Вы находитесь на странице: 1из 40

Thesis Supervisor Group Members

Dr Md. Farhad Hossain  Shahriar Sabrir Khan (1006014)


 Asif Ahmed Khan (1006032)
Associate Professor
Department of Electrical and Electronic Engineering
Bangladesh University of Engineering and Technology
 Motivation
 Thesis Contribution
 System Model
 Simulation Setup
 CDF Analysis
 Outage Probability Analysis
 Throughput Analysis
 Conclusion
 Future Opportunities
 References
DAS – One Base Station (BS),
Many Dispersed Antennas
 Antenna elements (AEs) remotely
dispersed
 Connected to base station through fiber-
optic cable or wireless
 Base station controls resource allocation
(a) Centralized Antenna System
(b) Distributed Antenna system
Why Distributed Antenna System (DAS)?

• Decreases statistical distance of user elements (UEs) to


serving antenna element (AE)
• Improves received SINR
• Better service to cell edge UEs
• Higher Capacity
• Lower Outage Probability
• Lower Power Requirement
• Greater chance of line of sight (LOS) transmission
Why Stochastic Geometry?

• Cellular network parameters are


inherently random
• Better analyzed using probabilistic
models
• Stochastic geometry is perfect to
model randomness of network layout
•Allows for mathematically tractable
and practical analysis
Research Trends in DAS

Uniform distribution of UEs


Equally spaced AEs in circular or hexagonal cell
Random Spatial distribution of AEs with Poisson Point Process
Outage Probability and Capacity analysis
Rural
Area
Is UE distribution truly uniform?

 NO!
Suburban
 Denser around population centers
Area
 Markets, Commercial areas etc.
 Tails off towards suburban and rural
Urban
regions Residential
 Non-uniform UE distribution should be
considered
Area
 Can be modeled as Gaussian
Distribution over the terrain
Commercial Area
Considered Gaussian Distributed Non-Uniform UE Distribution

Proposed Non-Uniform Network Layouts

Evaluated Performance of Proposed Models

CDF Analysis of Received SINR and Statistical Distance of UEs to Serving AE

Outage Probability Analysis

Throughput Analysis
Distribution of parameters

 Poisson Point Process (PPP)


 Homogeneous PPP for Uniform
Distribution
 Inhomogeneous PPP for Non-
uniform Distribution
 Voronoi Tessellation to model
cells
Poisson Point Process (PPP)

The random number of points of Ψ in  Independently scattered


a bounded set 𝑨 ⊂ ℝ𝟐 has a Poisson points
distribution of mean 𝛬 𝑨  Inherently random process

𝒎  Density of points depends on


𝚲 𝐀 parameter
𝑷 𝚿 𝐀 =𝒎 = 𝐞𝐱𝐩(−𝚲 𝐀 )
𝐦!  Can be homogeneous or
inhomogeneous
Homogeneous Poisson Point Process

𝚲 𝐀 𝒎
𝑷 𝚿 𝐀 = 𝒎 = 𝐞𝐱𝐩(−𝚲 𝐀 )
𝐦!
 Density measure for homogeneous
PPP with intensity λ,

𝚲 𝐀 = 𝛌|𝑨|
 Uniform distribution of points over
the space with intensity λ
Inhomogeneous Poisson Point Process
𝚲 𝐀 𝒎
𝑷 𝚿 𝐀 =𝒎 = 𝐞𝐱𝐩(−𝚲 𝐀 )
𝐦!
 Density measure for inhomogeneous PPP
with intensity function λ(y),
𝜦 𝑨 = න 𝝀 𝒚 𝒅𝒚

 Non-uniform distribution of points


 Modeled as Gaussian Distribution over
the terrain
 Intensity function in this case:
𝒏 |𝒚 − 𝝁|𝟐
𝝀 𝒚 = 𝟐
𝒆𝒙𝒑(− 𝟐
)
𝟐𝝅𝝈 𝟐𝝈
Six Network Layouts

Uniform BS with Non-Uniform BS with


No DAS No DAS

Uniform BS with Non-Uniform BS with


Uniform DAS Non-Uniform DAS

Uniform BS with Non- Non-Uniform BS with


Uniform DAS Uniform DAS
Network Layouts

Uniform BS with No DAS Non-Uniform BS with No DAS


Network Layouts

Uniform BS with Uniform DAS Non-Uniform BS with Non-Uniform DAS


Network Layouts

Uniform BS with Non-Uniform DAS Non-Uniform BS with Uniform DAS


LTE with OFDM
 Orthogonal Frequency Division
Multiplexing (OFDM)
 OFDM symbols are grouped into
Resource Blocks (RBs)
 180 kHz in the frequency domain
and 0.5ms in the time domain.
 Universal Frequency Reuse
assumed
 One RB available for each UE
 No intra-cell interference
Path Loss Model
 WINNER+ Non Line of Sight (NLOS) urban macro-cell path-loss model

𝐏𝐋 = 𝟒𝟒. 𝟗 − 𝟔. 𝟓𝟓 𝐥𝐨𝐠 𝟏𝟎 𝐡𝐁𝐒 𝐥𝐨𝐠 𝟏𝟎 𝐝 + 𝟓. 𝟖𝟑 𝐥𝐨𝐠 𝟏𝟎 𝐡𝐁𝐒 + 𝟏𝟒. 𝟕𝟖 + 𝟑𝟒. 𝟗𝟕 𝐥𝐨𝐠 𝟏𝟎 𝐟𝐜


 For ℎ𝐵𝑆 = ℎ𝐴𝐸 = 25𝑚, 𝑓𝑐 = 2𝐺𝐻𝑧
𝑷𝑳 = 𝟏𝟑𝟖. 𝟒 + 𝟑𝟓. 𝟕𝟒 𝒍𝒐𝒈𝟏𝟎 𝒅
 𝒅is distance in km.

Thermal Noise
 Johnson-Nyquistequationfor Thermal Noise
𝑷𝑵 = −𝟏𝟕𝟑. 𝟖 + 𝟏𝟎 𝒍𝒐𝒈𝟏𝟎 𝑩𝑾
 BW is the system Bandwidth
Parameter Value Parameter Value
Area Threshold SINR, 11.1dB
Carrier Frequency 2GHz 384kbps
Channel BW per 5MHz (25 RBs) BS Density per km2
Cell DAS Density per km2
Transmit Power per 20W Variance of AE 15
BS Distribution
Transmit Power per 0.8W Variance of UE 15
RB Distribution
Threshold SINR, -4.1dB Mean Total Number 188
64kbps of BSs
Threshold SINR, 7.9dB Mean Total Number 2740
384kbps of AEs
 Total no of points in inhomogeneous distributions is equal to the
total no of points in homogeneous distributionsfor fair comparison.
 Mean total no of points were calculated with

𝜦 𝑨 = න 𝝀 𝒚 𝒅𝒚

 Standard Deviation of Shadow Fading parameter = 8 dB


 Threshold SINR is -4.1dB, 7.9dB and 11.1dB for 64kbps, 384kbps
and 512kbps service classes respectively
 Monte Carlo Simulation Method was used
 Selective Transmission Scheme: UEs connect to closest AE
SINR
 Signal to Interference and Noise Ratio,

𝑷𝒊,𝒖
𝜰𝒊,𝒖 =
𝓘𝒊𝒏𝒕𝒆𝒓
𝒊,𝒖 + 𝓘 𝒊𝒏𝒕𝒓𝒂
𝒊,𝒖 + 𝓟𝑵

 No intra-cell interference due to OFDM


Outage Probability
 Probability of UE receiving a SINR lower than the threshold SINR
 For a threshold SINR Υth , Outage Probability of a UE,

𝑷𝒐𝒖𝒕
𝒊,𝒖 = 𝑷{𝜰𝒊,𝒖 ≤ 𝜰𝒕𝒉 }

 Network-wide average of all UEs for Overall Outage Probability

Throughput
 Calculated by Shannon Capacity Theory

𝑻𝒉𝒓𝒐𝒖𝒈𝒉𝒑𝒖𝒕 = 𝑩𝑾 ∗ 𝒍𝒐𝒈𝟐 𝟏 + 𝜰𝒊,𝒖


 Averaged for the UEs who are under coverage
CDF of Distance from Serving AE

 Near identical performance for


networks with DAS
 Range 0-1.6km for networks without
DAS
 Range 0-0.6km for networks with
DAS
 Significantly lower service distances
for networks with DAS
CDF of SINR

 Near identical performance for


networks with DAS
 Range -58dB to 77dB for networks
without DAS
 Range -52dB to 128dB for
networks with DAS
 Network with Non-Uniform BS and
Non-Uniform DAS performs best
Varying Threshold SINR
 Networks without DAS give almost
identical results
 70% Outage for threshold SINR of
11.1 dB
 Networks with DAS perform
better
 24-30% Outage Probability
 Non-Uniform BS with Non-Uniform
DAS produces best output
 Only 24% Outage for 11.1 dB
threshold SINR
Varying BS Density

 Nearly constant for traditional BS


networks
 Increases for networks with DAS
 Smaller distances from interfering
AEs
 Increased interference supersedes
higher received power
 Network with Non-Uniformly
distributed BSs and DAS has smaller
service distance and hence lowest
Outage Probability
Varying AE Density of DAS

 Outage performance remains


constant for networks without DAS
 For networks with DAS Outage
decreases with increased AE
density
 Due to smaller statistical service
distances, high SINR
 For AE density 15 per km2, DAS
networks exhibit about 50% less
outage
 Network with Non-Uniform BS and
Non-Uniform DAS performs best
Varying Shadow Fading Parameter
(64kbps Class)

 Outage increases in all cases


 Some UEs go into Outage and some
come out of Outage due to increased
fading parameter
 Most of the users are above threshold
due to smaller threshold SINR
 Greater probability to face Outage
 Networks with DAS perform better
 Combination of Non-Uniform BS and
Non-Uniform DAS yields best results
Varying Shadow Fading Parameter
(384kbps Class)
 Outage increases in networks with
DAS
 Outage decreases in networks
without DAS
 Some UEs Go into Outage and some
come out of Outage due to fading
parameter increase
 More UEs go into Outage for DAS
 More UEs Come out of Outage for
networks Without DAS
 Combination of Non-Uniform BS and
Non-Uniform DAS yields best results
Varying Shadow Fading Parameter
(512kbps Class)
 Outage increases in networks with
DAS
 Outage Decreases in Networks
without DAS
 Same mechanism as 384kbps class
 More Users Go into Outage for DAS
 More Users Come out of Outage for
Networks Without DAS
 DAS network models are robust
against changing Shadow Fading
 Combination of Non-Uniform BS and
DAS Yields Best Results
Varying Threshold SINR
 Networks with DAS perform much
better than networks without DAS
 Networks with DAS and Non-Uniform
BS give highest Throughput
 Initially Throughput increases with
threshold SINR
 After a certain threshold SINR,
Throughput decreases
 Non-Uniform BS distributions
produce higher Throughput due to
higher received SINR
 High SINR is due to lower service
distances from nearest AE densely
populated region
Varying AE Density of DAS

 Initially, all networks are without DAS


 Throughput increases with AE
density
 Networks with Non-Uniform
distributions perform better
 Higher SINR due to smaller distances
from serving AE
 AE density of 15 per km2 yields a
33% increase in Throughput
compared to traditional BS Networks
Varying BS Density

 Interfering AE distance increases with


BS density
 Leads to higher interference
 Interference Surpasses received power
gain due to shorter service AE
distance
 SINR and Throughput decreases
 Networks with Non-Uniform BS exhibit
highest Throughput
Varying Shadow Fading Parameter
(64kbps)

 Throughput increases with shadow


fading parameter
 Throughput of UEs going into Outage
not counted
 Only UEs under coverage contribute
to average Throughput
 Plots almost linear for σ>6dB
Varying Shadow Fading Parameter
(384kbps)

 Throughput increases with shadow


fading parameter
 Throughput of UEs going into Outage
not counted
 Only UEs under coverage contribute
to Throughput
 Plots almost linear for σ>6dB
Varying Shadow Fading Parameter
(512kbps)

 Throughput increases with shadow


fading parameter
 Throughput of UEs going into Outage
not counted
 Only UEs under coverage contribute
to Throughput
 Plots almost linear for σ>6dB
 The proposed model is well-suited for Non-Uniformly
distributed UEs
 Exhibits remarkable performance gains over traditional
systems
Significant reduction of Outage
Notable increase in Throughput
Robustness against varying network conditions

 DAS is conducive to LTE


 Constructing analytical model of inhomogeneous
network layout with DAS
 Designing optimum network layout
 QoS analysis and maximization considering Non-
Uniform UEs
 Integrating thinning processes e.g. Matern
Hardcore with the current model
 A. A. M. Saleh, A. J. Rustako, and R. S. Roman, “Distributed antennas for indoor radio
communications,” IEEE Trans. Commun., vol. 35, pp. 1245–1251, Dec. 1987.
 X.H. You, D.M. Wang, B. Sheng, X.Q. Gao, X.S. Zhao, M. Chen, “Cooperative Distributed Antenna
Systems for Mobile Communications,” IEEE Wireless Communications, vol. 17, pp. 35-43, Jun.
2010.
 W. Choi, J. G. Andrews, “Downlink Performance and Capacity of Distributed Antenna Systems in a
Multicell Environment,” IEEE Trans. wireless Commun., vol. 6, no. 1, pp. 69-73, Jan. 2007.
 J. Zhang and J. Andrews, “Distributed antenna systems with randomness,” IEEE Trans. Wireless
Commun., vol. 7, no. 9, pp. 3636–3646, Sep. 2008.
 D. Stoyan, W. S. Kendall, and J. Mecke, “Stochastic Geometry and its Applications.” Wiley,
Chichester, 2nd edition, 1995.
 M. Haenggi, “Stochastic Geometry for Wireless Networks,” Cambridge University Press, 1st edition,
2013
 “D5.3: WINNER+ Final Channel Models”, Page 75, Wireless World Initiative New Radio WINNER+,
June 2010