`Performance
`
`Konstantinos Dimou¹, Min Wang², Yu Yang¹, Muhammmad Kazmi¹, Anna Larmo3, Jonas Pettersson², Walter
`Muller¹, Ylva Timner²
`Ericsson Research
`Isafjordsjgatan 14 E, 146 80 Stockholm, Sweden¹, Laboratoriegränd 11, Luleå; Sweden2, Jorvas, Finland3
`konstantinos.dimou@ericsson.com; min.w.wang@ericsson.com
`
`Abstract— The 3GPP LTE system has been designed to offer
`significantly higher data rates, higher system throughput, and
`lower latency for delay critical services. This improved
`performance has to be provided and guaranteed under
`various mobility conditions. Hence, handover (HO) and its
`performance are of high importance. This paper investigates
`the performance of the handover procedure within 3GPP
`LTE in terms of HO failure rate and the delay of the whole
`procedure. System level simulations within a typical urban
`propagation environment, with different User Equipment
`(UE) speeds, cell radii and traffic loads per cell have been
`performed. The entire layer 3 signalling exchanged via air
`interface is considered in the simulations. In addition, errors
`at the Layer 1 (L1) control channels are taken into account.
`Simulation results show that the handover procedure within
`3GPP satisfies the goal of high performance mobility. Namely
`for cell radii up to 1 km and for UE speeds up to 120 km/h,
`the HO failure rate lies within the range of 0-2.2% even in
`high loaded systems. For medium and low loads even at
`speeds of 250 km/h, HO failure is below 1.3 %. In addition,
`simulation results show that the handover procedure is robust
`against L1 control channel errors.
`
`Keywords— Long Term Evolution (LTE); Handover (HO),
`Layer 1 control errors, RRC signaling, HO failure, HO delay,
`interruption time.
`
`INTRODUCTION
`I.
`3GPP has recently finalized the standardization of the
`Long Term Evolution (LTE) system within its release 8 [1],
`[2]. The radio interface is termed Enhanced UTRA (E-
`UTRA) and the radio network Enhanced UTRAN (E-
`UTRAN). Requirements for 3GPP LTE
`include
`the
`provision of peak cell data rates up to 100 Mbps in
`downlink (DL) and up to 50 Mbps in uplink (UL) under
`various mobility and network deployment scenarios.
`Namely, there is a requirement for mobility support with
`high performance up to speeds of 120 km/h [1]. An
`additional requirement is the uninterrupted provision of
`high data rates and services.
`A major difference of LTE in comparison to its 3GPP
`ancestors is the radio interface; Orthogonal Frequency
`Division Multiplexing
`(OFDM) and Single Carrier
`Frequency Domain Multiple Access (SC-FDMA) are used
`for the downlink and uplink respectively, as radio access
`schemes [3], [4]. Another distinctive characteristic of E-
`
`UTRA is the total lack of dedicated channels; hence, all the
`traffic and signalling is sent over shared channels for both
`directions of transmission-downlink and uplink. To support
`the data transmission over the physical shared channels, a
`number of L1 control channels is defined [4]. Another
`difference of LTE with the previous 3GPP releases lies in
`the radio access network architecture. The absence of a
`centralized network controller results in a distributed
`network architecture.
`In LTE only hard handover is supported. Considering
`then that handover creates an interruption time in the user
`plane, the handover performance in terms of success rate
`and delay of execution is of high importance.
`Previous papers have shown that LTE can achieve a
`good handover performance in terms of user throughput,
`handover delay, and handover failure rate [5], [6].
`However, therein the protocol procedures and control
`signaling messages exchanged during handover are not
`thoroughly considered. Moreover, the L1 control channel
`errors are not considered either.
`This paper investigates the HO performance using
`system level simulations, in a typical urban propagation
`environment. Performance results focus on handover
`failure rate and delay of the overall procedure including
`measurement reporting. Section II outlines the handover
`procedure within 3GPP LTE and describes its design
`choices. Section III presents the handover triggering
`procedure and the involved triggers, as well as their impact
`on HO performance. Section IV describes the L1 control
`channels and the possible impacts of losses in these
`channels. Sections V, VI and VII present the simulation
`model, performance metrics and results respectively.
`Finally, in section VIII results are discussed and the major
`conclusions are drawn.
`
`II. HANDOVER PROCEDURE WITHIN 3GPP LTE
`The HO procedure within 3GPP LTE is illustrated in
`Figure 1 [2]. The procedure starts with the measurement
`reporting of a handover event by the User Equipment (UE)
`to the serving evolved Node B (eNB). The UE periodically
`performs downlink radio channel measurements based on
`the reference symbols (RS); namely, the User Equipment
`(UE) can measure the reference symbols received power
`
`978-1-4244-2515-0/09/$25.00 ©2009 IEEE
`
`1
`
`SAMSUNG 1016
`
`
`
`RSRP (dBm)
`
`INTERRUPTION TIME
`
`Cell B
`
`HO
`hysterisis
`
`Event detection
`&
`Measurement
`Reporting
`
`CONNECTED TO CELL A
`
`Cell A
`
`CONNECTED
`TO CELL B
`
`Time To
`Trigger
`
`t0
`
`Measurement
`Period
`
`t3
`
`t4
`
`t5
`
`t6
`
`t7
`
`t8
`
`t9
`
`t10
`
`time
`(sec)
`
`Figure 2. Triggering of HO.
`HO hysterisis; this condition has to be satisfied for a period
`equal to the TTT.
`In Figure 2, an example of HO triggering within 3GPP
`LTE is illustrated. The event detected and reported is the
`so-called event A3 within 3GPP LTE [5]. A number of
`various handover triggering mechanisms combining these
`triggers with absolute ones may be defined. The baseline
`however of handover triggering is the one presented in
`Figure 2. It is noted here that the RSRP displayed in Figure
`2 is the output of a certain processing, which includes
`averaging of the latest RSRP values and their filtering [5].
`Previous works have shown that it is not a trivial task to set
`appropriately HO hysterisis and TTT, since the optimal
`setting depends on UE speed, radio network deployment,
`propagation conditions, and the system load [8]-[11]. The
`appropriate setting of HO
`triggers
`is of significant
`importance to the HO procedure, since the instant when the
`HO is triggered defines the radio propagation conditions to
`be met upon transmission of the HO-involved signaling;
`both for the messages transmitted in the serving and in the
`target cell.
`
`IV. L1 CONTROL CHANNELS
`The L1 control signaling [4] involves among others
`scheduling assignments, uplink scheduling requests (SR),
`channel quality estimation reports as well as HARQ
`feedback. Scheduling grants are mapped to the Physical
`Dedicated Control Channel
`(PDCCH), and HARQ
`feedback for uplink transmissions is mapped on the
`Physical Hybrid ARQ Indicator Channel (PHICH). Uplink
`scheduling requests and feedback for downlink HARQ
`processes are transmitted via the Physical Uplink Control
`Channel (PUCCH); eventually PUCCH
`is used for
`feedback of the downlink channel quality and other MIMO
`related channel state information. Since the format of
`PDCCH can vary dynamically, it has to be signalled as
`well to the UEs in the cell. The PDCCH transmission
`format is signaled via the Physical Control Format
`Indicator Channel (PCFICH).
`In case an error occurs upon transmission of control
`information on L1 control channels, additional delay is
`
`Figure 1. HO procedure.
`the reference symbols received quality
`(RSRP) and
`(RSRQ) [3]. If certain network configured conditions are
`satisfied, the UE sends the corresponding measurement
`report indicating the triggered event. In addition, the
`measurement report indicates the cell to which the UE has
`to be handed over, which is termed "target" cell. The
`triggering mechanism within the UE is described in detail
`in Section III.
`Based on these measurement reports, the serving eNB
`starts handover preparation. The HO preparation involves
`exchanging of signaling between serving and target eNB
`and admission control of the UE in the target cell. The
`communication interface between the serving and the
`target eNB is called X2 [7]. Upon successful HO
`preparation, the HO decision is made and consequently the
`HO Command will be sent to the UE. The connection
`between UE and the serving cell will be released. Then, the
`UE attempts to synchronize and access the target eNB, by
`using the random access channel (RACH). To speed up the
`handover procedure, the target cell can allocate a dedicated
`RACH preamble-included in HO command [2]-to the UE.
`Upon successful synchronization at the target eNB, this last
`one transmits an uplink scheduling grant to the UE. The
`UE responds with a HO Confirm message, which notifies
`the completion of the HO procedure at the radio access
`network part. It is noted that the described signaling
`messages belong to the Radio Resource Control (RRC)
`protocol [7].
`
`III. HANDOVER TRIGGERING
`As mentioned in section II, handover is triggered at the
`UE on the basis of triggers defined by the network. Namely,
`a set of triggers is signaled to the UE, one of them is named
`hysteresis, or "HO hysteresis", and the second one is called
`"Time To Trigger" (TTT) [7].
`The UE makes periodic measurements of RSRP and
`RSRQ based on the RS received from the serving cell and
`from the strongest adjacent cells. In case the handover
`algorithm is based on RSRP values, handover is triggered
`when the RSRP value from an adjacent cell is higher than
`the one from the serving cell by a number of dBs equal to
`
`2
`
`
`
`caused since the control information has to be retransmitted.
`A number of error scenarios upon control channels
`transmission exist; some of them impact the transmission
`of HO signalling; namely, errors upon transmission of DL
`assignment and UL grant via PDCCH, HARQ feedback
`over PUCCH and PHICH, as well as errors upon
`transmission of SRs via PUCCH.
`If an error happens during UL grant transmission, the
`UL HO signalling message, i.e. the measurement report or
`the handover confirm will be delayed until the reception of
`a new grant. In most of the cases, the eNB detects the grant
`error due to the lack of uplink transmission (DTX) at the
`TTI when the UL transmission is expected. Detection of
`DTX may trigger a new grant allocation by the scheduler.
`If no new grant is received, the UE sends a new SR.
`Similarly, errors at transmission of DL assignment delay
`the DL HO signalling, namely of the handover command.
`A new DL assignment may be scheduled upon the
`detection of missing HARQ feedback at a given TTI at the
`eNB. This detection occurs after half the HARQ round trip
`time (RTT). As shown in Figure 3, at least one more
`HARQ transmission is needed. In simulations presented
`here the HARQ RTT is set to 8 ms.
`Errors on HARQ feedback occur when NACK is
`erroneously decoded as ACK, or ACK as NACK.
`Compared to PHICH, PUCCH has additional errors; DTX
`is detected as ACK or NACK, or vice versa. In case ACK
`is erroneously interpreted as NACK or DTX, there is no
`explicit impact on the transmission delay of an HO-
`involved message. On the contrary, when NACK or DTX
`is decoded as ACK, the recovery from this error involves
`link layer retransmissions via the Radio Link Control (RLC)
`protocol [12], which might lead to delays of 50-60 ms.
`However, due to the low probability of negative reception,
`in combination with the low probability of this L1 control
`error happening, the impact from such an error is not
`considerable.
`Regarding the losses of SR over PUCCH, the UE is
`awaiting till the next time instant the transmission of uplink
`scheduling requests is allowed through PUCCH. The
`transmission interval of SR is set to 10 ms in simulations.
`
`Figure 3. Error upon transmission of DL assignment.
`
`V. SYSTEM MODEL
`A system level simulator featuring a radio network
`consisting of 21 hexagonal wrapped around cells is used.
`
`TABLE I. SIMULATION PARAMETERS
`Variable
`Value
`Cellular layout
`21 cells (7 sites)
`Cell radius
`288 m and 1000 m
`Traffic Type
`VoIP traffic (40 and 100 users
`per cell)
`20 Watts
`5 MHz
`2 GHz
`Okumura-Hata model
`3GPP Typical Urban
`{3, 50, 120, 250} km/h
`{0;1;3;5;10;15}%, target: 1%
`
`BS Tx Power
`System bandwidth
`Carrier Frequency
`Propagation model
`Channel Model
`UE speed
`UL grant / DL
`assignment error rate
`PUCCH NACK to
`ACK error rate
`PUCCH ACK to
`NACK error rate
`PHICH NACK to ACK
`error rate
`HO messages size
`
`{0;0.1;5;10;15}%, target: 0.1%
`
`{0;1;5;10;15}%, target: 1%
`
`{0;0.1;5;10;15}%, target: 0.1%
`
`Measurement report: 184 bits
`HO command: 288 bits
`HO confirm message: 112 bits
`20 ms
`
`UE RRC Processing
`Delay
`50 ms
`X2 Latency
`One site serves three cells and hosts one eNB. A fixed
`number of users are uniformly distributed over the area
`with randomly initialized positions. Users are moving with
`a specified speed at random directions. The UEs have
`active VoIP sessions throughout the whole simulation. The
`simulation time is set to 30 seconds. Both at the eNB and at
`the UE, one antenna for transmission and two antennas for
`receptions (Single Input Multiple Output, SIMO) are used.
`Simulation parameters and the VoIP traffic model conform
`to the 3GPP LTE simulation scenarios [13]. The most
`relevant simulation parameters are listed in Table I.
`Round Robin scheduling policy is used. RRC messages
`are prioritized over VoIP traffic. In addition, all the RLC,
`MAC and physical layer processing is implemented in
`conformance to 3GPP guidelines.
`
`VI. HANDOVER PERFORMANCE METRICS
`The performance of HO procedure is measured in terms
`of HO failure rate and overall delay.
`1. HO loss rate. A HO is considered as failed when the
`transmission of one RRC HO-involved message
`exceeds a predefined delay. In simulations this delay is
`set to 280 ms, accounting for 4-5 RLC retransmissions.
`2. The overall delay of the HO procedure:
` HO overall delay = t2 – t1
` (1)
`where t2 is the instant the HO confirm is received at
`the target eNB, and t1 is the instant when the
`measurement report is transmitted by the UE. Hence,
`the HO overall delay includes the delays due to:
`transmission of the measurement report, reception and
`processing of HO command, RACH procedure, and
`
`3
`
`
`
`In this high loaded scenario, the UEs are resource
`limited. This will result into the segmentation of uplink HO
`signalling, because only few resources are allocated to UEs.
`This limitation is more accentuated in UL, where the
`scheduler does not have knowledge of the UEs buffers
`contents. Consequently measurement reports and HO
`confirm messages cannot be prioritized over VoIP traffic,
`as in DL. Failures happen rarely upon transmission of the
`HO confirm, due to the much better propagation conditions
`in the new serving cell than the ones in the previous one.
`B. Impact of L1 Control Channel Errors on HO
`performance
`In simulations, the L1 control channel is designed as a
`channel with fixed error probability. Both the targeted error
`rates defined in [7] and significantly higher rates are tested.
`The simulated UE speeds, cell radii, and numbers of UEs
`per cell are the ones presented in the previous paragraphs.
`The focus is placed mainly on the delay of the HO
`procedure, since, errors at L1 control channels cause extra
`delays. Figure 6 shows the mean HO overall delay with
`different error probabilities on PDCCH and PCFICH. The
`cell radius is 1000 m. For the UL grant/DL assignment
`error scenarios, the mean HO overall delay increases from
`approximately. 92 up to 96.5ms even with an error rate of
`15%. The delay due to DL assignment errors can be
`estimated as:
`Delayadd_dl = Delaydl * DL assignment Error rate (2)
`
`where Delaydl is the delay added due to a DL assignment
`error. This delay will be typically the DL HARQ RTT
`interval, hence 8 ms; similarly, the delay due to the loss of
`an UL grant is:
`Delayadd_Ul = DelayUl * UL grant Error rate (3)
`where DelayUl is the delay due to the loss of an UL grant.
`As discussed in Section IV, this delay can vary between 1
`and 11 ms, since the UL SR transmission interval is 10 ms.
`Considering that in the overall handover procedure the
`transmission of two UL and of one DL messages is
`
`Figure 6. Mean HO overall delay with ULgrant /DL assignment
`errors in cells with radius1km.
`
`the reception and processing of HO confirm message
`plus delays via X2 and within serving and target eNB.
`
`SIMULATION RESULTS
`
`VII.
`A. HO Failure Rate
`This section shows simulation results for 40 and 100
`UEs per cell carrying VoIP traffic. Further results with
`lower loads are available in [8].
`Figure 4 shows the HO failure rate for the simulated UE
`speeds. The cell radius is 1 km. As expected, the increase
`in the UE speed and the system load has an impact on the
`HO failure rate. For the case of 40 UEs per cell, the HO
`failure rate remains under 1.3 %, even for the speed of 250
`km/h. For the case of 100 UEs per cell, the HO failure rate
`is acceptable up to the speed of 120 km/h. For the case of
`250 km/h, the HO failure rate increases to 4.58 %.
`
`HO failure rate (cell radius: 1 km)
`
`4.58
`
`2.28
`
`1.33
`
`0.87
`
`0.89
`
`0.30
`
`50km/h
`
`120km/h
`
`250km/h
`
`0
`0
`3km/h
`
`5,00
`
`4,00
`
`3,00
`
`2,00
`
`1,00
`
`0,00
`
`HO failure rate(%)
`
`40 VoIP UEs/cell
`100 VoIP UEs/cell
`Figure 4. HO failure rate for 40 and 100 VoIP UEs per cell.
`
`It is noted that in small cell sizes, neither the high speed
`nor the high load is a problem. This is achieved by the low
`latency in the X2 interface and in the fast processing within
`the eNBs. This low latency eliminates the risk of the UE
`loosing its connection with the serving cell, while still
`waiting for the HO Command. This is the main difference
`between LTE and its predecessors.
`In the specific scenario of 1 km cell radius, 250 km/h
`UE speed and highly loaded system, almost all of the
`handover failures are due to the transmission of the
`measurement reports, as Figure 5 indicates.
`HO failures (UE speed: 250 km/h, cell radius: 1 km)
`
`Measurement Report
`HO Command
`RACH
`HO Confirm
`
`Measurement
`Report 97%
`
`HO
`Command
`0%
`RACH 0%
`
`HO Confirm
`3%
`
`Figure 5. Percentage of failures occurring upon
`transmission of the different HO involved messages; UE
`speed 250 km/h and cell radius 1 km.
`
`4
`
`
`
`of measurement reports in uplink. This is because uplink is
`more resource limited, compared to DL, due to the need of
`UEs buffer estimation at the eNB. This will result into
`much more segmentations for UL HO signaling messages,
`which causes longer transmission delays.
`It has to be noted that even with these HO failure rates,
`the target of 1% drop rates can be achieved by using
`handover-or radio link-failure recovery mechanisms [7].
`The investigation of HO procedure featuring failure
`mechanisms would provide better insight on the impact of
`HO failures onto call drops.
`These results are obtained with the best performing set
`of triggers per scenario [8]. As also mentioned in section
`III, the setting of HO triggers is of primary importance for
`the design of a good performing HO procedure. Hence, it is
`inferred that adaptation of the HO triggers on the basis of
`speed, propagation conditions and cell sizes is needed.
`Considering the difficulties in adapting properly the HO
`triggers [9]-[11] an apposite solution is the use of a series
`of HO triggers. The RSRP variations observed in the
`various scenarios can be captured by the combination of
`HO triggers, resulting in improved HO performance in
`comparison to the one presented in this paper.
`
`[1]
`
`[2]
`
`[3]
`
`[4]
`
`[5]
`
`REFERENCES
`3GPP TR 25.913, “Requirements for evolved UTRA (E-UTRA)
`and evolved UTRAN (E-UTRAN),” version 7.3.0.
`3GPP TS 36.300, "Evolved Universal Terrestrial Radio Access
`(E-UTRA) and Evolved Universal Terrestrial Radio Access
`Network (E-UTRAN); Overall Description; Stage 2", version
`8.7.0, December 2008.
`3GPP TS 36.214, "Evolved Universal Terrestrial Radio Access
`(E-UTRA) Physical Layer - Measurements", version 8.5.0,
`December 2008.
`3GPP TS 36.213, "Evolved Universal Terrestrial Radio Access
`(E-UTRA) Physical Layer Procedures (Release 8)", version
`8.5.0, December 2008.
`Andras Racz, Andras Temesvary and Norbert Reider,
`“Handover Performance in 3GPP Long Term Evolution (LTE)
`Systems,” Proc. of Mobile and Wireless Communications
`Summit, 2007. 16th IST, pp.1 – 5, July 2007.
`[6] M. Anas, F.D. Calabrese, P.-E.Östling et. al., “Performance
`Analysis of Handover Measurements and Layer 3 Filtering for
`UTRAN LTE,” Proc. of PIMRC 2007, pp.1 – 5, Sept. 2007.
`3GPP TS 36.331, "Evolved Universal Terrestrial Radio Access
`(E-UTRA); Radio Resource Control
`(RRC); Protocol
`Specification (Release 8)", version 8.4.0, December 2008.
`Y. Yang, “Optimization of Handover Algorithms within 3GPP
`LTE,” MSc Thesis Report, KTH, February 2009.
`R4-081592, 3GPP TSG RAN WG4 Meeting #47bis, “System
`Simulation Results for Mobility State Detection based Cell
`Reselection”. Ericsson, June 2008.
`[10] R4-082062, 3GPP TSG RAN WG4 Meeting #48, “Performance
`of Mobility State Detection based Cell Reselection”. Ericsson,
`August 2008.
`[11] R4-082491, 3GPP TSG RAN WG4 Meeting #48bis,
`“Performance of Mobility State Detection based Cell
`Reselection-Further Simulation Results”. Ericsson, September
`2008.
`[12] A. Larmo, M. Lindström, M. Meyer, G. Pelletier, J. Torsner, H.
`Wieman, ”The LTE Link Layer Design”, IEEE Communications
`Magazine, April 2009 Vol. 47, No 4.
`3GPP TR 25.912, “Feasibility study for evolved UTRA (E-
`UTRA) and evolved UTRAN (E-UTRAN),” version 7.1.0.
`
`[7]
`
`[8]
`
`[9]
`
`[13]
`
`Figure 7. Mean HO overall delay with NACK to ACK
`errors in cells with radius of 288 m.
`involved,
`then, when
`the error probability upon
`transmission of UL grant/DL assignments are 15%, then
`the maximum added delay
`in
`the overall handover
`procedure is 11*0.15*2+8*0.15, i.e. approximately 4.5 ms,
`which corresponds to the simulation result of Figure 6.
`Figure 7 shows the mean HO overall delay with
`different HARQ ACK/NACK error probabilities in cells
`with radius of 288 m. With 15% NACK to ACK error rate,
`the mean HO overall delay increases approximately 2ms.
`Similarly, the average added delay due to HARQ NACK to
`ACK errors is estimated as:
`Delayadd = Delayrlcrx * NACK rate * Error rate
`
` (4)
`
`where Delayrlcrx is the delay due to one RLC retransmission,
`normally 50-60 ms. The NACK rate is 10%. In the overall
`handover procedure, this NACK to ACK error can occur
`three times, when the probability of error upon NACK
`transmission
`is 15%,
`the
`increase
`in delay
`is
`60*3*0.1*0.15, i.e. approximately 3 ms. Hence, the impact
`of a NACK to ACK error is smaller than that of UL
`grant/DL assignment error. The simulation results match
`the values obtained by formula (4).
`The added delay due to L1 control channel errors in the
`overall handover procedure is most of the times below 5ms.
`Considering that the delay for the overall handover
`procedure, which involves the transmission of three RRC
`messages, is below 100 ms and the maximum transmission
`delay threshold for each RRC message is 280ms, it is
`readily
`inferred
`that
`the
`impact from errors upon
`transmission of L1 control channels is insignificant.
`
`DISCUSSION AND CONCLUSIONS
`VIII.
`The results in this paper show that the HO mechanism
`within LTE meets the performance requirements even with
`errors at L1 control channels. One of the main reasons is
`the low latency in the radio access network and the low
`delay for HO procedures between the source and the target
`eNBs.
`The HO failure rate exceeds the threshold of 2% in large
`heavy loaded cells when the users move with speeds of 250
`km/h. The main reason for these failures is the transmission
`
`5
`
`