Quality of Service (QoS) is an important concept in any network which ultimately leads to network efficiency and customer satisfaction. In this PPT, we deal mainly with the Quality of Service aspects relating to Femto Access Point (FAP) of UMTS technology. PPT mainly deals with the Guaranteed Bit Rate (GBR) implementations.

1. Krishna Adapa V M
UMTS System Architecture

2.  UMTS QoS architecture and Evolution
 Service and User Prioritization Strategies
 QoS Management in RNS/FAP
 Release 99 Channels
 HSDPA Channels
 HSUPA Channels
 QoS Management in Core Network Elements
 QoS Management in Backhaul Network

3.  Scope: Quality of Service (QoS) is a network wide topic
and its actual scope in-terms of depth and width are
tremendous. This presentation is primarily meant to
focus on the Home Node B (HNB) NW element related
functionality pertaining to Guaranteed Bit Rate (GBR)
implementations. Other aspects are just touched to
provide relevant overview of the topic under
consideration. Some aspects are purposefully
simplified when compared to the functionality of
macro network element (NodeB/RNC)
 Acknowledgements: Simulations, Algorithms and
other concepts referred in this presentation are
obtained from listed references. They are, by no means,
original contributions from the author.

4. 3GPP Architecture Evolution

5.  Voice over IP (VOIP):
 Mouth to ear delays (200 to 300 msec)
 Packet RTT (150 to 200msec)
 Bit Rate (16 to 64 kbps) depending on CoDeCs
 Web browsing:
 High data rate (200 to 400kbps)
 Short packet RTT (<200msec)
 Streaming:
 Bit rate 64kbps to 128kbps
 Stable bit rates and small bit rate variations

6. UMTS
TE MT RAN CN CN TE
EDGE Gateway
NODE
End-to-End Service
TE/MT Local UMTS Bearer Service External Bearer
Bearer Service Service
Radio Access Bearer Service CN Bearer
Service
Radio Bearer RAN Access Backbone
Service Bearer Service Bearer Service
Physical Radio Physical
Bearer Service Bearer Service

7. TE MT RAN CN EDGE Gateway Ext.
Netw.
Adm./Cap. Adm./Cap. Adm./Cap. Subscr. Adm./Cap.
Transl. Control Control Control Control Control Transl.
Local Ext.
Service Service
Control UMTS BS UMTS BS UMTS BS Control
Manager Manager Manager
RAB
Manager
Local BS Radio BS Radio BS RA BS RA BS CN BS CN BS Ext. BS
Manager Manager Manager Manager Manager Manager Manager Manager
RAN RAN RA NS RA NS BB NS BB NS
ph. BS M ph. BS M Manager Manager Manager Manager
protocol interface service primitive interface

8. TE MT RAN CN EDGE Gateway Ext.
Netw.
Class
if.
Class
if. Cond.
Cond.
Cond. Mapper Mapper Mapper
Local BS Resource Resource Resource Resource Resource Resource External BS
Manager Manager Manager Manager Manager Manager
RAN phys. BS RAN Access network service BB network service
data flow with indication of direction

9.  Traffic Class
 Maximum Bit Rate
 Guaranteed Bit Rate
 Delivery Order
 Maximum SDU Size
 SDU Format Information
 SDU Error Ratio
 Residual Bit Error Ratio
 Delivery of Erroneous SDUs
 Transfer Delay
 Traffic Handling Priority
 Allocation/Retention Priority
 Source Statistics Descriptor

10.  QoS Parameter Request by the UE at PDP Ctx
Activation to SGSN. SGSN Admission Control.
 QoS parameter negotiation between SGSN – GGSN in
Create PDP context request/response. GGSN
Admission Control.
 QoS parameter negotiation between SGSN – FAP
through RAB Assignment Request/Response
 Radio Interface Protocol parameters derivation by FAP
Admission Control (RLC modes, SIR/BLER targets,
Transport channels selection, TFCS
selection/definition, Scheduling Priority Indicator
derivation, GBR configuration etc) and Radio Bearer
configuration
 Negotiated QoS profile to UE in PDP Context Accept
(if accepted)

11.  In R5: Allows Proxy Call Session Control Function (P-CSCF)
(Policy Decision Function - PDF) to provide dynamically the IMS
application characteristics to GGSN through Go interface
 GGSN acts as an enforcement point (PEP) and thus enables
dynamic QoS control by performing gating on the user data flows
 In R6: The PDF is logically separated from Application Function
(ex, P-CSCF) through Gq interface, allows dynamic QoS support
for non IMS domains
 Also allows other than SIP and SDP application session protocols
to be used
 Allows the possibility to multiplex several service
applications/multimedia components on to the same PDP context.
The PDF remains single and same QoS is applied to all
components.
 NOTE: This is just for information: Clearly 3GPP R5/R6 is a
significant architectural shift from the existing base architecture.
So, proposal is we will not implement this.

12. Subscription Profile AF
Repository
(SPR)
Rx
Sp Policy and Charging
Rules Function
(PCRF)
Online Charging System (OCS)
CAMEL Service Data Flow
SCP Based Gx
Credit Control
PCEF
Gy
GW
Gz
Offline
Charging
System
(OFCS)
NOTE: This is just for information: Clearly 3GPP R5/R6 is a
significant architectural shift from the existing base
architecture. So, proposal is we will not implement this.

14.  Diverse applications and diverse QoS requirements
 For the purpose of QoS management, applications can
be categorized (3GPP R99 categorization) in to:
 Conversational
 Streaming
 Interactive
 Background
 For the simplicity, we would propose maintaining four
distinctive QoS profiles at the Core Network each
correspond to one Traffic Class. These profiles shall be
configurable (be careful when changing since we
would only change to a value that we are sure we
could sustain in our network) and are picked by SGSN
depending on the Traffic Class received

15.  In many a network, operator would like to
prioritize a certain group of users depending on
their subscription/user profiles. For ex, users
could be categorized in to:
 Gold Users: Gold users are users by the virtue of
their subscription gets highest privilege in the
network
 Silver Users: Silver users are next to Gold users in
terms of their treatment (data of course!)
 Bronze Users: Bronze users receive the lowest
attention. Their QoE depends on the
presense/absence of other higher priority users in
the network at any given moment

16.  Priority assignment/differential treatment that
a user’s data flow gets in the network depends
on:
 The application characteristics and/or
 User prioritizations and/or
 Other operator policies
 For ex, we could imagine:
 Case1: An operator policy which is user category
centric
 Case2: An operatory policy which is application
centric

17. • I would propose
to make this
mapping
configurable to
leave the
flexibility to the
customer to
accommodate
their changing
requirements
• For ex, one
could assign
SPI15 to SPI4 to
MC0 to MC11
 A Management Class (MC) is a way that we would like to
treat a user’s data in the network

18. NOTE: Study is pending on the need for the requirement of
PDP context modification procedure and/or Secondary PDP
context support depending on different scenarios (explained
in later slides). For ex,
1) An application is launched when no PDP context exist. This
might lead to a primary PDP context activation with a
proper Traffic Class included
2) An application is launched when a primary PDP context is
already active (for ex, through a former web browsing
session). This might lead to a PDP context modification to
activate/re-negotiate a QoS profile which is the best of both
applications. Another possibility could be that it leads to a
Secondary PDP context activation depending on whether the
UE supports secondary PDP context

19.  In summary, there are practical difficulties in
selecting one of them:
 MS do not request any QoS
 Many MS support single PDP context
 The Application API being open, potential misuse of RT
PDP contexts
 One possibility is to implement non-standard
defined service differentiation in GGSN for ex, by
looking inside the IP flow through Layer4/7
lookup
 Once the user service is identified, then the QoS is
upgraded/downgraded through PDP context
modification. When several IP flows are existing
for the same PDP context, the one with the need of
maximum QoS can be provisioned

21.  Parameters relevant for the QoS treatment in the Core
Network for traffic differentiation are:
• Bandwidth Management and CPU utilization management
are two aspects that can be considered for QoS management
in CNE nodes
• We could consider some BW is allocated for each TC and the
AC will monitor this for the RT/NRT user admission
• Similarly for CPU thresholds can be defined for the AC
purposes for RT/NRT users
• These Topics are FFS

22.  There are three area of QoS treatment in CNE:
 Functions/procedures that implement QoS
inside the CNE nodes, for ex, as described
earlier:
 BW management
 CPU management
 QoS management at CNE edge interfaces for
the Traffic Management
 Interface towards external networks (DiffServe edge
functionality towards external IP network)
 Interface towards Iu, Gn (DiffServ edge functionality
towards radio and backbone network)

23.  SGSN and GGSN will mark the DSCP in the transport
IP header according the the traffic handling priorities
previously defined
 QoS aware (DiffServe supported functionality) routers
in the IP network would examine this DSCP filed and
treat according to the priorities defined
 An example of DiffServ marking table is include in
Appendix – C
 NOTE: We many not need to implement this
functionality of DSCP marking towards Iu since it
would be either a VLAN or LAN between SGSN and
FAP in our network (this is FFS)

26.  The traffic conditioning in GGSN is used to
monitor the ingress traffic characteristics (data
rate) against the negotiated QoS parameters
(Maximum Bit Rate - MBR/Guaranteed Bit Rate -
GBR)
 One typical example algorithm that could be used
(proposed as normative reference algorithm in
3GPP) is a Token Bucket Algorithm:
 The input to the bucket is the incoming bursty data
flow and the out put is the conditioned data flow
at the negotiated rate.

27. OK OK Non-compliant
TBC
L1<TBC L2<TBC L3>TBC
b
b-L1+r* T
b-L1
Time
Notation Used in the Above Algorithm:
 b – Bucket Size
 TBC – Token Bucket Counter (available tokens)
 L1, L2, L3 – Received Packet Lengths
 R – Conditioned Rate/Output rate

28. Release 99 Channels

29.  Transport Channel Selection
 Power Control (Not treated in length in this
presentation)
 Open Loop Power Control
 Fast Power Control (Inner Loop)
 Slow Power Control (Outer Loop)
 Admission Control (R99, HSDPA, HSUPA)
 Load Control
 Packet Scheduling (R99, HSDPA, HSUPA)
 Handover Control (Not dealt in this presentation)

30.  The type of transport channel selected is
influenced by for ex, the QoS attributes and
system performance.
 As an example, the delay sensitive applications
involving conversational traffic class can be
carried over DCH
 One idea would be to make the mapping table
operator configurable

31.  NGBR traffic is always admitted except in the case of
system overload, defined by:
• GBR traffic is admitted when in addition to above condition:
• Where:
– PtxTotal: Current total transmitted power: Available from FAP
common measurements
– PtxTarget: Target transmit power: Known at FAP
– PtxNC: Non Controlled power of GBR users: To be calculated
(TBD)
– DelP: Estimated power for the new GB users (see next Slide)

32.  The downlink power needed for the new GBR user can
be calculated by the open-loop method as follows:
• Where:
• R: Required Guaranteed Bit Error Rate (GBR): Known at FAP
• (Eb/No)dl: Downlink needed to achieve a level of error performance
(BLER). Derived from the service, coverage and capacity planning: To be
Derived (TBD)
• CPICH_Tx_power: Total transmitted CPICH power: Known at FAP
• (Ec/No)cpich: CPICH received Chip energy to Noise power. Available
from UE measurement report
• PtxTotal: Total transmitted power in the cell: Known at FAP
• Alpha: Orthogonality factor: Configurable/Tunable parameter at FAP

33.  An NRT bearer is admitted if:
• An RT bearer is admitted when above condition is fulfilled and if:
• PrxTotal: Received Total Wideband Power (RTWP): Available
from PICO L1 measurements
• PrxTarget: Target Received Total Wideband Power
(RTWPtarget): Known at FAP
• DelI: Increase of wide band power caused by RT user:
Unknown (To be Estimated)
• PrxNC: Non-controlled power received (GBR users + Other
cell users) – Need to see if this is Reference Total Wideband
Power (when reported by PICO L1)

34.  The increase in the received wide band power due to
admitting a GBR user can be calculated using:
 Derivative method:
• Integral method:
• The fractional load increase is calculated by (refer next slide):

35.  Calculate the CDMA uplink load which is the sum of the loads of each user
• Or by using (a simplified equation when the power of user Pk << Iown
NOTE: Above load equations are derived from general CDMA Eb/No notion:

36. • In some exceptional cases (exceptional when system is
designed properly), the system can be overload. The
overload condition is defined by:
 When such overloaded occurs, Load Control is used to bring
the system back under control. Different strategies exist:
 Reduce the NGBR bit rates
 Drop low priority connections
 Reduce GBR bit rates
 Drop GBR connections (for ex, based on SPI)

37.  Assign Radio Resource Priority (RRP) for each
bearer during Admission Control
 Dynamically manage the RRP according to the
served bit rate
 Determine the TFC set according to the QoS
parameters received (for ex, maximum and
minimum bit rates) for GBR and NGBR bearers
 For a GBR bearer, the resource requests are always
treated (provided the Ptarget criteria is met)
 The Pngbr is derived as the power available for
NGBR bit rate scheduling

38.  Different scheduling algorithms possible: Ex,
Fair Throughput Sharing, Fair Resource
Sharing
 In FT, the Pngbr is equally allocated to users
according to their priorities and request arrival
times.
 In FR, the Pngbr is allocated keeping in view of
the signal quality experience by users (for ex,
users nearer to BS/experiencing good
conditions get better share) . This is achieved
by TFC resizing.

39. High Speed Downlink Packet Access
(HSDPA) Channels

40.  Scheduling Priority Indicator (SPI): Indicates
the relative priority of each priority queue.
Range from 0 to 15 (15 being the highest
priority)
 Discard Timer (DT): Time to Live for a received
MAC-hs SDU. The timer is started when the
SDU is received at the queue. Range from
20msec to 7500msec
 MAC-hs GBR: Guaranteed no of bits that the
Mac-hs needs to deliver under normal
conditions.

41.  Non-HSDPA Power: Power of all codes not used
for HS-DSCH and HS-SCCH transmission
including common channels, simply termed as
non-HSDPA power
 HS-DSCH Provided Bit Rate: Number of MAC-d
PDU bits which are successfully transmitted over
last measurement period divided by the
measurement period. This is defined per SPI
 HS-DSCH Required Power: This is the power
required for all users to meet their GBR criteria,
reported per SPI. Non-GBR users are not included

42.  Unused power from DCH users is used for HSDPA scheduling
 In order to make sure that users admitted will meet their GBR
criteria, it is important to dimension the DCH power and HSDPA
power according to the capacity/coverage plans

43.  After transport channel selection, MAC-hs scheduler parameters
defined above are assigned (SPI, GBR (same as the one received)
and DT) as per policies defined.
 These policies could be, for ex, configurable at FAP/CNE, through
a table. The following is such an example template (Note: MAC-hs
GBR for Interactive/Background are optional)
NOTE: The above is just an example: In reality there shall have to be a correspondence between
the MCs defined earlier and the table above as detailed earleir

44.  A new HSDPA GBR user (or a user being
considered for the incoming handover) is admitted
provided the following is satisfied:
• Phsdpa: Total configured HSDPA power: Known at FAP
• Pnew: Estimated power required to satisfy the new GBR user
• Pspi(x): Estimated power required to satisfy already admitted
GBR users whose priority k greater than or equal to the
priority of the new user (priority based AC): Obtained
through PICO L1/L2 measurements
• Pscch: HS-SCCH estimated power: Known at FAP with some
in-accuracy
• P0: Power Offset used to compensate estimation errors for
Pnew, Pspi and Pscch: Configurable/Tunable Parameter

45.  The HS-DSCH SINR can be calculated from the following expression:
• SF16: Spreading Gain factor corresponding to Spreading Factor 16
• Ptx,dsch: HSDPA transmit power: Known at FAP
• Ptx,pilot: CPICH transmit power: Known at FAP
• Ppilot: CPICH Ec/Io: Obtained through UE measurement report
• Alpha: Orthogonality factor: Configurable/Tunable Parameter
• Pxtx,tot: Total Transmit Power: Known at FAP

46.  Calculate the maximum bit rate possible with the derived SINR
figure (possibly through a table stored at FAP obtained through
simulation studies - refer Annex for example)
 Power needed to serve the GBR for the new user is then calculated
as:
• Ptx,dsch: HSDPA transmit power to achieve the throughput f(p):
Known at FAP
• GBRnew: Guaranteed Bit Rate of the new GBR user: Known at FAP
• f(p): Derived throughput for the derived SINR: A priori knowledge
from simulation study

47.  For an incoming handover, the same priority
based admission control algorithm apply. Ie,
 The handover user with SPI x will have priority
admission over all users with SPI <= x
 Usually in a macro scenario, the Admission
Controls can implement special algorithms
during intra-RNC handover procedure to see if
the target cell can accommodate the GBR user.
 In the scenario of FAP no such special
algorithms are possible since the view is
limited to single cell

48.  Same principles as mentioned earlier for R99 load
control apply for HSDPA or for a mixed load
(DCH + HSDPA)
 Drop NGBR connections (note that the NGBR throughput
is autonomously managed by scheduler with
programmed SPI, so nothing much to be done there)
 Reduce GBR configuration for ex based on SPI
configuration
 Possibly drop less priority GBR connections
 Overload control action could start either from DCH or
from HSDPA (which one to be given importance could be
configured through an operator defined parameter)

49. High Speed Uplink Packet Access
(HSUPA) Channels

50.  The power utilization in the uplink when E-DCH
is configured is shown below:

51.  The Admission Control for HSUPA is based on
similar principles are R99 uplink admission (FFS)
 The Guaranteed Bit Rate (GBR) of HSUPA users is
achieved using the non-scheduled grants
 FAP programs PICO L1/L2 and UE with certain
grants (non-scheduled grants). UE uses there
grants to send uplink transmission whenever it has
data. PICO L1/L2 uses this value to reserve the
uplink resources for the GBR users
 Delay and Error performances parameters
derivation, related to HSUPA, from the QoS
attributes are FFS

52.  In conventional networks, the GPRS CN is based
on the IP transport network.
 Different solutions can be used for the service
differentiation
 Differentiated service handling
 Multi Protocol Label Switching (MPLS)
 L2/L1 related methodologies like VLAN feature of
ethernet
 Similar methods cold be used on IuB when the
backhaul is based on the IP transport. If ATM is
used, ATM specific methods like VBR, CBR can be
used for differentiated traffic handling

53.  Quality of Service (QoS) Concept and Architecture
– 3GPP 23.107
 QoS and QoE Management in UMTS Cellular
Systems - David Soldani, Man Li, Renaud Cuny.
John Wiley & Sons, LTD
 QoS Management in UMTS Terrestrial Radio
Access FDD Networks, D.Soldani, Dissertation for
PhD in Technology, Helsinki University of
Technology, October 2005
 Policy and Charging Control Architecture – 3GPP
23.203
 Quality Based HSDPA Access Algorithms – Claus
I. Pedersen, IEEE VTC 2005

54.  Channel Profile: Pedestrian A
 UE Speed: 3 km/h
 Target BLER: 10%
 User positions simulated with Geometric (G) factor defined
as own cell power to interference cell power received at UE
 Used MCS schemes are depicted in the table below

55.  HSDPA Throughput for a single code

57. Traffic class Conversational class Streaming class Interactive class Background class
Maximum bitrate (kbps) <= 256 000 (2) (7) <= 256 000 (2) (7) <= 256 000 (2) (7) <= 256 000 (2) (7)
Delivery order Yes/No Yes/No Yes/No Yes/No
Maximum SDU size (octets) <=1 500 or 1 502 (4) <=1 500 or 1 502 (4) <=1 500 or 1 502 (4) <=1 500 or 1 502 (4)
SDU format information (1) (5) (5)
Delivery of erroneous SDUs Yes/No/- Yes/No/- Yes/No/- Yes/No/-
Residual BER 5*10-2, 10-2, 5*10-3, 10-3, 5*10-2, 10-2, 5*10-3, 10-3, 10- 4*10-3, 10-5, 6*10-8 (6) 4*10-3, 10-5, 6*10-8 (6)
10-4, 10-5, 10-6 4, 10-5, 10-6
SDU error ratio 10-2, 7*10-3, 10-3, 10-4, 10-5 10-1, 10-2, 7*10-3, 10-3, 10-4, 10-3, 10-4, 10-6 10-3, 10-4, 10-6
10-5
Transfer delay (ms) 80 – maximum value 250 – maximum value
Guaranteed bit rate (kbps) <= 256 000 (2) (7) <= 256 000 (2) (7)
Traffic handling priority 1,2,3
Allocation/Retention priority (1)
- Priority Level 1,2, ..., 15 1,2, ..., 15 1,2, ..., 15 1,2, ..., 15
- Pre-emption Capability Yes/No Yes/No Yes/No Yes/No
- Pre-emption Vulnerability Yes/No Yes/No Yes/No Yes/No
Source statistic descriptor Speech/unknown Speech/unknown
Signalling Indication Yes/No