A Model for Characterizing Total Interference in Heterogeneous Cellular CDMA

1
A Model for Characterizing Total Interference in
Heterogeneous Cellular CDMA

• Keivan Navaie
• April 2003

2
Outline

• Overview
• Downlink interference modeling
• Effect of traffic characteristics on total
  interference
• Analytical and simulation results
• Analysis and discussion
• Applications
• Conclusion

3
Overview

• Third generation cellular networks will support
  multimedia and data applications
• CDMA seems to be the next generation physical
  layer technology
• CDMA performance is interference limited
• Interference management is a crucial topic in
  wireless cellular CDMA systems

4
Downlink Interference

• Data traffic is asymmetric, therefore downlink
  interference can be a performance bottleneck
• Asymmetric application require to transmit more
  data on the downlink than that of transmit on the
  uplink

5
Effect of Dynamics

• Both the user and the network dynamics affect the
  total interference in a cell
• Network dynamics, is due to the variation of
  traffic load in adjacent cell
• User dynamics include, the moving pattern, the
  channel variations, and the traffic
  characteristics
• It has been shown that air interface dynamics
  could improve network performance
• Users mobility and channel fading Tse01
• Water filling Holtz00,01
• Opportunistic scheduling Liu01

6
Total Downlink Interference Modeling

• To study the variations in the downlink
  interference, we characterize total interference
  in the downlink as a stochastic process
• Time scale for modeling
• We exploit total interference structure at the
  same time scale, when rate control and the
  admission control decisions are made

7
Where are we?
Call admission control Load control Congestion
control
Traffic characteristics
Fast power control Rate control
Channel variations
8
Downlink Total Interference Process

• I(n) is a discrete time series, corresponding to
  the value of the total interference in a window
  of length Tc
• Time resolution is Tc , Tplt Tc ltltTf

• Total interference is
• ?c(n) is the channel gain, and Pc(n) is the base
  station total transmit power

9
Base Station Total Transmit Power

• Transmit power of the base station c is

• Each call is described by
• A call start time ? cij
• J service types

10
Effect of Users Traffic

• Number of interferers
• (Call arrival statistics)
• Period of contribution of each call
• (Call duration statistics)
• Variations in call durations
• (Bit rate statistics)

11
How Traffic Characteristics are Modeled?

• What is the approach for modeling of voice
  traffic?
• Is this approach valid for multimedia and data
  traffic?

12
Traffic Modeling in Wireline Poisson Approach

• Traffic modeling in the world of telephony is
  based on two basic assumptions
• Poisson arrival process
• Poisson call duration
• These assumptions have enabled very successful
  engineering of telephone networks

13
Traffic Modeling in Wireline non-Poisson Approach

• It has been shown that the Poisson models cannot
  accurately characterize
• Poisson assumptions are not valid
• Aggregate traffic trace
• LAN and WAN traffic Paxon, Willinger
• Individual traffic Source
• Source model Digital video traffic Beran
• Source model Web traffic Crovella

14
Example of non-Poisson Call duration Heavy-tail
distribution

• Pareto type call duration
• P?kL(k)k-(?1)

15
Poisson Traffic in Wireless

• Poisson models have been used to model the users
  traffic effect on the total interference vit93
  everit95
• This results in the interference to be a Poisson
  process
• The effect of non-Poisson traffic on the
  network's performance was first pointed out in
  Quraishi97

16
Non-Poisson Traffic in Wireless

• The non-Poisson long range dependent (LRD)
  traffic and its impact on teletraffic efficiency
  of CDMA system have also been studied in Tsyb98
• In Zhang02, it is stated that the total
  interference in a packet based wireless network
  with bursty data traffic is self-similar. The
  self similar model is used to develop a scheme
  that performs rate and admission control.

17
Effect of Traffic Characteristic on the Total
Interference Statistics

• Theorem
• Suppose the EI(n) lt ? and Tf / Tc gtgt1 , we
  define
• as follow
• Then I(n) is as-s if, ?!c such that
• Therefore H1-?/2 and ? min?c.

18

• Timescale of the modeling (Tc) is such that
  relative channel variation is slow.
• Point of modeling is actually between inner and
  outer loop power control
• It is sufficient to be at least one
  non-Poisson traffic for I(n) to be self-similar

19
Call Duration Effect

• Assuming, Tf/Tc gtgt1 and a constant bit rate
  traffic in the call duration, then

• For voice only network
• Total interference I(n) will be short
  range dependent (H0.5)

• For mixed voice and heavy tail traffic
• Total interference I(n) will be long
  range dependent with
• (1/2ltHlt1)

20
Fading and Time Scale Effect

• Assuming Constant bit rate traffic in the call
  duration, then

• For voice only network
• Total interference I(n) will be short range
  dependent H0.5

• For mixed voice and heavy tail traffic
• Total interference I(n) will be long range
  dependent with 1/2ltHlt1 if slow fading, and Tf/Tc
  gtgt1
• Otherwise is short range dependent

21
Intuitions about Downlink Interference
Self-similarity

• Self similarity indicates that there are exist
  extended periods of either strong or weak
  interference
• Periods of high level interference cause the
  quality of service to be degraded
• Also, the existence of periods of light
  interference level, results in network
  under-utilization

22
Simulation Environment

• We have considered a two tire hexagonal cell
  configuration, along with a wrap around technique
  
• Universal Mobile Telecommunication System (UMTS)
  
• 1500/s fast power control
• 12.2 kbps voice (Eb/I05dB), Poisson 5 Erlangs
• 32kbps data (Eb/I03dB), Pareto ?1 1.5, E?1
  2, Fixed bit rate
• 64kbps data (Eb/I02dB), Pareto ?2 1.8, E?2
  1.5, Fixed bit rate
• Tc10 ms
• Slow fading with ?c8dB and Tf100ms
• Uniform spatial distribution
• Theoretically H0.75

23
Simulations Results Total interference
24
Simulations Results Total Interference (contd)
25
Simulations Results Fading Effect
26
Discussion of Results

• In the heterogeneous CDMA network, (with certain
  conditions on the channel and traffic
  characteristics) total interference is
  self-similar
• We showed that, this can be occurred within
  actual network configurations
• The characteristics of self similar process could
  potentially affect the services that require
  tight QoS requirements

• Q How can we use total interference behavior to
  improve system performance?

27
Discussion of Results (contd)

• Self-similarity implies the existence of a
  nontrivial predictive structure
• This structure can be exploited for interference
  management, and system performance improvement

28
An ApplicationNon-real time data scheduling
29
Total Downlink Interference
30
System specifications

• Non-real time data in presence of the background
  services
• Background services voice and multimedia, and
  connection oriented data
• Users predict the value of total interference in
  the next control window using a linear predictor
• Self-similarity in the interference is modeled as
  a fGn process
• fGn is a Gaussian Self-similar process which can
  be specified by H, variance and mean

31
Scenario
Evaluating the total available power for non-real
time users
Non-real time traffic scheduling
User predicts the value of received interference
32
Scheduling

• In each control period Tc, we use time domain
  scheduling
• It is shown that this scheme is throughput
  optimal
• Simulation results also show that this simple has
  bigger admission region

33
Simulation results Admission region
34
Conclusion

• Traffic characteristics affect total interference
• We derived the condition on channel and traffic
  characteristics that cause the total interference
  be as-s
• This characteristic affect the applications with
  tight QoS requirements
• These characteristics can be used for
  multiservice system to improve total system
  throughput

35
A.1 Channel Model

• Distance related attenuation (d-?) along with
  log- normal slow-shadowing
• Gudmunsons slow fading model with standard
  deviation ? and frequency 1/Tf
• ?(n) is a zero mean WGN with variance
  ?2(1?)/(1-?)
• Denote

36
A.2 Outage Characteristics

• It is shown that for Markovian aggregate process
  the overflow asymptotical is
• This change to the following for the case of the
  LRD process

37
A.3 Optimal Linear Prediction

• A self-similar process with given H can be
  parsimoniously represented by a fGn process
• For fGn covariance is for k?Z,
• So the optimal ?-step linear predictor with M tap
  is

38
A.4 Self-Similarity The Mathematics

• Self-similarity manifests itself in several
  equivalent fashions such as
• Slowly decaying variance
• The variance of the sample decreases more slowly
  than the reciprocal of the sample size
• For most processes, the variance of a sample
  diminishes quite rapidly as the sample size is
  increased, and stabilizes soon
• For self-similar processes, the variance
  decreases very slowly, even when the sample size
  grows quite large
• Long range dependence
• For most processes (e.g., Poisson, or compound
  Poisson), the autocorrelation function drops to
  zero very quickly (usually immediately, or
  exponentially fast)
• For self-similar processes, the autocorrelation
  function drops very slowly (i.e., hyperbolically)
  toward zero, but may never reach zero
• Non-summable autocorrelation function

39
Autocorrelation Function
1
Typical long-range dependent process
0
Autocorrelation Coefficient
Typical short-range dependent process
-1
lag k
0
100
40
A.4 Simulation Results Delay Performance
41
A.5 Time Efficiency
42
A.6 Overview (contd)

• Interference management
• _at_ frame level Power and rate control
• _at_ call level Call admissions and load control
• Multiservice CDMA has to support different
  service types each having complicated traffic and
  QoS characteristics

?Multi-Service vs. voice-only network