGREdge - GSMMT = 5, Th3a - Th2d = 4

1400

1200

1000 Additional WCDMA Distance [m]

coverage limited scenario) ™ per link DCH TX power exceeding threshold (downlink – typical for capacity limited scenario) ™ channel estimation failing because of too weak common pilot (downlink – typical for pilot polluted scenario)

A radio link failure takes place, according to the standard [10], if, in CELL_DCH state, the timer T313 expires. T313 starts if the last N313 consecutive CRCs of the received transport blocks are incorrect and were incorrect for the last 160ms. The N313 and T313 default settings are 20 and 3s, respectively. For simplicity, the emulator assumes a call drop if, for a duration of 3s, path-loss forces the max. UE TX power to exceed 24 dBm, CPICH Ec/No combined across the ASET drops below -18 dB, or the strongest ASET CPICH Ec/No drops below -20 dB. Operation at -20dB is required [9].

800

REFERENCES 600

[1]

EcIo Th2d = EcIo Th2d = EcIo Th2d = EcIo Th2d =

400

-9 [dB] -11 [dB] -13 [dB] -15 [dB]

200

0 -118

-116

-114

-112

-110 -108 -106 -104 RSCP Threshold for starting CM (event 2d) [dBm]

-102

-100

-98

Figure 10: Gain in coverage measured in distance along the drive route as a function of Event 2d RSCP and Ec/No thresholds for a minimum CM measurement duration of 5s (load, coverage boundary) The investigation above was limited to the WCDMA coverage boundary. Two more scenarios need to be kept in mind: outdoor to indoor transition and coverage hole. IV. CONCLUSIONS We have proposed intra-frequency HO parameters as well as inter-system HO parameter settings designed to optimize inter-system HO performance. These recommendations helped to significantly reduce call drop rates in commercial networks.

WCDMA Compressed Mode Triggering Method for IRAT Handover, Zhang Zhang, WCNC 2004, March 2004 [2] D.Lugara, J.Tartiere, L.Girard, “Performance of UMTS to GSM handover algorithms”, proceedings of PIMRC 2004 [3] Dino Flore, Christopher Brunner, Francesco Grilli, and Vieri Vanghi, “Intra-Frequency Cell Reselection Parameter Optimization in UMTS”, proceedings of ISWCS 2005 [4] Andrea Garavaglia, Christopher Brunner, Dino Flore, Ming Yang, and Francesco Pica, “Inter-System Cell Reselection Parameter Optimization in UMTS”, proceedings of PIMRC 2005 [5] WCDMA (UMTS) Deployment Handbook, Christophe Chevallier, Christopher Brunner, Andrea Garavaglia, Kevin Murray, Kenneth Baker, John Wiley & Sons Ltd, estimated publication date: July 2006. [6] 3GPP TS 25.211 “Physical channels and mapping of transport channels onto physical channels (FDD)” [7] 3GPP TS 25.212 “Multiplexing and channel coding (FDD)” [8] 3GPP TS 25.215 “Physical layer; Measurements (FDD)” [9] 3GPP TS 25.133 “RRC Requirements for support of radio resource management (FDD)” [10] 3GPP TS 25.331 “RRC Protocol Specification”

APPENDIX A: CALL DROP MODEL Call drops are due to either the up- or the downlink breaking (in addition to the GSM NL mismatch explained in Section II.C which is not taken into account here). More specifically, call drops occur because of ™ UE TX power exceeding threshold (uplink – typical for

6 While the uplink can be affected if additional power required in compressed mode (SF/2) is not available, the DL DCH power cap is doubled for the compressed frames.