Handover Paper Final

Mobility Management and Capacity Analysis for High Speed Downlink Packet Access in WCDMA Klaus I. Pedersen, Antti Toskala, Preben E. Mogensen Nokia Networks Email: [email protected] Abstract—In this paper we discuss two options for supporting mobility management for HSDPA-users. Especially issues related to direct change of the serving HS-DSCH cell are addressed, where a handover from a HSDSCH in the source cell to a HS-DSCH in the target cell is made. It is concluded that change of the serving HS-DSCH cell without any intermediate channel switching to DCH seems to be a promising solution from a performance point of view as it allows to have full HSDPA cell coverage and high capacity gain from introducing HSDPA. Uplink HS-DPCCH coverage is also addressed for users with their DCH in soft handover mode.

I. I NTRODUCTION High speed downlink packet access (HSDPA) was introduced in 3GPP UTRAN Release 5 [1]. HSDPA includes a number of downlink performance enhancing features such as adaptive modulation (QPSK/16QAM) and coding, a new physical layer retransmission mechanism with soft combining of retransmissions (Hybrid ARQ), and location of the medium access control layer (called MAC-hs) in the base station (NodeB). Location of the MAC-hs in the Node-B implies that data for HSDPA-users are being buffered in the Node-B, and also the HSDPA packet scheduler is in the Node-B. The shared transport channel for HSDPA is the high speed downlink shared channel (HS-DSCH). Not many studies have been published where mobility management for HSDPA-users is addressed in details. Mobility management, or equivalently handover management, is quite different for HSDPA compared to dedicated channels (DCH) as specified in Release’99. Mainly because the HS-DSCH is only transmitted from one cell to the user (i.e., soft handover transmission from multiple cells as for the DCH is not supported) and secondly, because the system architecture for HSDPA is more decentralized so functions are moved from the central radio network controller (RNC) to the Node-B. The architecture for HSDPA is sketched in Fig. 1, where part of the protocol stack and the functional split between the serving RNC, the Node-B, and the user equipment (UE) is illustrated. The choice of handover strategy for HSDPA-users also influences the cell coverage and capacity of HSDPA. We will therefore discuss two different handover algorithms for HSDPA-users. In this context, issues related both to downlink and uplink coverage, as well as capacity considerations are addressed. The paper is organized as follows: Two different options for handling HSDPA mobility are discussed in Section II. Section III includes results for the soft handover statistics, uplink coverage considerations for HSDPA-users in the soft handover area, and capacity results. Concluding remarks are in Section IV.

II. M OBILITY MANAGEMENT FOR HSDPA USERS A. Mobility options Let us assume that an HSDPA-user is receiving data on the HS-DSCH in the source cell as indicated in Fig. 2. As there can only be one cell transmiting data on the HS-DSCH to the user, this cell is called the serving HS-DSCH cell. If the user moves to a new target cell, there are basically two different methods for implementing mobility management. One option is first to make a channel switching from HS-DSCH to DCH when the user moves into the soft handover area and then potentially again make channel switching from HS-DSCH to DCH when it exits the soft handover area. In this context, the soft handover area is defined as the area where the user is having an active set size larger than one for its DCH. However, by using this method users in the soft handover area are excluded from using the HS-DSCH and are therefore not able to benefit from the potential advantages that HSDPA offers. An alternative option for handling mobility is therefore to make a direct handover from the HS-DSCH in the source cell to the HS-DSCH in the target cell. The latter option is referred to as changing the serving HS-DSCH cell. The advantage of changing the serving HSDSCH cell without using an intermediate channel switching to DCH is that full HSDPA coverage is thereby supported, i.e., all HSDPA-users in the network can receive data on the HSDSCH independently of their position. Notice from Fig. 1 that the serving RNC is responsible for both handover control and channel type selection. Serving HS-DSCH cell Soft handover area (potential channel switching to DCH)

Active set size>1 User Source cell

The user is moving from the source cell into the target cell

Target cell

Fig. 2. An HSDPA-user that is receiving data on the HS-DSCH is moving from the source cell (currently serving HS-DSCH cell) into a new target cell.

In order to compare the two considered candidate options for HSDPA mobility handling, we will compare the methods in terms of performance and also take a closer look at the implications of performing a change of the serving HS-DSCH cell. Due to the de-centralized HSDPA architecture, intra and in-

0-7803-8521-7/04/$20.00 (C) 2004 IEEE

Node-B

Serving RNC

User equipment (UE) RLC

Handover Control

Data MAC-hs: - MAC-hs packet scheduler - Data buffer for each user - Hybrid ARQ manager - MAC-hs flow control - etc.

RLC Channel type selection

MAC-d HS-DSCH Frame Protocol

Iub

MAC-d

HS-DSCH Physical Layer Frame Protocol

Uu (air interface)

MAC-hs: - Hybrid ARQ manager - Re-ordering - etc. Physical Layer

Fig. 1. Simplified block diagram of the functional split between the serving RNC, the Node-B, and the user equipment (UE) for HSDPA.

ter Node-B change of the serving HS-DSCH cell are discussed separately.

cell. Serving RNC

B. Inter Node-B HS-DSCH cell change Both synchronized and asynchronous change of the serving HS-DSCH cell is supported in Release’5. However, in this study we focus on synchronized cell changes. The decision to make an inter Node-B change of the serving cell is made by the serving RNC and is typically triggered by a measurement from the user, such as measurement event 1d, which is called measurement event for the best serving HS-DSCH cell [7]. This measurement event reports the best serving HS-DSCH cell to the serving RNC based on a measurement of the primary common pilot channel (P-CPICH) for the potential candidate cells for serving HS-DSCH cell. Once the serving RNC decides to initiate an inter Node-B change of the serving HS-DSCH cell as indicated in Fig. 3, a synchronized radio link reconfiguration prepare message is sent to the drifting RNC and the Node-B that controls the target cell, as well as a radio resource control (RRC) physical channel reconfiguration message to the user1 . Among others, these messages specify the time where the actual cell change is made, which is typically 300-500 ms from the time where the serving RNC decides to make the cell change. However, until the time where the actual cell change is made, the source cell is still allowed to transmit to the user on the HS-DSCH. At the time where the cell change is implemented, the MAC-hs for the user in the source cell is reset, which basically means that any buffered protocol data units (PDU) for the user are deleted, including the pending PDUs in the Hybrid ARQ manager. At the same time index, the flow control unit in the MAC-hs in the target cell starts to request PDUs from the serving RNC. It is therefore desirable to control the amount of buffered PDUs in the source cell, so that only few PDUs are buffered at the time where the actual cell change is made. This can for instance be achieved by stop forwarding PDUs from the RNC to the source cell at the time where the RNC decides to make the cell change. Provided that the source cell is able to transmit the buffered PDUs within the 300-500 ms time-window before the actual cell change is made, no PDUs are deleted in the source 1 For a detailed signaling flow diagram for change of the serving HS-DSCH cell, see the 3GPP specifications [9]-[10].

Drifting RNC

Iur

Iub

Iub

Node B 1

2

Node B

Source cell 3

1

2

3

Target cell The user is moving to cell #1 under another Node B User

Fig. 3. Sketch of an HSDPA user that is making an intra Node-B change of the serving HS-DSCH cell.

However, if it anyway happens that PDUs are deleted in the source cell prior to the change of the serving HS-DSCH cell, then these PDUs must be recovered by higher layer retransmissions such as RLC retransmissions. When the RLC protocol realizes that the PDUs it has originally forwarded to the source cell are not acknowledged, it will initiate retransmissions, which basically means forwarding the same PDUs to the target cell that was deleted in the source cell. In order to reduce the potential PDU transmission delays during this recovery phase, the RLC protocol at the user can be configured to send an RLC status report to the UTRAN at the first time incident after the serving HS-DSCH cell has been changed [8]. This implies that the RLC protocol in the RNC can immediately start to forward the PDUs that were deleted in the source cell prior to the HS-DSCH cell change. C. Intra Node-B HS-DSCH cell change A synchronized intra Node-B serving HS-DSCH cell change is similar to the synchronized inter Node-B serving HS-DSCH cell change. The only difference is that potentially buffered PDUs in the source cell’s MAC-hs are not necessarily deleted at the time where the actual cell change is made. Provided that the Node-B supports MAC-hs preservation, the buffered PDUs in the source cell are moved to the target cell’s MAC-hs, so that no PDUs need to be recovered by higher layer retransmission. In this respect, an intra Node-B serving HS-DSCH cell change

0-7803-8521-7/04/$20.00 (C) 2004 IEEE

is simpler to control, compared to an inter Node-B cell change.

a relatively high transmit power when the user is in soft handover mode [6].

III. A NALYSIS HS-DPCCH with L1 Ack/Nack/CQI

A. Soft handover statistics

Percentage of users in different soft handover states [%]

Fig. 4 shows the probability for different soft handover states, such as the probability of being in 2-way softer handover between cells on the same Node-B, 2-way soft handover between cells on different Node-Bs, etc. These results are obtained from Monte-Carlo network simulations, assuming a standard 3-sector network topology where each cell is using a 65 degree antenna. Users in the network are assumed to be uniformly distributed. A simple single slope path loss model is applied with a path loss exponent of 3.5. Shadow fading is modelled with a lognormal random variable with 8 dB standard deviation. The soft handover state of the different users depends on the relative received quality of the P-CPICH from the different cells and the soft handover window which triggers addition of a cell to the users active set. It is observed that for a soft handover window of 4 dB, 5% of the users are in 2way softer handover. Hence, these 5% of the users are likely to require an intra Node-B change of the serving HS-DSCH cell. However, some of users in 3-way soft handover may also be candidates for intra Node-B serving HS-DSCH cell change if two of the soft handover legs are associated with the same Node-B. 50

Users with active set size>1

45 40 35 30

Users in 3-way soft handover

25 20 Users in 2-way soft handover (under different Node-Bs)

15 10 5 0 2.0

Users in 2-way softer handover (on the same Node-B) 2.5

3.0

3.5 4.0 4.5 Soft handover window [dB]

5.0

5.5

6.0

Fig. 4. Soft handover statistics from a homogeneous 3-sector network.

B. Uplink coverage considerations It only makes sense to allow HSDPA-users in the soft handover area to receive data on the HS-DSCH if there is sufficiently good uplink coverage for the high speed dedicated physical control channel (HS-DPCCH) that carries the Layer1 Ack/Nack’s and the channel quality indicator (CQI) report. The HS-DPCCH is power controlled relative to the uplink DCH that the user also transmits. However, as the DCH is in soft handover mode while the HS-DPCCH is only decoded at the serving HS-DSCH cell, this creates some potential power control imperfections, which implies that the HS-DPCCH requires

Return channel for TCP ack’s (DCH) 12.2 kbps AMR speech (DCH) User equipment

Node-B TCP download on HSDPA (HS-DSCH) 12.2 kbps AMR speech (DCH)

Fig. 5. Sketch of uplink and downlink channels for the considered uplink coverage example.

Let us therefore consider the example pictured in Fig. 5, where it is assumed that the user is in soft handover mode while downloading data on the HS-DSCH over the transmission control protocol (TCP) in parallel with an ongoing 12.2 kbps speech call on DCH. In the uplink, the user also transmits a 12.2 kbps speech call on DCH in parallel with a DCH return channel for TCP Ack’s and the HS-DPCCH. Assuming that a TCP Ack of 40 bytes is sent on the uplink DCH return channel for every received TCP packet of 1500 bytes on the DS-DSCH (i.e., header compression is not being used), we can calculate the required bandwith of the DCH return channel. For a HSDPA-user on the cell edge that is receiving data on the HS-DSCH with 500 kbps, the required uplink bandwidth of the DCH return channel is thus 13.4 kbps, so it is assumed in the following that the bit rate of the DCH return channel is set at 16 kbps. In addition to the channel pictured in Fig. 5, there is also a 3.4 kbps signaling radio bearer on DCH in both the uplink and downlink. Let us assume that the network has been dimensioned for 64 kbps uplink data coverage on DCH for cases with no HSDPCCH transmission, for a maximum user transmit power of 125 mW. Given this starting point, the uplink link budget is known so the required transmit power for the uplink speech channel and the DCH return channel can be found, and therefore also the available power for HS-DPCCH transmission. Given these assumptions, link level simulations have been conducted for a user in 3-way soft handover. The link level simulations have been repeated for different power offsets between the HS-DPCCH and the uplink dedicated physical control channel (DPCCH) using different degrees of Ack/Nack repetition for the HS-DPCCH. The link level simulations assumes a Node-B with two receive antennas per cell and a standard Rake receiver. It is found that the HS-DPCCH detection error probabilities listed in Table I can be fulfilled for a user in soft handover provided that the Ack/Nack is transmitted on the HS-DPCCH with repetition two, i.e. is sent twice before being decoded [11]. The latter is possible without reaching the maximum user transmit power. Hence, under the discussed assumptions, uplink coverage is not a problem for HSDPA users in the soft handover area. Notice that without transmission of the uplink speech channel, the bit rate of the uplink DCH can be much higher than in the

0-7803-8521-7/04/$20.00 (C) 2004 IEEE

TABLE I U PLINK HS-DPCCH ERROR DETETCTION PROBABILITIES .

Pr(Ack|N ack) Pr(N ack|Ack) Pr(Ack|DT X)

User in soft handover 0.01% 1% 4%

User in non-soft handover 0.01% 1% 1%

The HS-DPCCH detection error probabilities listed in Table I for users in non-soft handover mode can be achieved without using Ack/Nack repetition as there are no HS-DPCCH power control problems for such users. Furthermore, in the special case where the user is in softer handover, the HS-DPCCH can be received at both cells in the user’s active set and thereby benefit from the same softer handover diversity combining as the DCH so that Ack/Nack repetition is not needed in this special case. However, minimizing the HS-DPCCH transmit power allows more power for DCH transmission and therefore higher uplink DCH data rates while at the same time downloading data on HS-DSCH. C. Downlink capacity considerations Let us first estimate the HSDPA bearer gain over DCH for a user with an active set size of one and two. The HSDPA bearer gain is defined as PDCH (1) W = PHSDP A where PDCH is the average require transmit power to serve a user on DCH with a given average bit rate and PHSDP A is the average required power to serve a user on HSDPA with the same bit rate. Notice here that PHSDP A includes the average transmit power used for the user on both the HS-DSCH, the high speed shared control channel (HS-SCCH), and associated DPCH. The associated DPCH is assumed to only carry Layer-3 signaling with a bit rate of 3.4 kbps. As the HSDSCH is assigned constant transmit power and is time shared between multiple HSDPA-users, the average HS-DSCH transmit power used to serve the user with a given bit rate depends on the scheduling frequency of the user, and therefore also on the MAC-hs packet scheduling strategy. Here we will consider two cases; (i) Blind scheduling where the user is scheduled independent of the radio channel conditions and (ii) Intelligent scheduling where the user is only scheduled during good fast fading conditions. Hence, the scheduling frequency is higher for blind scheduling compared to intelligent scheduling. As an example, the popular proportinal fair (PF) scheduler [3]-[4] can be characterized as an intelligent scheduler, while the simpler sequential round robin (RR) scheduler belongs to the class of blind schedulers. Results for (1) at a target bit rate of 64 kbps are obtained from Monte-Carlo simulations, where link adaptation and Hybrid ARQ with chase combining is simulated for the HS-DSCH [2]. The HS-DSCH link adaptation algorithm is assumed to

perfectly track the radio channel quality at the HSDPA-user. The HS-SCCH is assumed to be power controlled with a target BLER of 1%. The DCH is simulated with both inner loop fast power control every slot and an outer loop power control algorithm. The radio channel between the Node-B and the user is modeled according to the ITU Vehicular-A power delay profile. Results for the HSDPA bearer gain over DCH are plotted in Fig. 6 as a function of the G-factor for a user with an active set size of one. The G-factor is defined as the ratio between the average experienced own cell interference and other cell interference at the user. It is observed that the HSDPA bearer gain is on the order of 2.7 for intelligent scheduling, while it equals 1.2-1.4 for blind scheduling. For both schedulers, the HSDPA bearer gain increases slightly as function of the G-factor. 3.0 2.8 Intelligent scheduling

Bearer gain of HSDPA over DCH [-]

considered example.

2.6 2.4 2.2 2.0 1.8 1.6 1.4 Blind scheduling

1.2 1.0 0

2

4

6 G-factor [dB]

8

10

12

Fig. 6. Estimated HSDPA bearer gain over DCH for a bit rate of 64 kbps. The active set size equals one.

Fig. 7 shows the HSDPA bearer gain over DCH for a user in 2-way soft handover. Hence, the DCH is in soft handover mode and is transmitted from two cells with equal power. The fast fading in the two soft handover legs is assumed to be uncorrelated. The HS-DSCH and HS-SCCH is only transmitted from one of the cells. The results are presented for an average branch power ratio of 0 dB and 3 dB between the two soft handover legs. Notice that the HSDPA bearer gain is slightly lower for a user in soft handover compared to a user in non-soft handover for the case where the branch power ratio equals 0 dB. This behavior is observed because the DCH benefits from the additional diversity that soft handover provides, and therefore the HSDPA bearer gain over DCH decreases. However, for a branch power ratio of 3 dB, the aforementioned soft handover gain on the DCH turns into a loss since the DCH is transmitted with equal power from the two cells even though there is an unbalance of 3 dB between the soft handover legs. The latter is the reason for the higher HSDPA bearer gain over DCH when the branch power ratio is increased. Given the results for the HSDPA bearer gain over DCH, the HSDPA cell capacity gain over a cell with only DCH-users can

0-7803-8521-7/04/$20.00 (C) 2004 IEEE

3.0

dover area are using the DCH, while users in the rest of the network are using the HS-DSCH. For this case, GainP F = 1.96, which indicates a reduced HSDPA cell capacity gain if mobility for HSDPA users is supported via channel switching to DCH as the users move through the soft handover area. The latter result assumes PF MAC-hs scheduling. However, if the soft handover area is made smaller so it covers less than 40% of the users, the relative loss in HSDPA cell capacity from not supporting users in the soft handover area on HS-DSCH becomes much less.

Bearer gain of HSDPA over DCH [-]

2.8 2.6 2.4 Intelligent scheduling 2.2

Branch power ratio between the soft handover legs:

2.0

0 dB 3 dB

1.8 1.6

IV. C ONCLUDING REMARKS

Blind scheduling

1.4 1.2 1.0 -3

-2

-1

0 G-factor [dB]

1

2

3

Fig. 7. Estimated HSDPA bearer gain over DCH for a bit rate of 64 kbps. The user’s DCH is in 2-way soft handover.

be approximated as GainF T
Q[PnonSHO WnonSHO + PSHO WSHO ],

(2)

where Q expresses the fraction of the total Node-B power that is allocated to HSDPA, PnonSHO and PSHO are the probability that a user is in non soft handover and soft handover mode, respectively. Similarly, WnonSHO and WSHO are the HSDPA bearer gain over DCH for a user in non soft handover and soft handover mode, respectively. If users in the soft handover area are not allowed to use the HS-DSCH, then WSHO = 1. Notice that (2) assumes fair throughput (FT) in cell, where all users in the cell are served with the same bit rate. If a PF MAC-hs packet scheduler [3]-[4] is used, the cell capacity gain can be expressed as (3) GainP F
C · GainF T , where the constant C expresses the additional gain from given higher bit rates to users that are close to the Node-B compared to users that are far from the Node-B. According to [1], a good approximation is to use C = 1.3. According to [5] the cell capacity gain of HSDPA over DCH in a macro cellular environment with a Vehicular-A power delay profile is estimated to be a factor 2.4 (140%), in the case where 75% of the Node-B power and 15 HS-PDSCH codes are allocated for HSDPA, i.e. corresponding to Q = 0.75. The results in [5] are obtained from quasi-static network simulations, assuming PF MAC-hs scheduling. Using (2) and (3) with WnonSHO = 2.7 and WSHO = 2.6 according to the results in Fig. 6 and Fig. 7, and PnonSHO = 0.6 and PnonSHO = 0.4, yields GainP F = 2.59. Hence, there is a reasonable good match between the results in [5] and the estimate according to (3) when comparing the two gain figures of 2.4 and 2.59. Thus, (2) and (3) provide a reasonably good first order estimate of the potential HSDPA cell capacity gain before running more accurate system level simulations. By setting WSHO = 1 we obtain an estimate of the HSDPA capacity gain over DCH in the case where users in the soft han-

Two different options for support of mobility management for HSDPA-users have been discussed, using either channel switching from HS-DSCH to DCH as the user moves close to neighboring a cell, or by using a direct change of the serving HS-DSCH cell. The algorithm for making a change of the serving HS-DSCH cell has been presented, where potential optimization options were addressed to avoid deletion of buffered PDUs in the source cell. The results for the achievable HSDPA cell capacity gain indicate that the best solution is to use direct change of the serving HS-DSCH cell during handover. However, the option that involves intermediate channel switching from HS-DSCH to DCH could also be applied with a marginal loss of the HSDPA capacity gain if the soft handover area is minimized by using a small soft handover window. The latter is possible by using different soft handover parametrization for HSDPA-users so the area for DCH switching is reduced. Notice furthermore that the potential temporary interruptions during channel switching from HS-DSCH to DCH are not considered harmful for delay tolerant packet traffic. It is furthermore demonstrated that good uplink HS-DPCCH coverage can be maintained for users in the soft handover area, even though the HS-DPCCH power control is sub-optimal, since the uplink DCH is in soft handover mode. R EFERENCES [1]

H. Holma, A. Toskala (Ed.), ”WCDMA for UMTS”, Third Edition, Wiley, 2004. [2] F. Frederiksen, T.E. Kolding, ”Performance and modeling of WCDMA/HSDPA transmission/H-ARQ schemes”, IEEE Proc. VTC, pp. 472-476, September 2002. [3] R.C. Elliott, W.A. Krzymieh, ”Scheduling Algorithms for the cdma2000 Packet Data Evolution”, IEEE Proc. VTC, September 2002. [4] J.M. Holtzman, ”CDMA Forward Link Water Filling Power Control”, IEEE VTC, pp. 1663-1667, May 2000. [5] T. E. Kolding, K. I. Pedersen, J. Wigard, F. Frederiksen, P. Mogensen, ”High Speed Downlink Packet Access: WCDMA Evolution”, IEEE Vehicular Technology Soceity News, Vol. 50, pp. 4-10, Feb. 2003. [6] C. Hermann, P. Nickel, ”High Speed Downlink in UMTS: How to Improve Packet Loss Probability in Soft Handover and at Higher Mobile Terminal Speeds”, Proc. IEEE VTC, October 2003. [7] 3GPP TS 25.331, ”Radio Resource Control (RRC); Protocol Specification”, Version 6.1.0, March 2004. [8] 3GPP TS 25.322, ”Radio Link Control (RLC) protocol specification”, Version 6.0.0, December 2003. [9] 3GPP TS 25.308, ”High Speed Downlink Packet Access (HSDPA); Overall description; Stage 2”, Version 6.1.0, March 2004. [10] 3GPP TR 25.877, ”High Speed Downlink Packet Access: Iub/Iur protocol aspects”, Version 5.1.0, June 2002. [11] TSG RAN WG1, Meeting 27 in Oulu, Finland, ”Performance of uplink HS-DPCCH in SHO with different error probability requirements”, July 2002.

0-7803-8521-7/04/$20.00 (C) 2004 IEEE