1

Latency in HSPA Data Networks

Siddharth Mohan, Rohit Kapoor, Bibhu Mohanty
Abstract:
Latency and throughput are two critical performance metrics of a communication network. Recently, a
lot of attention has been focused on improving throughput (or spectral efficiency) of Wireless Wide Area
Networks (WWANs) through the use of physical and MAC layer techniques, such as higher order
modulation, MIMO and aggregation of bandwidth (multi-carrier). While some data applications directly
benefit from the higher data rates, for many applications high data rates do not translate to improved
user experience unless the latency is low. In this paper, we examine the Control plane (C-plane) and
User plane (U-plane) latencies in an HSPA data system. We show that significant C-plane latency
reduction can be achieved in HSPA by carrying signaling on HS-DSCH and E-DCH channels (as opposed to
dedicated channels, which is the practice now). We also compare the C-plane and U-Plane latencies of
HSPA and LTE, which have comparable spectral efficiency [1].
1. Introduction
3
rd
Generation Partnership Project (3GPP) has standardized Code Division Multiple Access (CDMA) based
packet-switched air-interfaces for downlink and uplink, called High-Speed Downlink Packet Access
(HSDPA) and High-Speed Uplink Packet Access (HSUPA) respectively, together referred to as High Speed
Packet Access (HSPA) [2][4] l PSA
dedicated channels. These dedicated channels are typically low rate channels (3.4 kbps), which allows
larger spreading factors to be used S
signaling messages tend to be fairly large, the use of these low rate dedicated channels leads to call
setup latencies that are in the range of a few seconds. Newer HSPA deployments are considering
carrying signaling on HSPA channels (i.e., HS-DSCH on the downlink and E-DCH channels on the uplink).
Such configurations not only allow better statistical use of power and code resources, but also allow
signaling to benefit from higher rate HSPA channels. We will show that significant improvements to call
setup latency can be achieved by carrying signaling on HSPA channels.
In parallel with the evolution of HSPA, 3GPP has also standardized an Orthogonal Frequency Division
Multiplexing (OFDM) based air-interface, called Long Term Evolution (LTE) [3]. Though the spectral
efficiencies of the two systems are comparable, Rel-9 LTE provides higher peak rates (due to its wider
20MHz bandwidth) compared to Rel-8/Rel-9 HSPA, which allows a User Equipment (UE) to receive on 10
MHz bandwidth with carrier aggregation. This paper focuses on the call setup and user plane latencies of
the two systems. Call setup latency determines how quickly the user can start to receive service, while
user plane latency is important since the performance of applications such as web browsing is very
sensitive to the Round Trip Time (RTT) to the server.
2

In this paper, we compare call setup latency of HSPA when signaling is carried on dedicated channels
versus HSPA channels. We also compare call setup and user plane latencies of HSPA and LTE. A detailed
breakdown of the call setup as well as user plane latencies of HSPA and LTE systems is available in the
Appendix.
2. Classification of Latency
Latency can typically be classified as:
User Plane Latency: Loosely speaking, User plane (U-plane) latency is the delay seen by an application in
exchanging data with a server. A metric that is commonly used to characterize U-plane latency is the
PING delay, i.e., the delay from a UE sending a small PING to the first IP node in the network and
receiving the PING response. This PING latency is the definition of U-plane latency used in this paper.
Control plane latency: Control plane (C-plane) latency, also known as call setup latency, is the latency
for a User Equipment (UE) to transition to a state where it can send/receive data.
The definition of call setup latency merits further clarification. The 3GPP specification for HSPA allows
the UE to camp
1
with low battery consumption in either RRC idle or CELL_PCH (or URA_PCH, which is
similar to CELL_PCH for the purposes of this paper) states [4]. In these states, the UE only needs to listen
to the network in pre- uL
during the rest of the DRX (Discontinuous Reception) cycle. Camping in CELL_PCH state has a number of
benefits: the signaling cost and delay to transition the UE from CELL_PCH to a state where the UE can
CLLLlACP uL
battery consumption in CELL_PCH and idle states is similar (assuming same DRX cycles for both).
Given the two types of camped states (idle and CELL_PCH) allowed by the HSPA specification, we
consider two definitions of call setup, one from RRC idle to CELL_DCH (we refer to this as call setup from
idle) and one from CELL_PCH to CELL_FACH (we refer to this as call setup from connected). Legacy HSPA
networks typically transitioned the UE to RRC idle state once data transfer was finished, so UEs
experienced the latency of call setup from idle. However, with the deployment of CELL_PCH, we expect
that UEs will typically camp in CELL_PCH, and only very long continuous inactivity (of the order of tens of
minutes or more) may cause them to be transitioned to RRC idle. Thus, in future HSPA networks, we
expect CELL_PCH (as opposed to idle) to be the default camped state from which the UE will initiate
most of its data transactions.
While CELL_PCH is supported in the specifications since R99, an Enhanced version of CELL_PCH was
added in Rel 7 [4]. Enhanced CELL_PCH allows the UE to retain a dedicated H-RNTI in CELL_PCH state,
which means that the UE is primed to send/receive data.

1
Camping traditionally refers to camping of the UE in Idle mode. In this paper, we apply the term camping to mean
user being inactive, which corresponds to Idle, CELL_PCH and URA_PCH states for HSPA, and connected state with
long DRX for LTE.

3

In the case of LTE, the only two RRC states are idle and connected. In connected state, the UE is allowed
to camp with a long discontinuous receive (DRX) cycle [5]: for the purposes of this paper, this can be
considered the equivalent of the CELL_PCH state in HSPA. Thus, for LTE, the term call setup from idle
refers to a transition from RRC idle to a state where data can be sent/received, and the term call setup
from connected refers to a transition from RRC connected with a long DRX cycle to a state where data
can be sent/received.
3. Overview of Results
In this section, we summarize the key results of this paper. We present the following comparisons:
Comparison 1: Call setup from Idle

C-Plane Scenario Signaling on
Dedicated Channels
for HSPA (ms)
Signaling on
HSPA Channels
(ms)
LTE (ms)
MO 1717 546 188
MT 1920 749 365
Table 1: Latencies of Call setup from idle for HSPA (when carrying signaling on Dedicated channels or HSPA channels) and LTE
Comparison 2: Call setup from connected

C-Plane Scenario CELL_PCH to
CELL_FACH for
HSPA (ms)
Enhanced
CELL_PCH to
CELL_FACH for
HSPA (ms)
Long DRX to
Connected for
LTE (ms)
MO 82 14 13
MT 242 174 173
Table 2: Latencies of Call setup from connected for HSPA (when starting from CELL_PCH or Enhanced CELL_PCH) and LTE

Comparison 3: U-Plane latency

U-Plane Process HSPA (ms) LTE (ms)
PING latency 38 32
Table 3: U-Plane latencies of HSPA and LTE
A detailed analysis of how the numbers in this section are derived, including call flows and assumptions,
is shown in the Appendix.

4

Conclusion
In this paper, we analyzed the latency of call setup from idle for a legacy WCDMA network where
signaling is carried on dedicated channels. We showed that a significant (~1.2 sec) gain in latency of call
setup from idle can be achieved by carrying signaling on HSPA channels. We showed that LTE further
reduces the latency of call setup from idle latency by ~360 ms compared to an HSPA system where
signaling is carried on HSPA channels. One of the key reasons that contribute to this gain for LTE is that
the Radio Bearer Setup message is assumed to be a synchronized message for HSPA, whereas the
corresponding message in LTE is unsynchronized. Unsynchronized Radio Bearer Setup is supported by
the HSPA specification, but we did not consider it in this paper since it is not commonly deployed.
We also compared the latencies of call setup from connected between HSPA and LTE systems and
showed them to be similar, if the Enhanced CELL_PCH state, introduced in Release 7 of the HSPA
specification, is considered. If starting from legacy CELL_PCH, HSPA latency is 69-75 ms more than that
of LTE. We expect that UEs will typically camp in connected camped state (i.e., CELL_PCH for HSPA and
connected with long DRX cycle for LTE), and thus latency of call setup from connected is likely to drive
responsiveness of the system when the UE is camped.
Finally, we compared the U-plane latency (or PING request-response latency) of HSPA and LTE systems.
We showed that, with similar H-ARQ operating points and similar processing latencies of network nodes,
the U-plane latencies for HSPA and LTE are similar.
5

Appendix
A1. Transport Network Technology and Architecture for HSPA and LTE
systems
Figure 1 shows the network architecture for UMTS and the Evolved Packet System (EPS) flat architecture
for LTE. In the EPS flat architecture, there is user plane direct tunneling between the Serving Gate way
(S-GW) and the eNodeB. For EPS, RNC functions are partly in the eNodeB and partly in the the MME,
SGSN functions are in the MME, GGSN functions are in the S-GW and the PDN-GW.

Figure 1: Transport Network Architecture for HSPA and LTE
A2. WCDMA Latency of Call Setup from Idle when Signaling is carried
on Dedicated Channels
In this section, we present a breakdown of the typical latency of call setup from idle in legacy 3GPP
WCDMA networks. This section focuses on the latency when signaling is carried over dedicated
channels, whereas Section A3 compares this with the case where signaling is carried over HSPA channels
(i.e., HS-DSCH and E-DCH channels for downlink and uplink respectively). We consider both mobile
originated (MO) as well as mobile terminated (MT) cases.

6

Mobile Originated Call Setup from Idle
The call flow for a Mobile Originated (MO) call setup from idle for a packet switched (PS) call is shown in
Figure 2 [4]. Some of the key procedures involved in call setup are:
1. RRC Connection Establishment Procedure: The purpose of this procedure is to transition the UE
from RRC idle mode to RRC connected state. The messages involved are RRC Connection
Request initiated by the UE, RRC Connection Setup sent by the RNC in response to the RRC
Connection Request and RRC Connection Setup Complete sent by the UE.
2. GPRS Attach Procedure: 1 uL
the packet switched core network. The messages involved in this procedure are the Initial Direct
Transfer sent by the UE to the core network and the Security Mode Command initiated by the
core network to activate ciphering and integrity protection. The GMM Attach Accept sent to the
UE completes the attach procedure.
3. PDP Context Activation Procedure: The purpose of this procedure is to assign an IP address to
the UE and create a data path from the packet network to the UE. The steps involved are
Activate PDP Context sent by the UE to the GGSN (not shown inFigure 2), radio link setup
between the RNC and the NodeB, setting up of radio bearers between the RNC and the UE, and
Activate PDP Context Accept sent to the UE.
Table 4 shows the breakdown of the latency for the steps shown in Figure 2. The numbers in Table 4
assume that signalling radio bearers are mapped to 3.4 kbps dedicated channels with 40ms TTI. The
sizes of the messages exchanged in the MO call flow are shown in Table 4. Other assumptions, including
message processing latencies of various nodes, are shown in Section A6. Note that the call setup latency
could change if the message processing latencies of nodes and/or the backhaul (RNC-Node B) latency
are different than what is assumed.
From Table 4, the average total delay incurred in a MO PS call setup from idle is ~ 1.7sec, when signaling
is carried on 3.4 kbps dedicated channels.



7


Figure 2: MO PS Call Flow for Call Setup from Idle for WCDMA

8


4. Security Mode Command 191
SGSN to RNC, RNC to NodeB delay 6
UE Phy layer and RRC layer
processing delay for DL msg
8
RRC Security mode command DL
Tx*
80
UE Packet preparation 4
RRC Security mode complete UL
Tx*
80
NodeB and RNC Processing 7
NodeB to RNC, RNC to SGSN delay 6
5. Downlink direct transfer (GMM
Attach Accept)
100
RNC to NodeB delay for DL 5
UE Physical layer and RRC layer
processing delay for DL message
8
GMM Attach Accept DL Tx * 80
RNC to SGSN and SGSN to RNC
Delay
2
SGSN Processing 5


6. Uplink Direct Transfer 130
UE Packet preparation 4
Activate PDP Context Request UL
Tx*
80
NodeB Phy and RNC Processing 7
NodeB, RNC and SGSN delay 7
Radio Link Prepare/Ready/Commit 27
SGSN Processing 5

7. RAB Assignment 951
SGSN-RNC Delay 1
Radio Bearer Setup DL Tx* 800
UE Packet preparation 4
Radio Bearer Setup Complete UL
Tx*
40
NodeB Phy and RNC Processing 7
NodeB to RNC, RNC to SGSN delay 6
RNC to NodeB delay for DL 5
UE Physical layer and RRC
processing delay for DL message
8
Activate PDP context accept DL
Tx*
80

Total MO PS latency 1717
Table 4: MO PS Latency for Call Setup from Idle when Signaling is Carried on 3.4 kbps Dedicated Channels
Process Delay
(ms)
Process
Latency
(ms)

1. Random Access 72
SIB 7 Uplink interference
parameter delay (Assuming
80ms cycle for SIB7 broadcast)
40
Preamble
PRACH period 0.67
PRACH transmit time 1.33
PRACH processing at NodeB 4
AICH Transmit time 2
AICH response processing 2

Data
RRC connection request 10
Node B Phy processing 2
Node B - RNC Delay and RNC
processing
10

2. RRC Connection Setup
Procedure
174
Radio Link Setup 17
RRC Connection Setup DL Tx* 40
RNC to NodeB delay for DL 5
UE Physical layer and RRC layer
processing delay for DL
message
8
NodeB Synchronization 20
UE Packet preparation 4
RRC Connection Setup
Complete UL Tx*
80
NodeB and RNC Processing and
delay (Happens in parallel when
UE is transmitting Attach
Request)
0

3. Initial Direct Transfer (GMM
Attach Request)
99
GMM Attach Request UL Tx* 80
NodeB, RNC and SGSN
Processing
12
NodeB, RNC and SGSN delay 7

*Refer Section A3
9

Mobile Terminated Call Setup from Idle
For Mobile Terminated (MT) call setup from idle, the call flow is similar to Figure 2, with the addition
that the UE receives a paging message before it sends a RRC Connection Request message. The paging
procedure adds an additional delay (shown in Table 5) for the MT call compared to the MO call. For MT
PS call setup from idle, the average total delay is ~ 1.9 sec.
Process Delay(ms) Comments/Explainations
Expected Delay due to Paging Cycle of 320ms 160 Average delay is mean of paging cycle
GGSN processing 5
Processing delays across network elements
SGSN processing 5
SGSN-RNC delay 1
RNC processing 5
RNC-Node B delay 5
Node B processing 2
Paging Delay 20 Delay in transmitting the paging indicator
and paging channels on the downlink
Total Page delay 203
Table 5 : Latency involved in Paging the UE
A3. WCDMA/HSPA Latency Improvements for Call Setup from Idle due
to carrying Signaling on HSPA channels
In this section, we focus on latency of call setup from idle when signaling is carried on HSPA channels.
Table 6 compares, for the various messages exchanged in the call flow, the air interface latency involved
when carrying signaling over HSPA versus 3.4 kbps dedicated channels. The delay of the messages
shown in Table 6 is obtained by dividing the size of the message by the channel rate, i.e., 3.4 kbps for
dedicated channel. For signaling carried over HSPA channels, 2ms TTI is assumed for E-DCH. Other
assumptions, including processing latencies of various nodes and average number of H-ARQ
transmissions for HS-DSCH and E-DCH, are shown in Section A6. The following explains the latency
numbers for two of the messages from Table 6 (the latency reduction for other messages follows a
similar argument):
1. RRC Connection Setup (NodeB to UE):
The size of RRC connection Setup is typically ~100 to 120 bytes. When sent over FACH with a
SCCPCH channel rate of 30Kbps*, the transmission delay is 30ms to 40ms. When sent over FACH
mapped to HS-DSCH channels (as in Enhanced CELL_FACH, which is a Release 7 feature), RRC
Connection Setup can be transmitted using 2ms TTI. Selecting a conservative 320 bit payload to
cover the cell edge (due to absence of CQI/ACK in CELL_FACH), the transmission delay is 6ms.
2. Radio Bearer Setup (NodeB to UE):
The typical size of the Radio Bearer Setup message is ~120bytes. This requires ~300ms to be
transmitted over a 3.4kbps dedicated channel. Moreover, since Radio Bearer Setup is typically a
synchronized message (i.e., it includes an Activation Time), the Activation Time needs to account for
RLC retransmission(s) as well as other network delays. Thus, typically a conservative activation time
* All messages except RRC connection setup are carried over the dedicated channel. The RRC
connection setup message is carried over the FACH channel.
10

of 500ms is selected, which leads to a latency of ~800ms, when carried over dedicated channels.
When transmitted over HS-DSCH channels, the transmission time of this message would be 3.2ms
(assuming 1.1 average H-ARQ transmissions and 1 ms waiting time for TTI). Moreover, the Activation
Time in this case does not need to be as conservative as that for dedicated channel, since a smaller
time needs to be budgeted for RLC retransmission(s). Thus, the total latency for this message,
assuming a 200ms Activation Time, is 203.2 ms. It may be noted here that WCDMA/HSPA also
supports an Unsyncronized Radio Bearer Setup message (with A 1
the potential to further reduce the C-plane latency. Unsynchronized Radio Bearer Setup has not
been considered in this paper.
No: OTA message Size Air Interface delay when
Signaling on Dedicated
Channels (ms)
Air Interface delay when
Signaling on HSPA (ms)
1 RRC Connection Setup 107 40 6
2 RRC Connection Setup
Complete
25 80 10
3 Initial direct transfer GMM
Attach Request
21 80 10
4 Security mode command 25 80 3.2
5 Security mode complete 25 80 10
6 Downlink direct transfer
GMM Attach Accept
10 80 3.2
7 Uplink direct transfer- Activate
PDP context
25 80 10
8 Radio Bearer Setup 126 800 (including Activation
Time of 500 ms)
203.2 (including Activation
Time of 200 ms)
9 Radio Bearer Setup Complete 10 40 10
10 Activate PDP context Accept 14 80 3.2
Total Delay 1440 269
Table 6: Air Interface Latency comparison when carrying Signaling over Dedicated and HSPA channels
Table 7 shows a comparison of the latencies of call setup from idle for HSPA when signaling is carried on
dedicated and HSPA channels. Refer to Table 10 for a detailed breakdown of HSPA latency numbers for a
546 ms MO call setup from idle. From Table 6 and Table 7, we can see that when signaling is carried on
HSPA channels, the transmission of signaling messages over the air interface contributes a small
percentage ((69 ms =269ms activation time of 200ms) out of a total of 546 ms) of this latency and the
rest is contributed by other factors such as backhaul, core network, Activation Time and SIB reading. MT
call latency is obtained by adding the paging delay (203ms from Table 5) to the MO delay.
C-Plane Process Signaling on
Dedicated Channels
(ms)
Signaling on
HSPA Channels
(ms)
MO Delay of Call Setup from Idle 1717 546
MT Delay of Call Setup from Idle 1920 749
Table 7: Comparison of Latency of Call Setup from Idle when Signaling is carried on Dedicated and HSPA channels
11

A4. HSPA Latency of Call Setup from Connected
For the case of call setup from connected, we assume that the UE camps in CELL_PCH state. The HSPA
specifications support legacy CELL_PCH as well as Enhanced CELL_PCH (introduced in Rel 7). Enhanced
CELL_PCH allows the UE to retain a dedicated H-RNTI in CELL_PCH state, which means that the UE is
primed to send/receive data. In comparison, legacy CELL_PCH requires the UE to exchange a Cell
Update/Cell Update Confirm with the network before uplink/downlink data can be exchanged. While
Enhanced CELL_PCH offers latency benefits over legacy CELL_PCH, networks are only now starting to
deploy legacy CELL_PCH. Thus, we analyze the call setup latency considering both Enhanced and legacy
CELL_PCH. Note that apart from the case when the user starts a data session after a long inactivity
period, the user will likely be in CELL_FACH state. From this state, the latency for call setup from
connected is similar to that from the Enhanced CELL_PCH state (for similar DRX cycle in CELL_FACH.
Latency of Call Setup from Connected for Enhanced CELL_PCH
We assume that the state to which the UE transitions where it can send/receive data is CELL_FACH
(specifically Downlink and Uplink Enhanced CELL_FACH). As mentioned before, Downlink and Uplink
Enhanced CELL_FACH (defined in Rel 7 and Rel 8 respectively) allow the use of HS-DSCH and E-DCH
channels, thus enabling high-speed data transfer in CELL_FACH state.
Figure 3 shows the procedure for an MO call setup from connected from Enhanced CELL_PCH to
Enhanced Uplink CELL_FACH: the two steps required before data can be transmitted are: (a) sending
RACH preamble and receiving acknowledgment on the Access Indicator Channel (AICH) and (b) the
sending of a configurable number of pilot-only preambles. Though the minimum duration for which
pilot-only preambles need to be sent is 2 TTIs (or 4 msec), for a more typical setting, the network may
configure a slightly larger number (example 5) of pilot-only preambles. Thus, we expect the delay in this
procedure to be in the range of 14-20 msec. A breakdown of this latency is shown in Table 8.
Note that as part of Uplink Enhanced CELL_FACH, a default uplink grant is carried in the SIBs, so that the
UE can start transmitting without requiring the network to assign an uplink grant. Moreover, Uplink
Enhanced CELL_FACH allows a default value of Uplink Interference to be sent in SIB5, and a UE
supporting Uplink Enhanced CELL_FACH does not need to read the dynamic value of Uplink Interference
from SIB7 before starting transmission.
For an MT call setup from connected 11l paging
cycle. Since the UE is in Enhanced CELL_PCH (with dedicated H-RNTI), it can directly receive data in the
11l paging cycle, without needing to go through the Cell Update procedure. The delay
involved in this procedure on the average is half the length of the paging cycle, i.e., 160 msec (Table 5).
Note that in Enhanced CELL_PCH, the UE still monitors the Paging Indicator Channel (PICH) to determine
whether to decode HS-uSCP 11Is. On receiving data, the UE autonomously transitions to
CELL_FACH.

12


Figure 3: Call Setup from Connected for Enhanced Cell PCH callflow
Call Setup from Connected for Enhanced Cell PCH
Delay(ms)
Total Delay (ms)
PRACH period 0.67
14-20
PRACH transmit time 1.33
PRACH processing at NodeB 4
AICH Transmit time 2
AICH response processing 2
E-DCH Transmission backoff 4-10 (2-5 pilot
premables)
Table 8: Breakdown of Latency of MO Call Setup from Connected for Enhanced cell PCH
Latency of Call Setup from Connected for Legacy CELL_PCH
Comparing to camping in Enhanced CELL_PCH, camping in legacy CELL_PCH incurs the additional latency
of the UE sending a Cell Update message to the RNC, and receiving a Cell Update Confirm from the RNC.
This is expected to incur an additional latency of 68.2 msec (when signaling is carried on HS channels),
bringing the MO call setup from connected to 82.2-88.2 msec and the MT call setup from connected to
248 msec (MT latency = MO latency + half the length of the paging cycle, i.e., 160ms). The breakdown of
this additional latency is shown in Table 9.
Call Setup from Connected for Legacy Cell PCH
Delay(ms)
Total Delay (ms)
Latency for Enhanced Cell PCH 14-20
82.2 to 88.2
Packet preparation for Cell Update 4
Cell Update Uplink Transmission 10
NodeB Phy Processing 2
NodeB to RNC delay 5
RNC processing delay 5
RNC to NodeB delay for DL 5
Cell Update confirm DL Tx 3.2
UE Physical layer processing delay for decoding DL message 3
UE RRC processing delay for DL message 5
UE Packet preparation for PCRC 4
PCRC Uplink Transmission 10
NodeB Phy Processing 2
NodeB to RNC delay 5
RNC processing delay 5
Table 9: Breakdown of Latency of MO Call Setup from Connected for Legacy Cell PCH
13

A5. Latency Comparison of HSPA and LTE
A5.1 Latency Comparison of Call Setup from Idle
In this section, we show the call flow of MO call setup from idle for an LTE system (Figure 4). We identify
some key differences between the call setup procedures for LTE and HSPA. We also compare the
latency of call setup from idle of an HSPA system (where signaling is carried on HSPA channels) with that
of an LTE system. For the comparison, we assume similar H-ARQ operating points and similar processing
latencies of network nodes for HSPA and LTE.

Figure 4: Call flow for MO Call Setup from Idle for LTE
14

Comparing Figure 4 (LTE) with Figure 2 (HSPA), the key differences are:
1. In LTE, the Attach Request (a NAS PDU) is piggybacked on the RRC Connection Setup Complete
message. In HSPA, the Attach Request is sent seperately after RRC Connection Setup Complete.
2. In LTE, the UE needs to get an uplink grant before it can send the RRC Connection Setup
Complete message.
3. The Activate PDP Context message is not sent in LTE since an IP address is assigned to the UE in
response to Attach Request (this is possible due to LTE being a PS only system). In addition, a
default radio bearer (default EPS bearer) is setup in response to the Attach Request and the UE
can start transmitting data after sending RRC Connection Reconfiguration Complete. However,
the MME needs to wait for the Attach Complete message sent by the UE to arrive before it can
start forwarding downlink data.
4. The RRC Connection Reconfiguration Complete message in LTE does not have an Activation
Time, while the corresponding message (i.e., Radio Bearer Setup) in HSPA is typically sent with
an Activation Time.
MO Call
HSPA Procedure HSPA delay
(ms)
LTE Procedure LTE delay
(ms)
--SIB reading time 40 --SIB reading time 40
--Random Access Preamble 11 --Random Access Preamble 9
--RRC connection request
--RRC connection Setup
78 --RRC connection request
--RRC connection Setup
--Scheduling Request
--UL grant
19.8
--RRC connection setup
complete
--GMM Attach Request
--Activate PDP context Request
63 --RRC Connection Setup
Complete (Attach Request)
-- S1 connection establishment
37.9
--Security Mode Command 44.2 --Security Mode Command 32.7
--GMM Attach Accept
--Radio Bearer Setup
--Activate PDP context Accept
311.4 --Attach Accept 48.7
Total 546 188
Table 10: Comparison of MO Call Setup Latency from Idle between HSPA and LTE
Table 10 shows that LTE reduces the MO call setup latency from idle by ~360 ms compared to HSPA
(when signaling is carried on HSPA channels).
MT Call
Procedure HSPA (ms) LTE (ms)
MO latency (fromTable 10) 546 188
Paging latency (from Table 5 for HSPA and Table 15 for LTE) 203 177
Total MT latency 749 365
Table 11: Comparison of MT Call Setup Latency from idle between HSPA and LTE
15

For MT call setup from idle, the UE needs to be paged in addition to the procedures detailed in Table 10.
Table 11 shows that LTE has a lower MT call setup latency from idle of ~385ms compared to HSPA.
The following are the key reasons for the lower call setup latency for LTE compared to HSPA:
1. The use of synchronized Radio Bearer Setup for HSPA, while the corresponding message (RRC
Connection Reconfiguration) for LTE is unsynchronized. This leads to an additional latency of
~200 ms for HSPA. Note that, unsynchronized Radio Bearer Setup is also supported in the
specifications for HSPA, but is not considered in this paper due to the fact that most of the
current deployments use synchronized setup.
2. NodeB power control synchronization for HSPA after entering CELL_DCH: this leads to an
additional latency of 20ms for HSPA.
3. Messages exchanged between NodeB and RNC for radio link setup for HSPA. This leads to an
additional latency of ~50ms for HSPA.
4. All RRC signaling messages experience the backhaul delay for HSPA, whereas in LTE, RRC
messages terminate at the eNodeB. This leads to an additional latency of ~50ms for HSPA.
A5.2 Latency Comparison of Call Setup from Connected

Figure 5: MO Call Setup from Connected for LTE
Figure 5 shows the procedure for MO call setup from connected for LTE, i.e, the steps required to reach
a state where data can be transmitted starting from RRC connected with long DRX cycle. The delay in
this procedure (as shown inTable 12) is 13 ms.
For MT call setup from connected 11l u8x
The delay involved in this procedure on the average is half the length of the DRX cycle, i.e., 160 msec
(since a DRX cycle of 320 msec is assumed in this paper).
Call Setup from Connected Delay (ms)
Total Delay (ms)
PRACH period, transmit time and processing time 4 13
RA response scheduling delay and transmission (ms) 2
RA response processing 3
Waiting for PRACH opportunity 4
Table 12: Latency of Call Setup from Connected for LTE
16

Comparing the latency of call setup from connected for HSPA and LTE (i.e., Sections A4and A5.2
respectively), we find that:
the latency for HSPA and LTE is similar when the starting state in HSPA is Enhanced CELL_PCH,
and
the latency for HSPA is ~69-75 ms larger than LTE when the starting state in HSPA is legacy
CELL_PCH.
A5.3 User Plane Latency comparison
In this section, we evaluate the U-plane latencies for HSPA and LTE for a 32 byte PING packet. We
assume similar H-ARQ operating points for HSPA and LTE (1.1 average number of transmissions for
downlink and 1.5 average number of transmissions for uplink) and the use of scheduled grants for
HSUPA (note that with the use of non-scheduled grants, the U-plane delay of HSPA can be further
reduced). From Table 13, we see that, with similar H-ARQ and backhaul assumptions, the U-plane
latencies for HSPA and LTE are similar. We explain some of the latencies from Table 13 below:
1. UL Resource Assignment latency is the latency involved in assignment and transmission of grants
for transmission of data on the uplink.
2. UL Data Transmission latency is the latency involved in packet preparation, waiting for TTI
boundary and transmission of data on the uplink.
3. Network Node Processing latency is the latency involved in processing at the RNC and the GGSN
for HSPA (note that the 1-tunnel architecture standardized in Rel 7 allows the SGSN to be
bypassed for user plane traffic) and the SGW and the PGW for LTE.
4. DL Transmission latency is the latency involved in packet preparation, waiting for TTI boundary
and transmission of data on the downlink.
Procedure HSPA delay (ms) LTE delay (ms)

UL Resource Assignment 8 8
UL Data Transmission 8 4
Node B Decoding 3 3
Backhaul delay 5 5
Network node processing ( RNC/GGSN for
HSPA and SGW/PGW for LTE)
2 2
Backhaul 5 5
DL Transmission 3 2
UE Decoding and processing 4 3
Total U-Plane Latency 38 32
Table 13: U-Plane Latency Comparison between HSPA system and LTE system

17

A6. System Assumptions
Table 14, Table 15 and Table 16 show processing delays and other delay assumptions that have been
used in the analysis in this paper. Some of these assumptions are obtained from base station/RNC
vendors and others are reasonable estimates.
HSPA System Assumptions
ms

HSPA System Assumptions
ms
RNC processing (ms) 5 RACH UL data Tx (ms) 10
Node B Processing (ms) 2 DL Synchronization (ms) 0
RNC-Node B Delay (ms) 5 UL Synchronization (ms) 20
SGSN-RNC Delay (ms) 1 UE Packet Preparation (ms) 4
SGSN processing (ms) 5 UE Phy processing (ms) 3
GGSN processing (ms) 5 Avg Number UL H-ARQ for E-DCH 1.5
UE RRC Processing (ms) 5 Avg Number DL H-ARQ for HS-PDSCH 1.1
NB Scheduling Delay (ms) 1 Downlink FACH Transmission
NB Phy Processing (ms) 2 Bits per transmission 320
Paging Cycle (ms) 320 Number of repeats 3
Paging Delay (ms) 20 Message Size (bytes) 100
PRACH period (ms) 1.33 Radio Link Setup

PRACH Tx time (ms) 12 RNC processing (ms) 5
PRACH processing NB (ms) 4 RNC-Node B Delay (ms) 5
AICH Tx time (ms) 2 Node B Processing (ms) 2
AICH response processing (ms) 2 Node B RNC Delay (ms) 5
Table 14: HSPA System Assumptions

Table 15: LTE System Assumptions

LTE System Assumptions ms
Paging 177
Half of the Paging Cycle (ms) 160
SGW processing (ms) 5
S11 delay (ms) 1
MME processing (ms) 5
DL S1-MME delay (ms) 5
eNB processing (ms) 1

DL Data Transmission 10.8
eNB scheduling delay (ms) 1
DL Data Tx time (ms) 1.8
UE Phy processing (ms) 3
UE RRC processing (ms) 5
LTE System Assumptions ms
UL Data Transmission 20.9
SR period 5
SR delay (ms) 2.5
SR tx (ms) 1
SR processing (ms) 2
UL scheduling delay (ms) 1
UL grant Tx (ms) 1
UL data Tx delay (ms) 3
UL data Tx (ms) 3.4
eNB Phy processing (ms) 2
eNB RRC processing (ms) 5
18

Process Delay
(ms)
Process
Latency
(ms)

3. Security Mode Command 32.7
NAS security mode
command**
10.8
NAS security mode command
complete*
20.9
eNodeB processing, MME
processing
6
UL S1-MME delay, DL S1 -
MME delay
10
eNodeB processing 1
RRC security mode
command**
10.8
RRC security mode command
complete*
20.9

4. Attach Accept 48.7
RRC reconfiguration
RRC connection
reconfiguration**
10.8
Attach Complete* 20.9
Update Bearer
eNodeB and MME processing 6
UL S1-MME delay, DL S1 -
MME delay
10
eNodeB processing 1

Total MO delay for LTE 188.1

Table 16: LTE Idle Call Setup MO Latency breakdown
References
[1] 3G Americas 'Transition to 4G, 3GPP Broadband Evolution to IMT-Advanced`-
http://www.ngmn.org/fileadmin/user_upload/News/Partner_News/3G_Americas_RysavyResearch_HSPA-LTE_Advanced_FINAL.pdf
[2] 3GPP TS 25.214, 'Physical Layer Procedures (F DD).
[3] 3GPP TS 36.214, 'Evolved Universal Terrestrial Radio Access (E-UTRA); Physical layer; Measurements.
[4] 3GPP TS 25.331, 'Radio Resource Control (F DD).
[5] 3GPP TS 36.331, 'Evolved Universal Terrestrial Radio Access (E-UTRA); Radio Resource Control (RRC); Protocol specification.

Process Delay
(ms)
Process
Latency
(ms)

1. Random Access 69.8
SIB 2 reading time (Assuming
80ms cycle for SIB2 broadcast)
40
Preamble Total
PRACH period, transmit time
and processing time
4
RA response scheduling delay
and transmission (ms)
2
RA response processing 3

Data Total
UE transmission delay,
transmission time
2
RRC connection request* 1
eNodeB PHY and RRC
processing
7
RRC connection setup** 10.8

2. RRC Connection Setup
Complete/Attach Request
37.9
RRC Connection Setup
Complete*
20.9
S1 establishment
eNodeB and MME processing 6
UL S1-MME delay, DL S1 -
MME delay
10
eNodeB processing 1
*Refer Table 15 DL data, **Refer Table 15 UL Data