Fair Resource Allocation with Guaranteed Statistical QoS for Multimedia Traffic in Wideband CDMA Cel

1
Fair Resource Allocation with
GuaranteedStatistical QoS for Multimedia
Trafficin Wideband CDMA Cellular NetworkLiang
Xu, Member, IEEE, Xuemin (Sherman) Shen, Senior
Member, IEEE, andJon W. Mark, Fellow, IEEE

• Presented by Ahmed Abdelhalim

2
Motivation

• The need of a Dynamic Fair Resource Allocation
  Scheme for Wideband CDMA taking into account
• The characteristics of channel fading
• Intercell Interference

3
Dynamic Resource Allocation

• Dynamically vary the resource shares of
  applications according to
• Variation of traffic load
• Channel Conditions
• Why?
• To achieve efficient resource utilization
• Increase network throughput

4
Problem

• We have to ensure fairness while efficiently
  supporting QoS for multimedia traffic in Wireless
  networks.
• It is a tradeoff between wireless resource
  efficiency and level of satisfaction among users

5
Conventional Time-scheduling Approach

• Used in based and hybrid time-division/code-divisi
  on multiple access (TD/CDMA)-based wireless
  networks
• Requires high complexity due to intensive
  computation for the virtual time of each packet

6
Opportunistic Scheduling Approach

• Allocates more bandwidth to MSs with good
  channels conditions
• Adv.
• It takes into account the channel fading
  conditions
• Dis.
• Lacks short-term fairness

7
A Generalized Processor Sharing Approach (GPS)

• Allocates the resources according to weights
  assigned to the users.
• The user rates are optimized for each scheduling
  period.
• It guarantees the required minimum channel rates
  and adapts to the changes in channel conditions.

8
Proposed Work

• a dynamic fair resource allocation scheme which
  efficiently support both real-time and
  non-real-time (multimedia) traffic with
  guaranteed statistical quality of service (QoS)
  in the uplink of a wideband code division
  multiple access (CDMA) cellular network.
• The Scheme uses GPS to allocate channel resources
  taking into account the characteristic of channel
  fading and intercell interference

9
System Model

• Direct Sequence code division multiple access
  (DS-CDMA)
• A data signal at the point of transmission is
  spread over a wider frequency spectrum according
  to the Spreading Factor
• This is achieved by the MODULU-2 addition of
  Pseudorandom sequences to each data bit.

10
Rate-Scheduled DS-CDMA System mechanism
11
The Mechanism

• Each MS generates a sequence of packets which
  enter a buffer after error control coding.
• The packetized information sequence is then
  converted to a DS-CDMA signal and transmitted
  over the wireless channel to the base station
• The BS uses a single-user Rake receiver to detect
  the signal from each MS.
• The channel rate is dynamically allocated by the
  rate scheduler at the BS.
• The spreading factor is adjusted according to the
  scheduled channel rate.

12
The Mechanism (cont.)

• An SIR threshold corresponding to the target BER
  and the allocated channel rate is set at the
  receiver side.
• When a new channel rate is scheduled, the SIR
  threshold is adjusted accordingly.
• The home BS measures the received SIR and
  compares it with the SIR threshold.
• When the actual SIR is lower (higher) than the
  threshold, a feedback control signal is sent to
  the transmitter to increase (decrease) the
  transmission power.

13
Channel and Traffic Models

• Multi-path Fading Channel
• Pij PTi rij -µ 10ij/10Xij
• Where
• Pij The received power at the jth BS
• PTi Transmitted power of mobile i
• rij the distance between the ith mobile and the
  jth BS
• µ The loss exponent
• ij Gaussian random variable with zero mean
• Xij characterizes the multipath fading between
  MSi and BSj

14
Traffic Model and QoS Requirements

• The GPS discipline is applied in resource
  allocation for both real-time and non-real-time
  traffic.
• The bandwidth requirements of all the flows are
  characterized by positive real numbers Ø1 Ø2 .
  . . Ø N.
• GPS fairness is achieved when the following
  inequality holds
• Si(t1,t2)/Sj(t1,t2) Ø1/Ø2
• Where
• Si(t1,t2) The amount of service received by the
  flow i in an interval (t1,t2)

15
Resource Allocation for Real-Time Traffic

• Real-time traffic is scheduled before
  non-real-time traffic in each scheduling period
• The resource allocation is carried out jointly
  via fair scheduling and admission control
• The admission control assigns fixed weights to
  real-time traffic so that the required
  statistical delay bounds can be guaranteed

16
The Delay Bound

• Each traffic source is shaped by a Leaky-Bucket
  regulator. i.e.
• Permits are generated at a fixed rate
• Packets are only released in the network when a
  permit is available
• It is specified by a unique 5-tuple(Rm,i, si, Di,
  Li)
• Where
• ?i Permits rate Rm,i constraint on
  the peak rate
• s i token buffer size Di Required delay
  bound
• Li losses due to transmission errors and
  excessive queuing delay
• The QoS requirement is
• Pti (t) gt Di lt Li
• where ti (t) is the actual delay for user I
  traffic arrived at time t

17
Static-weight code division GPS Scheme (SW-CDGPS)

• The scheduler checks the total bandwidths
  requests from all MSs.
• If it doesnt exceed the channel capacity then
  each user is granted the requested BW. Otherwise
  the resource allocation vector (S1, S2 ,.,Sn)
  is computed iteratively
• Remaining capacity channel capacity Total BWs
  granted in the previous iteration
• A fair share of the remaining capacity is
  calculated for each MS whose requested bandwidth
  has not been granted, according to its GPS weight

18
SW-CDGPS (cont.)

• For the user whose requested bandwidth is less
  than its fair share of the remaining capacity
  computed in any iteration, its request will be
  fully granted otherwise, the users Svi will be
  determined in a later iteration.
• In the final iteration, each remaining MS will be
  given a fair share of the remaining capacity, but
  no more than its requested bandwidth.

19
Fairness Indes

• For non-real-time traffic we define the Fairness
  index Fim
• Fim Si(t1,t2)/Ød,i - Sm(t1,t2)/ Ød,m
• Si(t1,t2)/ Ød,i
• The fairness index indicates the difference
  between the services (normalized by weights)
  received by the two MSs.
• A Fairness bound Ø is guaranteed with probability
  T
• PFim gt Ø lt T

20
Dynamic-weight code division GPS Scheme (DW-CDGPS)

• For non-real-time traffic without stringent delay
  requirements, throughput can be made beneficial
  by dynamically adjusting the assigned weights
  according to channel conditions.
• The amount of change in the variable weight
  possible per operation varies directly with the
  parameter ?
• The larger the ? the looser the statistical bound
  can be guaranteed

21
Simulation

• The target cell and its 18 neighboring cells in
  the first and second tiers are simulated
• MSs are uniformly distributed in the service
  area.
• Each Rayleigh fading channel with a maximum
  Doppler shift of 5Hz is generated using the
  Jakes simulator

22
Hexagonal layout of cells
23
Performance Parameters

• Delay
• Throughput
• GPS Fairness

24
Simulation 1

• each cell has 12 homogeneous greedy data users.
• Three cases, where ?1 10 100,respectively,
  are simulated for the DW-CDGPS scheme, compared
  to the SW-CDGPS scheme with static weight.
• The fairness index Fim is measured once every 20
  slots in the simulation run for a total of 30,000
  slots

25
Results 1
26
Comments

• It can be seen that the statistical fairness
  bound can be regulated effectively by adjusting ?
• When ?0 the scheme reduces to the static weight
  scheduling scheme

27
Simulation 2

• 12 homogeneous Poisson data traffic flows are
  simulated
• delay and throughput performances of users in
  cell 0 are compared.
• For the DW-CDGPS, ? is set to be 100,
  corresponding to a large fairness bound
• The traffic load is defined to be the sum of
  average arrival rates of all data flows. The
• Throughput and traffic load are normalized by
• the maximal achievable throughput under
• SW-CDGPS when traffic load is high.

28
Results
29
Comments

• It is shown that the DW-CDGPS scheme can improve
  the maximal uplink throughput significantly
  compared to the SW-CDGPS scheme
• The throughput gain is seen only when traffic
  load is high since when the traffic load is low,
  the throughput equals to the total traffic
  arrival rate.

30
Results
31
Comments

• It can be seen that, when the traffic load is
  high (around 1), the average delay can be reduced
  up to 70 percent, by DW-CDGPS as compared to the
  SW-CDGPS.
• This implies that the short-term unfairness
  introduced by the DW-CDGPS can actually benefit
  the delay performance in the uplink when traffic
  load is high due to a high throughput that can be
  achieve
• When traffic load is low, the delay performance
  can still be improved slightly

32
Conclusions

• An efficient dynamic fair resource allocation
  scheme has been proposed for supporting
  multimedia traffic in the uplink of wideband CDMA
  cellular networks with QoS satisfaction.
• The proposed scheme guarantees a statistical
  delay bound for real-time traffic and a
  statistical fairness bound for non-real-time
  users
• The proposed scheme allows for a flexible
  trade-off between the generalized processor
  sharing (GPS) fairness and efficiency in resource
  allocation and is an effective way to maximize
  the radio resource utilization under the fairness
  and QoS constraints.