260
`
`LTE for UMTS — OFDMAand SC-FDMA Based Radio Access
`
`normal AMR of12.2 kbps. AMR-WBoffers clearly better voice quality than AMR-NBwith the
`samedata rate and can be called wideband audio with narrowbandradio transmission. The radio
`bandwidth is illustrated in Figure 10.1 and the audio bandwidth in Figure 10.2. The smallestbit
`rates, 1.8 and 1.75 kbps, are used for the transmission of Silence Indicator Frames (SID).
`This chapter considers AMR codecrates of 12.2, 7.95 and 5.9kbps. The resulting capacity
`of 12.2 kbps would also be approximately valid for AMR-WB 12.65 kbps.
`Whencalling outside mobile networks, voice transcoding is typically required to 64kbps
`Pulse Code Modulation (PCM) in links using ITU G.711 coding. For UE-to-UEcalls, the
`transcoding can be omitted with transcoder free or tandem free operation [1]. Transcoding
`generally degrades the voice quality and is not desirable within the network.
`
`AMR
`
`
`
`
`
`
`
`
`
`
`
`
`
`4.75 kbps
`
`1.8 kbps
`
`AMR-WB
`
`23.85 kbps
`
`18.2
`
`8.85 kbps
`
`6.6 kbps
`
`1.75 kbps
`
`
`
`
`
`
`
`
`
`
`
`
`
`Figure 10.1 Adaptive Multirate (AMR) Voice Codec radio bandwidth
`
`
`
`
`
`
`
`Human ear
`20-20000 Hz
`
`
`
`Wideband AMR
`50-7000 Hz
`
`Narrowband
`AMR
`300-3400 Hz
`
`Figure 10.2. Adaptive Multirate (AMR) Voice Codec audio bandwidth
`
`280
`
`W2235-00012675
`
`280
`
`

`

`Voice over IP (VoIP)
`
`261
`
`There are also a number of other voice codecs that are generally used for VoIP. A few
`examples are G.729, using an 8kbps coding rate, and Internet Low Bit Rate Codec (iLBC),
`using 13 kbps whichis used, for example, in Skype and in Googletalk.
`
`10.3 VoIP Requirements
`
`There is a high requirementset for the radio network to provide a reliable and good quality
`voice service, Some ofthe main requirements are considered below.
`The impact of the mouth-to-ear latency on usersatisfactionis illustrated in Figure 10.3. The
`delay preferably should be below 200 ms, whichis similar to the delay in GSM or WCDMA
`voice calls. The maximum allowed delay for a satisfactory voice service is 280 ms.
`IP Multimedia Subsystem (IMS) can be deployedto control VoIP. IMSis described in Chapter
`3. IMSprovidesthe information about the required Quality of Service (QoS)to the radio network
`by using 3GPPstandardized Policy and Charging Control (PCC) [3]. The radio network must
`be able to have the algorithmsto offer the required QoS better than Best Effort. QoS includes
`mainly delay, error rate and bandwidth requirements. QoS in LTE is described in Chapter8.
`The voice call drop rates are very low in the optimized GSM/WCDMAnetworkstoday — in
`the best case below 0.3%. VoIP in LTE mustoffer similar retainability including smooth inter-
`working between GSM/WCDMAcircuit switched (CS) voice calls. The handover functionality
`from VoIP in LTE to GSM/WCDMACSvoiceis called Single radio Voice Call Continuity
`(SR-VCC)and described in detail in section 10.10.
`The AMR 12.2 kbps packetsize is 31 bytes while the IP header is 40-60 bytes. IP header
`compression is a mandatory requirementfor an efficient VoIP solution. IP header compres-
`
`
`
`
`LLL Aa
`
`"CQY
`
`
`
`
`
`
`verysatisfied
`
`
`
`
`
`E-modelrating(R)
`
`
`
`
`
`
`WWW)
`
`%
`us all
`
`Uy
` G.114_FO1
`yy dissatisfiedLi
`
`users
`
`Mouth-to-ear delay (ms)
`
`Figure 10.3 Voice mouth-to-ear delay requirements [2]
`
`281
`
`W2235-00012676
`
`281
`
`

`

`262
`
`LTE for UMTS — OFDMAand SC-FDMA Based Radio Access
`
`sion is required both in UE and in eNodeB. All the VoIP simulations in this chapter assume
`IP header compression.
`The IP connectivity requires keep alive messages when the VE doesnot have a phonecall run-
`ning. The frequency of the keep alive messages dependson the VoIP solution: operator IMS VoIP
`can usefairly infrequent keep alive messages since IMSis within the operator’s own network and
`nofirewalls or Network Address Tables (NAT)are required in between. The internet VoIP requires
`very frequent keep alive message to keep the connectivity open throughfirewalls and NATs. The
`frequent keep alive message can affect UE power consumption and networkefficiency.
`VoIP roaming cases need further attention especially if there are some LTE networksdesigned
`for VoIP and data, while some networks are designed for data only transmission withoutthe
`required voice features. VoIP roaming also requires IMS and GPRS roaming agreements and the
`use of visited GGSN/MMEmodel. Oneoption is to use circuit switched (CS) calls whenever
`roaming with CS Fallback for LTE procedures. Similarly CS calls can also be used for emer-
`gency calls since 3GPP Release 8 LTE specifications do not provide the priority information
`from the radio to the core network nora specific emergency bearer.
`
`10.4 Delay Budget
`
`The end-to-end delay budget for LTE VoIP is considered here. The delay should preferably be
`below 200 ms, which is the value typically achieved in the CS network today. We use the fol-
`lowing assumptionsin the delay budget calculations. The voice encoding delay is assumedto
`be 30 ms including a 20 ms frame size, 5 ms look-ahead and 5 ms processing time. The receiv-
`ing end assumes a 5 msprocessing time for the decoding. The capacity simulations assume a
`maximum 50 msair—interface delay in both uplink and downlink including scheduling delay
`and the time required for theinitial and the HARQretransmissions of a packet. We also assume
`a5 ms processing delay in the UE, 5 ms in eNodeB and | ms in the SAE gateway. The transport
`delay is assumed to be 10 msandit will depend heavily on the network configuration. The delay
`budgetis presented in Figure 10.4. The mouth-to-ear delay is approximately 160 ms with these
`
`200 5
`
`
`
`0
`
`
`
`180
`
`160
`
`140 5
`120 5
`
`ms 100 5
`
`80 =
`60 4
`40 -
`
`20
`
`MSAE GW
`& Transport
`[@ eNodeBprocessing
`Downlink transmission
`.
`.
`Uplink transmission
`ff UE processing
`@ AMRdecoding
`© AMRencoding
`
`LTE voice end-to-end delay
`
`Figure 10.4 LTE voice end-to-end delay budget
`
`282
`
`W2235-00012677
`
`282
`
`

`

`Voice over IP (VoIP)
`
`263
`
`assumptions,illustrating that LTE VoIP can provide lower end-to-end latency than CS voice
`calls today while still providing high efficiency.
`
`10.5 Scheduling and Control Channels
`
`By default the LTE voice service uses dynamic scheduling, where control channels are used to
`allocate the resources for each voice packet transmission and for the possible retransmissions.
`Dynamic scheduling gives full flexibility for optimizing resource allocation, but it requires
`control channel capacity. Multi-user channel sharing plays an important role when optimizing
`the air interface for VoIP traffic. Because each user is scheduled with control signaling with
`the dynamic scheduler, the control overhead might becomea limiting factor for VoIP capacity.
`One solution in downlink for reducing control channel capacity consumption with a dynamic
`scheduler is to bundle multiple VoIP packets together at Layer | into one transmission of a
`user. The packet bundling is CQI based and is applied only for users whose channel conditions
`favor the use of bundling. The main benefit from the bundlingis that more users could be fitted
`to the network with the same control channel overhead as users in good channel conditions
`are scheduled less often. The drawback from the bundling is that bundled packets experience a
`tighter delay budget, but the negative impactof this to VoIP capacity can be kept to a minimum
`by applying bundling for good enough users that are not relying on HARQretransmissions.
`From the voice quality perspective it is also important to minimize the probability of losing
`consecutive packets. This can be achieved by makingthe link adaptation for the TTIs carrying
`bundled packets in a more conservative way leading to a reduced packeterror rate for the first
`transmission. Due to the low UE transmission power and non-CQIbased scheduling, packet
`bundling is not seen as an attractive technique in LTE uplink.
`The 3GPPsolution to avoid control channellimitation for VoIP capacity is the Semi-Persistent
`Scheduling (SPS) method [4, 5], where the initial transmissions of VoIP packets are sent without
`associated scheduling control informationbyusing persistently allocated transmission resources
`instead. The semi-persistent scheduling is configured by higher layers (Radio Resource Control,
`RRC), and the periodicity of the semi-persistent scheduling is signaled by RRC.Atthe beginning
`of a talk spurt, the semi-persistent schedulingis activated by sending the allocated transmission
`resources by Physical Downlink Control Channel (PDCCH)andthe UE stores the allocation
`and uses it periodically according to the periodicity. With semi-persistent scheduling only
`retransmissions and SID frames are scheduled dynamically, implying that the control channel
`capacity is not a problem for the semi-persistent scheduler. On the other hand, only limited
`time and frequency domain scheduling gains are available for semi-persistent scheduling. The
`semi-persistent allocation is released during the silence periods. Semi-persistent scheduling
`is adopted for 3GPP Release 8. The downlink operation of the semi-persistent scheduleris
`illustrated in Figure 10.5.
`In the following the impact of control channel capacity to VoIP maximum capacity with
`dynamic scheduleris illustrated with theoretical calculations.
`It is assumed that the number of Control Channel Elements (CCE) is approximately 20
`per 5 MHz bandwidth. Two,four or eight CCEs can be aggregated per user depending on the
`channel conditions in low signal-to-noise ratios. We further apply the following assumptions
`for the calculations: voice activity 50%, voice packet arrival period 20ms, downlink share of
`the traffic 50% and numberof retransmissions 20%. Theresults are calculated without packet
`
`283
`
`W2235-00012678
`
`283
`
`

`

`264
`
`LTE for UMTS — OFDMAand SC-FDMA Based Radio Access
`
`Talk burst
`
`Talk burst
`
`20 ms
`
`20 ms
`
`20 ms
`
`20 ms
`
`i
`
`ACK
`
`THM
`
`TT
`
`Tk
`
`NACK
`
`J = Downlink allocation
`= = Semi-persistent allocation
`%
`os
`% = Retransmission
`
`SEER,
`
`>
`
`ACK
`
`THT
`
`ee
`ACK
`
`sete
`ACK
`
`Figure 10.5 Semi-persistent scheduling in downlink
`
`bundling and with two packet bundling. To simplify the calculations it is further assumedthat
`SID frames are not taken into account. VoIP maximum capacity can be calculated by using
`Equation 10.1.
`
`eriod|ms] - Packet_bundling- Downlink_share. —————
`Max_users=-——#Cs Packet
`
`#CCEs_per_user—Voice_activity 1+ BLER
`
`(10.1)
`
`The calculation results are shown in Figure 10.6 for a 5 MHz channel. As an example, the
`maximumcapacity would be 330 users without CCE aggregation and without packet bundling.
`Assuming an average CCE aggregation of three brings the maximum capacity to 110 without
`packet bundling and 220 with packet bundling. These theoretical calculations illustrate the
`importance that multi-user channel sharing has for VoIP system performance, and further-
`more, the maximum gain in capacity that packet bundling may provide with the dynamic
`
` 500
`
`4 Packet bundling = 1
`@ Packet bundling = 2
`
`450
`
`400
`
`
`
`
`
`350 5
`
`300 5
`
`250
`
` Usersper5MHz
`
`200 +
`
`150 5
`
`100
`
`50 5
`
`0
`
`Figure 10.6 Maximumcontrol channellimited capacity with fully dynamic scheduling
`
`Average CCE aggregation
`
`284
`
`W2235-00012679
`
`284
`
`

`

`Voice over IP (VoIP)
`
`265
`
`scheduler. According to the system level simulations, the average size of CCE aggregationis
`approximately 1.5 for a scheduled user. Comparisonofthe simulated downlink capacities! for
`a dynamic scheduler (presented in section 10.6) with the theoretical upper limits showsthat
`the simulated capacities without packet bundling match rather nicely the theoretical calcula-
`tions, being approximately 5% lower than the theoretical maximum. With packet bundling the
`simulated capacities are notably below the theoretical upperlimits, as in practice the probability
`for using bundling is clearly limited below 100%.
`
`10.6 LTE Voice Capacity
`
`This section presents the system level performance ofVoIP traffic in LTE at a 5 MHz system
`bandwidth. VoIP capacity is given as the maximum numberof users that could be supported
`within a cell without exceeding 5% outage. A user is defined to be in an outageif at least 2%
`of its VoIP packetsare lost (i.e. either erroneous or discarded) during the call. VoIP capacity
`numbers are obtained from system level simulations in a macro cell scenario | [6], the main
`system simulation parameters being aligned with [7]. An illustration of a VoIP capacity curve
`is presented in Figure 10.7, which shows how the outage goes up steeply when the maximum
`capacity is approached. This enables running the system with a relatively high load (compared
`to the maximum capacity) while still maintaining low outage.
`The results of VoIP capacity simulations are summarized in Table 10.1 for three different
`AMRcodecs (AMR5.9, AMR 7.95 and AMR 12.2) and for both dynamic and semi-persistent
`schedulers. The capacity is 210-470 users in downlink and 210-410 users in uplink, which
`
`
`
`0.14
`
`0.12 -
`
`0.1 -
`
`2° 2co
`
`0.04 -
`
`capacity
`
`:
`
`-
`
`-
`
`230
`
`240
`
`
` 8Outageprobability
`0.02
`
`180
`
`190
`
`200
`
`210
`Numberof users
`
`220
`
`Figure 10.7 An example of VoIP capacity curve
`
`‘Simulated capacities should be up-scaled by 12.5% to remove the impact of SID transmissions on the
`capacity.
`
`285
`
`W2235-00012680
`
`285
`
`

`

`266
`
`LTE for UMTS — OFDMAand SC-FDMA Based Radio Access
`
`Table 10.1 VoIP capacity in LTE at 5 MHz
`
`
`
`VoIP codec AMR7.95 AMR 12.2 AMR5.9
`
`
`
`Downlink capacity
`
`Dynamic scheduler, without bundling
`Dynamic scheduler, with bundling
`Semi-persistent scheduler
`
`Uplink capacity
`
`Dynamic scheduler
`Semi-persistent scheduler
`
`210
`410
`470
`
`230
`410
`
`210
`400
`430
`
`230
`320
`
`210
`370
`320
`
`210
`240)
`
`corresponds to 42-94 users per MHz percell in downlink and 42—82 users per MHzpercell in
`uplink. The lower AMRrates provide higher capacity than the higher AMRrates. The AMR
`codec data rate can be defined by the operator allowing a tradeoff between the capacity and
`the voice quality. The lower AMRrates do notincrease the capacity with dynamic scheduling
`due to the control channellimitation.
`The results also show that voice capacity is uplink limited, which can be beneficial when
`there is asymmetric data transmission on the samecarrier taking more downlink capacity. The
`downlink offers higher capacity than uplink because the downlink scheduling uses a point-to-
`multipoint approach and can be optimized comparedto the uplink.
`In the following, the simulated capacities are analyzed in more detail. All the supporting
`statistics presented in the sequel are assumingload asclose as possible to the 5% outage.
`Asdescribedin section 10.5, VoIP system performance with the dynamic scheduleris limited
`by the available PDCCHresources, and therefore the dynamic scheduler cannot fully exploit
`the Physical Downlink Shared Channel (PDSCH)air interface capacity as there are not enough
`CCEsto schedule the unused Physical Resource Blocks (PRBs). This is illustrated in Figure
`10.8 for downlink, which contains a cumulative distribution function for the scheduled PRBs
`per Transmission Time Interval (TTI) with AMR 12.2. The total number of PRBs on 5 MHz
`bandwidth is 25. As can be seen from Figure 10.8, the average utilization rate for the dynamic
`scheduler is only 40% if packet bundling is not allowed.
`Dueto the control channellimitations, the savings in VoIP packet payload size with lower
`AMRrates are not mapped to capacity gains with the dynamic scheduler, if packet bundling is
`not used. The different codecs provide almost identical performance in downlink whereasthere
`are small gains in capacity in uplink with lower AMRrates. This uplink gain originates from
`the improved coverage as more robust MCS could be used when transmitting a VoIP packetin
`uplink — this gain does not exist in the downlink direction as it is not coverage limited due to
`higher eNodeB transmission power.
`Whenpacket bundling is enabled, the average size of an allocation per scheduled user 1s
`increased and hence the air interface capacity of PDSCH can be more efficiently exploited,
`whichleads to an increased average PRButilization rate of 70%. Gains of 75-90% are achieved
`for capacity when packet bundling is enabled. The probability of using packet bundling in a
`macro cell scenario is approximately 70% for AMR 12.2 codec,and it slightly increases when
`the VoIP packet payload decreases. This explains the small (< 10%) gains in capacity achieved
`by reducing VoIP packet payload.
`
`286
`
`W2235-00012681
`
`286
`
`

`

`Voice over IP (VoIP)
`
`267
`
`
`
`
`
`—— Dynamic w/o bundling
`og)oc77" Dynamic w bundling
`
`0.4
` 094 7% Semi-persistent
`
`a7F
`
`0.6
`
`0.4
`
`0.3
`
`0.2
`
`
`
`15
`Total number of RBsin use per TT!
`
`20
`
`25
`
`Figure 10.8 Cumulative distribution function for scheduled Physical Resource Blocks (PRB) per TTI
`in downlink (out of a total of 25 PRBs on 5 MHz bandwidth)
`
`The performance of the semi-persistent scheduler does not suffer from control channel
`limitations, as initial transmissions are scheduled without associated control information by
`using persistently allocated time and frequency resources instead. The difference in control
`channel consumption for simulated packet scheduling methods is presented for the downlink
`direction in Figure 10.9, which contains the cumulative distribution function for the total
`number of consumed CCEsper TTI. With the dynamic schedulerall control channel elements
`are in use 70% of the time, if bundling is not allowed. With packet bundling, load is higherat
`5% outage and hence control channel capacity could be more efficiently exploited implying
`that all CCEsare used almost 100% of the time. With the semi-persistent scheduler, the control
`channel overhead is clearly lower, the probability of using all CCEsbeingonly slightly higher
`than 10%. Similar observations can be made from the distribution of consumed CCEsper TTI
`in the uplink direction, which is presented in Figure 10.10. As expected, the CCE consumption
`for semi-persistent scheduling is much lowerthanthat for dynamic scheduling. Here we assume
`that the total number of CCEs reserved for downlink/uplink scheduling grants is 10.
`The performance of the semi-persistent scheduler is not limited by the control channel
`resources, butit is limited by the available PDSCH bandwidth, whichis illustrated in Figure
`10.8, showing that the average PDSCH utilization rate for a semi-persistent scheduler is more
`than 90%. As the performanceofthe semi-persistent scheduleris userplane limited, the size of
`a VoIP packethas a significant impact on the capacity: an approximately 35% higher capacity
`is obtained if AMR 7.95 is used instead of AMR 12.2, whereas the use of AMR 5.9 instead of
`AMR7.95 provides capacity gains of only 10%. The reason for the reduced gain for AMR 5.9
`is presented in Figure 10.11, which showsthedistribution ofthe size of the persistent resource
`allocation in terms of PRBsforall simulated codecs in the downlink direction. According to
`
`287
`
`W2235-00012682
`
`287
`
`

`

`268
`
`LTE for UMTS — OFDMAand SC-FDMA Based Radio Access
`
`|
`
`T
`
`
`T
`
`O.9+
`
`—s— Semi-persistent
`
`Dynamic w/o bundling
`sono Dynamic w bundling
`
`T
`
`:
`
`O
`
`|
`
`
`
`
`0
`
`2
`
`6
`4
`Total number of CCEs in use per TTI
`
`8
`
`10
`
`Figure 10.9 Cumulative distribution function for the total number of CCEs per TTI in downlink
`
`
`
`
`
`
`1
`
`0.9-
`
`0.8 -
`
`
`
`Semi-persistent
`tooten Dynamic
`
`
`
`
`
`0.7
`
`0.6 -
`
`0.5 -
`
`cdf
`
`0.4 -
`
`0.3-
`
`0.2
`
`0.1 -
`
`0)
`
`
`
`
`
`
`
`
`crceeees
`
`
`
`:
`poceceeed
`
`
`possessed
`pusenened
`
`
`gonsesenes
`ewannn nen een eee eeenmmeneet nn nnnee
`0
`10
`2
`
`6
`4
`Total number of CCEs in use per TT!
`
`8
`
`Figure 10.10 Cumulative distribution function for the total number of CCEs per TTIin uplink
`
`the distribution, the size ofthe persistentallocation is one (1) PRB for almost 50%ofthe users
`with AMR 7.95, and hence no savingsin allocated bandwidth can be achieved forthose users
`when using AMR5.9 instead of AMR 7.95. In the downlinkthesize of the persistent allocation
`
`288
`
`W2235-00012683
`
`288
`
`

`

`Voice over IP (VoIP)
`
`269
`
`60;
`
`r
`
`t
`
`i
`MMB AMR 5.9
`SAMA 7.95 —
`
`
`
`
`probability[%]ai
`
`nmQ
`
`Probability distribution functionforthe size of the persistent resource allocation for
`Figure 10.11
`different codecs in downlink
`
`Size of persistent allocation [PRB]
`
`is calculated dynamically from wideband CQI for each talk spurt separately, whereas in the
`uplink, due to practical constraints,the size ofthe persistent resource allocation is decided from
`the path loss measurements of a user. The reason for the uplink solution is that the sounding is
`not accurate enough with a large numberofusers. Thesize of the persistent resource allocation
`has the following distribution during the uplink simulation: with AMR12.2 all the users have
`a persistent allocation of size two (2) PRBs, whereas with AMR7.95 half of the users have a
`persistent allocation of size one (1) PRB, and half of the users have a persistent allocation of
`size two (2) PRBs. With AMR 5.9, 90% of the users have a persistent allocation of size one
`(1) PRB, and others havea persistent allocation of size two (2) PRB. An approximately 33%
`higher capacity is obtained if AMR 7.95 is used instead of AMR 12.2. Furthermore, AMR 5.9
`provides approximately 28% gains in capacity over AMR 7.95. The reduced gain is mainly
`due to a slightly increased numberof retransmissions for AMR 5.9.
`When comparing the schedulers in the downlink direction for experienced air interface
`delay, it is observed that due to the control channellimitations the relative amountofpackets
`experiencing delay close to the used delay bound (50 ms) is clearly higher for the dynamic
`scheduler. With packet bundling, control channellimitations can be relaxed, and therefore delay
`critical packets can be scheduled earlier. Bundlingitself adds delays in steps of 20 ms.For the
`semi-persistent scheduler, the time the packet has to wait in the packet scheduler buffer before
`initial scheduling takes place dependson the location of the persistent resource allocation in the
`time domain, which hasa rather even distribution with a full load. Therefore the packet delay
`distribution for the semi-persistent scheduler is smooth. In all these schemes, retransmissions
`are scheduledat the earliest convenience, after an 8ms HARQcycle. Dueto the first HARQ
`retransmission the distributions plateau at 1Oms. A cumulative distribution function for the
`packet delay is presented in Figure 10.12.
`
`289
`
`W2235-00012684
`
`289
`
`

`

`270
`
`LTE for UMTS — OFDMAand SC-FDMA Based Radio Access
`
`— Dynamic w/o bundling|+
`soecee Dynamic w bundling
`
` —€— Semi-persistent
`
`
`
`5
`
`0
`
`10
`
`|
`30
`20
`Packet Delay [ms]
`
`40
`
`50
`
`Figure 10.12 Cumulative distribution function for the experiencedair interface delay in downlink
`with AMR 12.2
`
`When summarizing VoIP system performance in downlink,it is concludedthat the perfor-
`mance of the dynamic scheduleris control channel limited without packet bundling and hence
`the semi-persistent scheduler is able to have 50-125% higher capacities than the dynamic
`scheduler. The performanceof the semi-persistent scheduleris user plane limited and the gains
`in capacity over the dynamic scheduler are smallest for AMR 12.2, which hasthe highest data
`rate amongst the simulated codecs and hence the biggest VoIP packet size. When the packet
`bundling is used with AMR 12.2, the control channel limitation for the performance of the
`dynamic scheduleris less significant compared with lowerrate codecs as the numberof sup-
`ported users at 5% outage is lower. Therefore, the dynamic scheduler can have a 15% gain
`in capacity over the semi-persistent scheduler if AMR 12.2 is used with the packet bundling.
`When bundling is used with lower rate codecs, the control channel capacity starts to limit
`the performance of the dynamic scheduler due to the high number of supported users at 5%
`outage, and hence the semi-persistent scheduler achieves 8—15% higher capacities than the
`dynamic scheduler.
`When comparing the performances ofdifferent scheduling methods in the uplink, it is
`observed that(similar to downlink) the performanceof the dynamic scheduleris purely control
`channel limited whereas the semi-persistent scheduler suffers much less from control channel
`limitation due to a looser control channel requirement. Therefore, the semi-persistent scheduler
`is able to have 40-80% capacity gains over the dynamic scheduler with AMR 7.95 and AMR
`5.9. Moreover, even with AMR 12.2 the semi-persistent scheduler achieves a 14% higher
`capacity than the dynamic scheduler.
`The presented downlinkresults are obtained by using frequency dependent CQIreporting.In
`practice the large numberof supported VoIP users in LTE maynecessitate the use of wideband
`CQIto keep the overhead from CQIfeedback at a reasonable level. This would mean a lower
`
`290
`
`W2235-00012685
`
`290
`
`

`

`Voice over IP (VoIP)
`
`271
`
`Table 10.2 Relative loss in capacity (%) due to increased CQI
`granularity
`
`
`CQI reporting sub-band size
`
`
`4 PRBs
`Wideband CQI
`
`Semi-persistent scheduling
`Dynamic without bundling
`Dynamic with bundling
`
`2%
`0%
`7%
`
`7%
`3%
`15%
`
`uplink signaling overhead from CQI feedbackatthe cost of reduced capacity, as the frequency
`domain scheduling gains will be lost. To keep the capacity reduction as small as possible, the
`impactof lost frequency domain scheduling gains to performance should be compensated with
`more efficient use of frequency diversity. Therefore the use of the semi-persistent scheduler
`becomes even moreattractive for VoIPtraffic in LTE, as the performanceof the semi-persistent
`scheduleris less dependenton the frequency domain scheduling gains than the dynamic sched-
`uler. This is illustrated in Table 10.2, which presents the relative losses in capacity with AMR
`12.2 whenincreasing the size of CQI reporting sub-band in the frequency domain. Relative
`losses are calculated against the presented capacity numbers, which were obtained assuming
`narrowband CQIreporting [7].
`According to the simulations, the capacity of the dynamic scheduler is reduced by 15%
`due to the use of wideband CQI, whereas for the semi-persistent scheduler the corresponding
`loss is only 7%. Hence with wideband CQI the semi-persistent scheduler provides a similar
`performance to the dynamic scheduler for AMR 12.2, and for lower rate codecs the gains in
`capacity over the dynamic schedulerare increased further compared to the gains achieved with
`narrowband CQI.
`Finally, in the light of the above results analysis, it seems that semi-persistent schedulingis
`the mostattractive scheduling method for VoIP traffic in LTE. Nevertheless, as the used per-
`sistent allocation is indicated to the user via PDCCHsignaling, some sort of combination of
`dynamic and semi-persistent scheduling methods seems to be the best option for VoIP in LTE.
`Furthermore, when comparing the performances of downlink and uplink, it is concluded that
`the VoIP performance in LTEis uplink limited. Hence downlink capacity can accommodate
`additional asymmetric datatraffic, e.g. web browsing.
`
`10.7 Voice Capacity Evolution
`
`This section presents the evolution of the voice spectral efficiency from GSM to WCDMA/
`HSPAand to LTE. Mostof the global mobile voice traffic is carried with GSM EFR or AMR
`coding. The GSM spectral efficiency can typically be measured with Effective Frequency Load
`(EFL), which represents the percentageofthe timeslots offull rate users that can be occupiedin
`case of frequency reuse one. For example, EFL=8% correspondsto 8% x8 slots/200 kHz =3.2
`users/MHz. The simulations and the network measurements show that GSM EFR can achieve
`EFL 8% and GSM AMRcan achieve 20% assuming all terminals are AMR capable and the
`network is optimized. The GSM spectral efficiency can be further pushed by the network fea-
`ture Dynamic Frequency and Channel Allocation and by using Single Antenna Interference
`Cancellation (SAIC), known also as Downlink Advanced Receiver Performance (DARP). We
`assume up to EFL 35% with those enhancements. For further information see [8] and [9]. The
`
`291
`
`W2235-00012686
`
`291
`
`

`

`272
`
`LTE for UMTS — OFDMAand SC-FDMA Based Radio Access
`
`overhead from the Broadcast Control Channel (BCCH)is not included in the calculations.
`BCCHrequires higher reuse than the hopping carriers.
`WCDMAvoice capacity is assumedto be 64 users with AMR 12.2 kbps, and 100 users with AMR
`5.9kbps on a 5 MHzcarrier. HSPA voice capacity is assumed to be 123 users with AMR 12.2 kbps,
`and 184 users with AMR 5.9kbps. For HSPA capacity evolution, see Chapter 13.
`Capacity evolutionis illustrated in Figure 10.13. LTE VoIP 12.2 kbps can provide efficiency
`which is 15x more than GSM EFR.The high efficiency can squeeze the voice traffic into a
`smaller spectrum. An example calculation is shownin Figure 10.14 assuming 1500 subscrib-
`
`
`
`
`
`
`
`
`
`
`
` UserperMHz
`
`100
`
`90
`
`80
`
`70
`
`60
`
`50
`
`40
`
`
`
`
`
`
`
`10
`
`T
`
`T
`
`GSM
`
`EFR
`
`GSM
`
`AMR
`
`:
`
`. i
`_ -
`T
`T
`T
`T
`T
`GSM WCDMA HSPA_ HSPA5.9 HSPA 5.9 LTE VoIP LTE VoIP
`
`DFCA
`
`5.9kbps 12.2kbps
`
`kbps
`
`kbps with 12.2 kbps 5.9 kbps
`Ic
`
`Figure 10.13 Voice spectral efficiency evolution (IC = Interference Cancellation)
`
`
`
`
`
`
`
`30
`
`25
`
`
`
`
`
`GSM EFR
`
`GSM
`
`AMR
`
`GSM WCDMA
`
`HSPA- HSPA5.9 HSPA 5.9 LTE VoIP LTE VoIP
`
`DFCA
`
`5.9kbps
`
`12.2kbps
`
`kbps
`
` kbpswith 12.2 kbps
`IC
`
`5.9 kbps
`
`Figure 10.14 Spectrum required for supporting 1500 subscribers per sector at 40 mErl
`
`292
`
`W2235-00012687
`
`292
`
`

`

`Voice over IP (VoIP)
`
`273
`
`ers per sector and 40 mErltraffic per subscriber. GSM EFR would take 25 MHz of spectrum
`while LTE will be able to carry that voicetraffic in less than 2 MHz. LTE canthen free up more
`spectrum for data usage.
`
`10.8 Uplink Coverage
`
`Uplink coverage can be maximized when UEtransmits continuously with maximum power.
`Since VoIP packets are small (< 40 bytes), they can easily fit into | ms TTI. The AMRpackets
`arrive every 20 msleading to only 5% activity in the uplink. The uplink coverage could be
`improved by increasing the UE transmission time by using retransmissions and TTI bun-
`dling. These solutionsare illustrated in Figure 10.15. The number ofretransmissions must
`be limited for VoIP to remain within the maximum delay budget. The maximum number
`of retransmissions is six assuming a maximum 50 mstotal delay, since the retransmission
`delay is 8 ms in LTE.
`TTI bundling can repeat the same data in multiple (up to four) TTIs [4,10]. TTI bundling
`effectively increases the TTI length allowing the UEto transmit for a longer time. A single
`transport block is coded and transmitted in a set of consecutive TTIs. The same hybrid ARQ
`process numberis used in each of the bundled TTIs. The bundled TTIs are treated as a single
`resource where only a single grant and a single acknowledgement are required. TTI bundling
`can be activated with a higher layer signaling per UE. The trigger could be, for example, UE
`reporting its transmit poweris getting close to the maximum value.
`The resulting eNodeB sensitivity values with retransmission and TTI bundling are shown
`in Table 10.3. The eNodeB sensitivity can be calculated as follows:
`
`eNodeB_sensitivity |dBm]=—174 + 10 log, (Bandwidth) +Noise_figure+SNR (10.2)
`
`1 ms
`
`No TTI bundling, no
`retransmissions
`
`20 ms
`
`20 ms
`
`No TTI bundling
`with retransmissions
`
`retransmissions
`
`4 TTI bundling with
`
`a = First transmission
`A = Possible retransmission
`
`Figure 10.15 TTI bundling for enhancing VoIP coverage in uplink
`
`293
`
`W2235-00012688
`
`293
`
`

`

`274
`
`LTE for UMTS — OFDMAand SC-FDMA Based Radio Access
`
`Table 10.3 Uplink VoIP sensitivity with TTI bundling[11]
`
`
`
` Numberof TTIs bundled I 4
`
`
`
`Transmission bandwidth
`Numberof retransmissions
`Required SNR
`Receiversensitivity
`
`360 kHz (2 resource blocks)
`6
`—4.2dB
`—120.6dBm
`
`360kHz (2 resource blocks)
`3
`-8.0dB
`—124.4dBm
`
`The eNodeB receiver noise figure is assumed to be 2dB, two resource blocks are used
`for voice and no interference margin is included. The bundling can improve the uplink voice
`coverage by 4 dB.
`The WCDMANodecBsensitivity can be estimated by assuming Eb/NO=4.5 dB [12], which
`gives —126.6dBm. To get similar voice coverage in LTE as in WCDMAweneed to apply
`TTI bundling with a sufficient numberof retransmissions. In terms of Hybrid-ARQ (HARQ)
`acknowledgement and retransmission timing, the TTI bundling methodillustrated in Figure
`10,16 is adopted for the LTE specifications [4,10]. According to this method four (4) subframes
`are in one bundle for the Frequency Division Duplex (FDD) system. Within a bundle, the opera-
`tion is like autonomousretransmission by the UE in consecutive subframes without waiting for
`ACK/NACKfeedback. The Redundancy Version (RV) on each autonomousretransmission in
`consecutive subframes changes in a pre-determined manner.
`HARQacknowledgementis generated after receiving the last subframe in the bundle. The
`timing relation betweenthe last subframe in the bundle and the transmissioninstant of the HARQ
`acknowledgementis identical to the case of no bundling. If the last subframe in a bundle