Sie sind auf Seite 1von 78

Adaptive Spectrum Sharing in TV White Space

Muhammad Asif Raza

Advisor: Prof. Heung-no Lee


Committee Members: Prof. Jalel Ben Othman
Prof. Lim Hyuk
Prof. Hyo-Sung Ahn
Prof. Sang-Seon Byun

Ph.D. Thesis Defense

InfoNet Lab,
School of Electrical Engineering and Computer Science
Gwangju Institute of Science and Technology (GIST)

October 24, 2017


Presentation Breakdown

Introduction TVWS System


Architecture

Two Constraints
QoS Provisioning Two Solutions WSO Accommodation

Problem Formulation Problem Formulation

Algorithm Design Algorithm Design

Simulation & Analysis Simulation & Analysis

Conclusion

2
Achievements

3
Introduction
• Global mobile data traffic growth is forecasted to reach 15.9 Exabyte
per month by 2018 [4].
• TV spectrum, due to low loss and excellent propagation characteristics,
can participate in delivering such voluminous traffic.
• TV channels, not being used by licensed services at a particular location
and at particular time, refer as TV White Space (TVWS) [5].
• Unregulated access to TVWS however, may result in coexistence
issues, for example [5],
– Indiscriminate usage of free resource
– Interference due to dissimilar network technologies
– Diversity in network size
• Therefore, IEEE presented a standard 802.19.1 to regulate access to
TVWS among users operating on dissimilar network technologies [6].
[4] http://www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-networking-index-vni/VNI_Hyperconnectivity_WP.pdf
[5] T. Baykas et. al., “Developing a standard for TV white space coexistence: Technical challenges and solution approaches,” IEEE Wireless Comm., vol. 9, issue 1, pp.
10-22, Feb. 2012.
4
[6] IEEE IT and info. exchange b/w systems—Local and metro. area networks—Specific requirements—Part 19: TV WS coexistence methods, IEEE std. 802.19.1, 2014.
TVWS System Architecture
• WSO represents a white space device
(WSD) or network of WSDs in the
TVWS system.
• TVWS DB contains information on the
allowed TV channels in the TVWS
system.
• CE represents a WSO in the system and
acts as communication bridge between
WSO and CM

• CM provides coexistence services e.g. channel allocation to WSOs.


• Coexistence discovery and information server (CDIS) performs WSOs registration in the
system and provides coexistence discovery service to CMs.
• Coexistence set: for a WSO w, it is set of WSOs who effect the performance of WSO w.
• Neighboring CMs: CMs serving the WSOs in a coexistence set (CS).
5
WSO QoS Provisioning in Channel
Allocation

6
Why QoS provisioning in channel
allocation in TVWS?
• As of 2012, mobile video traffic is more than half of global wireless
traffic [8].
• Similarly, the content delivery networks are expected to carry 71
percent of total internet traffic by 2021 [9]
• Content delivery multimedia applications have stringent bandwidth
requirements to meet the quality of service demand of the users [10].
• Similarly, as an important component, QoS provisioning is required
in some wireless networks, such as the IEEE 802.11e and the IEEE
802.11p networks.
• Most of the TVWS sharing mechanisms in the literature do not
consider QoS requirements of the WSOs during channel allocation
process.
[8] “Cisco Visual Networking Index: Global Mobile Data Trac Forecast Update, 2016-2021,” Tech. report Cisco Inc., San Jose,
CA, USA, Feb. 2017..
[9] “The Zettabyte Era: Trends and Analysis,” white paper, Cisco Inc., San Jose, CA, USA, Jun. 2017.
[10] R. A. Cacheda et. al., “QoS Requirements For Multimedia Services,“ Resource Management in Satellite Networks, ISBN:
978-0-387-36897-9, pp 67-94, New York, NY, USA: Springer US, 2007.
7
Related Work
[11] A max-min fairness algorithm that results in fair TVWS sharing among
neighboring CMs is presented.
[12] TVWS sharing in cellular networks is discussed. The heuristic based
approaches are adopted to defining greedy algorithms to tackle the
identified TVWS sharing problems.
[13] Zhang et al., adapt ecology based species competition model to
develop a coexistence mechanism among heterogeneous networks. The
scheme enables each coexisting network to adjust achieved bandwidth
per its QoS requirements dynamically.
[14] Bian et al., have implemented the concept of FR in sharing a single TV
channel among Cognitive Radios.
In [12], Hessar and Roy have presented an FR method in cellular networks
operating in TVWS. Moreover, the proposed algorithm orthogonalizes
WSOs in frequency domain only.
8
TVWS Sharing Problem
Given a set of available TV channels, a set of CMs with each CM
having at least one WSO registered in it and WSOs’ channel
demands; allocate the TV channels among WSOs such that the
following objectives are achieved.
1. Maximize the system throughput,
2. Minimize unfairness in spectrum allocation among WSOs
registered in neighboring CMs,
3. Fulfill the QoS (throughput) demand of the allocated WSOs.
A coexistence decision making (CDM) system is designed to solve
the above problem and implement a novel frequency reuse method.
9
Proposed CDM System
802.19.1: WSO
1
WSO Registration Clause Requirements CDIS
1 2

(w) , {CS}
= ( , , , , ) (1) Node:1 Requirements Collection

Requirements
Collection 1 ,  ,

Subgradient Frequency
{CS} Interference Y
Algorithm Reuse
2a 2b Matrix 1 Frequency
Reuse
Channel Allocation 2
Subroutine 2
CDM System
X'k
,   ,   ,   ,    ,   ,   ,  
Node:3 Frequency Reuse
Xk
Optimization Problem: Maximize (ST,
Frequency
Fairness in allocation)
Reuse
Subject to: Constraint satisfied.
X'k
X 14
Node:2 Channel10
Allocation
Channel assignment as an
optimization problem: abstract
• Given:
– A set of WSOs, ≔ { , ,…, },
– A set of CMs, ≔ { , ,…, },
– WSO w accessible channels, ≔ , ,
×
– WSO w desired bandwidth, ℬ MHz,
– WSO w channel quality, ≔ , ,
×
– WSO w channel occupancy demand, ≔ , ,
×
– WSO w allocated transmission power, ≔ , ,
×
– WSOs’ channel demands, ≔ , , , , , and
– TV channels in the system, ≔ { 1, 2, ⋯ , }
• Channel allocation problem can be defined as:
• Find an optimal subset of WSOs ( ⊆ ) sharing the channels such that:
– Balanced maximization in fairness and system throughput is achieved.
– Constraint 1 is implemented, i.e., satisfy QoS demand of each allocated WSO.

11
Calculations
• Let WSO w desired QoS (throughput) is defined as:
T , = , log 1 + SINR , ,∀ ∈ ,∀ ∈ (2)
• Let WSO w desired number of TV channels are defined as:

= ,∀ ∈ ,∀ ∈ (3)
• Let WSO w channel capacity is defined as:
, = log 1 + SINR , ,∀ ∈ ,∀ ∈ (4)
• Let , defines the normalized rate achieved by CM c as:
, ,
, = ,∀ ∈ (5)
, + ,

1, if   , = 0,
where defines Kronecker delta function as: :=
,
, 0, otherwise.
, is an indicator variable that if channel j is allocated to WSO w = 1 or not = 0

12
Problem formulation
max.   log  c, j  1 (6a )
cC jJ

subject to x  Z c , c  C
c
(6b)
X j TO j  T j , j  J , c  C (6c)
c T
c T
x1  N   , c  C (6d )
xc  0,1 , c  C (6e)

• (6b) ensures WSO w can operate on channel j if j is an accessible channel of w.


• (6c) ensures the WSOs scheduled in a channel j preserve the system QoS
performance,
• (6d) ensures that the number of channels allocated to the cth CM is restricted by
the number of channels desired by its WSOs.
• (6e) makes the decision variable as binary variable.
For brief intro to proportional fairness, see Appendix A
13
Piece-wise Linear, Relaxed Problem
max. P  X, λ  
X
 
cC jJ
 
F c , j  λ T Z c  x c  (7a )

subject to : X j T O j  T j , j  J , c  C (7b)
c T
c T
x 1  N   , c  C (7c )
x c  0,1, c  C (7d )

where , is linear approximation of original function log , +1 .


See appendix A for linearization.
The parameter : = , is Lagrangian multiplier. The Lagrangian
×
relaxation can be defined as,
h  λ   max P  X , λ  : constraints (7b), (7c ), (7d ) (8)
X

Subgradient algorithm is defined to solve the optimization problem in (8).

14
Subgradient algorithm for relaxed
optimization problem
Step 1: Perform input parameters initialization
Step 2: Given input parameters, solve the relaxed problem and obtain X k .
Step 3: Perform frequency reuse as in Table 2 and get Xk .
Step 3: Use Xk to compute the value of the primal objective function, and fairness index value.
Step 4: Use Xk to compute:
 h 
k
 Subgradient vector of the dual objective function h with λ as: h λ  k
  k , w  ,
 w, j 
 Value of the dual objective function h
define step _ size
 Dual var iable :  
λ k 1  max λ k  step _ size  h,0
Step 5: If step _ size  0.001 or iteration _ number > max_ iterations stop; otherwise go to Step 2.

15
Frequency reuse: Interference
matrix
• Let encoded CS of WSO w is: w, j  I w, m  j  , m  W
– , = 1 if WSO m interferes WSO w on channel j.
– Interference matrix on channel j is:

  I1,2  j   I1, w  j   I1,W  j  


 
    (9)
y  j  :  Iw,1  j  Iw,2  j     Iw,W  j  
 
   
I  j  I  j   I  j    
 W ,1 W ,2 W ,w

• Interference matrix :
Y   y 1 y  2   y  J   (10)

16
Frequency reuse subroutine

Input : W , λ k , Xk  X k ,  , CS.


Output : Xk
Step 1: Given CS, generate encoded CS: w, j , w  W , j  J and interference matrix, Y.
Step 2: Given Xk , update Y matrix as:
1) If channel j is allocated to WSO w, set WSO w entriesin Y equal to zero,or
2) If channel j is allocated to another WSO, m and WSO w interfers it,
set wth WSO entries in the matrix y  j  equal to zero.
Step 3: Define a set of WSOs who do not get channel, W   W .
Step 4: While interference matrix or unallocated WSO set is non-empty, do:
 Given W , solve relaxed problem using linear programming to obtain X k .
 Update Xk as: Xk  Xk  X k .
 Update W  as: remove wth WSO entry in W  if it gets a channel.
 Update Y as in Step 2.

17
Performance measurement
Simulation and comparison

18
Defining variables on subdomains
1. Given occupancy range (0-1); define WSO channel occupancy
demand on the range as,
 Low: ≔ 0, 0.33 ,
 Medium ≔ 0.33, 0.66 ,
 High: ≔ (0.66, 1]
2. Given TVWS, define three sub-domains of as,
 ≔ 0, 33 % ,
 ≔ 33, 66 % ,
 ≔ (66, 100]%
3. Given WSO_set (WSOs that can operate on a channel), define
three levels of Interference as,
 ≔ 0, 33 % WSO_set,
 ≔ 33, 66 % WSO_set,
 ≔ (66, 100]% WSO_set
 Define WSOs channel information on these sub-domains
19
Simulation Setup
• Simulation Parameters:
– Total number of WSOs: W= 32
– Total CM: C= 8
– WSOs per CM = 4
– Number of channels: J= {2,4,6, , 16}
– Allocated channels: ( )= to be determined
• Following comparative metrics are used:
 Fairness in CM allocation
 System throughput
 WSOs satisfaction from the allocation
• Subgradient algorithm is then implemented in MATLAB and
simulation results are shown next.
20
Fairness comparison result

2
 c
 T 
Fairness index value (FIV) calculated on H  T 1 , T 2 , , T C    cC  2 (11)
normalized throughput achieved by CMs as, C Tc  
21 cC
Fairness result discussion
• For low and medium subdomains, FIV result of proposed algorithm with
FR implementation reaches to 1
– It is due to equal share of channels bandwidths among neighboring CMs
• FIV result of Scheme in [17] also reaches to 1 as it also implements FR.
• For high subdomain, FIV of none of the schemes reaches to 1. It is due
to:
– Proportional fairness (PF) in the proposed scheme
• PF balances between fairness and system throughput.
– Orthogonal channel allocation reduces fairness in allocation in comparative
schemes.
– Maintaining lexicographic throughputs in [18] reduces fairness in allocation.
• Fairness in the proposed allocation case is still higher than the
comparative schemes.

22
System throughput gain

Given, WSO channel occupancy demand , , and channel SINR, the system
throughput gain is calculated as, ST     xwc , j Owc , j b log 2 1  SINRwc , j  (12)
cC jJ wW c
23
System throughput result discussion
• The proposed scheme achieves higher ST gain than the
comparative schemes in most cases of the number of channels in
the system.
– It is because, the proposed scheme applies a joint time-frequency FR concept which
accommodates a higher number of WSOs in the available TV channels.
– The Scheme in [17] applies FR concept in frequency domain only.
• The scheme in [17] achieves higher ST gain as number of channels
in the system are low.
– It is because, the Scheme in [17] focus on maximizing the throughput while
proposed scheme implements PF.
• The ST gain in the proposed allocation case is still higher than the
comparative schemes as the number of channels increases.

24
WSO satisfaction

 c

 x 
 jJ w, j 
Graph shows percentage of WSOs  
 nw
satisfied from allocation, calc. as, wW c
25
S  100  W c
(13)
cC
WSO satisfaction discussion
• Proposed scheme and scheme in [18] achieves similar satisfaction
result.
• Proposed scheme vs scheme in [17]: better satisfaction result.
– It is due to implementing FR in joint time-frequency domain that helps satisfy
channel demands of high number of WSOs.
• For low occupancy range, satisfied WSO rate is higher.
– It is due to accommodating larger subset of WSOs on a channel.

26
Algorithm scalability test:
algorithm runtime

• Graph shows effect of different TVWS size on the algorithm execution time.
• Different network sizes are considered for measuring performance under
different load on the algorithm

27
Algorithm complexity discussion
• Algorithm took few tens of milliseconds to complete the
allocation.
– Measured time is on small networks and large networks with small TVWS
• For large, highly dense environment, considering large TVWS
size, it takes about few hundred of milliseconds.
• The reasons are:
– Linprog iterates unless it converges or some objective function or any
constraint tolerance is reached.
• The large number of iterations require comparatively more CPU time.
– Big matrices for occupancy and bandwidth requires considerable time for
the page loading from memory to CPU registers.
• The allocation time for large, highly dense networks is still
acceptable when
– Users are either fixed or do not maneuver frequently.

28
WSO Accommodation in Channel
Allocation

29
TVWS Availability Status in USA

[Courtesy: https://www.google.com/get/spectrumdatabase/channel/]
30
WSO Accommodation in Congested
Spectrum Environment
• Being free to use status of TVWS, each coexisting WSO has an equal
right to have an access to TVWS.
• However, in highly congested spectrum environments, quite a few TV
channels are available for unlicensed transmission.
• On the other hand, a rapid increase in the deployment of wireless
connections and mobile devices is observed; 8 billion in 2016 [16].
• Thus, satisfying the channel demands of wireless networks (WSOs) is
restricted due to limited spectrum availability.
• Consequently, the channel allocation mechanism needs to relax the
channel demands of the WSOs to accommodate as many as WSOs in the
available wireless spectrum (TVWS in this case).
• Meanwhile maximize performance metrics like ST, fairness in allocation,
WSO satisfaction, TVWS utility.
[16] https://www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-networking-index-vni/mobile-white-
paper-c11-520862.html
31
Related Work
In [15] Bansal et. al., define the TV channel sharing problem as a vector of
lexicographic ordering of throughputs of an access point (AP) which is then
transformed into a graph coloring problem.
[17] Bahrak and Park modeled the spectrum-sharing problem as a MOP,
which was then scalarized using a weighted-sum approach and formulated
using a modified Boltzmann machine.
[18] A GA-based reliability model is defined to assign channels to mobile
hosts based on the reliability of the base station. The objective is to enhance
the overall reliability of the mobile network system.
In [19] a new cognitive radio network model with heterogeneous primary
users operating simultaneously via multi-radio access technology is
developed. It uses a GA-based scheme to obtain an optimal solution in
terms of power and bandwidth.

32
TVWS Sharing Problem
Given a set of available TV channels, a set of WSOs registered in
neighboring CMs and WSOs’ channel demands; allocate the TV
channels among WSOs optimizing network performance while
accommodating as many as WSOs in the available TV whitespace.

• We need to define CDM system to implement the TVWS


sharing problem.

33
Proposed CDM System
802.19.1: WSO
1
WSO Registration Clause Requirements CDIS
1
2

(w) , {CS}
Node:1 Requirements Collection

Requirements
Collection 1

EvCo Indicator ∗
Algorithm Function = ( , , , ) (14)
2a Mapping 2b
Channel Allocation 2

CDM System
,    ,    ,   ,   ,   ,
Ind. Fun.
Mapping
MOP: Maximize (ST, Fairness, Satisfaction,
Contiguity, Homogeneity)
Subject to: Constraint satisfied.
EvCo

43
Node:2 34 Channel Allocation
Channel assignment as an
optimization problem: abstract
• Given:
– A set of WSOs, ≔ { , ,…, },
– WSO w desired bandwidth, ℬ MHz,
– WSO w channel quality, ≔ , ,
×
– WSO w channel occupancy demand, ≔ O , ,
×
– WSO w allocated transmission power, ≔ , ,
×
– WSOs’ channel demands, ≔ , , , , and
– TV channels in the system, ≔ { 1, 2, ⋯ , }
– Let us calculate =T , , = and = , as in (2), (3) and (4),
respectively.
• Channel allocation problem can be defined as:
• Find an optimal allocation ( ∗ ) sharing the channels such that:
– Maximization in fairness, system throughput, and WSO satisfaction is achieved.
– Second constraint is implemented, i.e., relaxing WSOs channel demands to accommodate
as many as WSOs.

35
Objective Functions Formulation
1. Fairness in allocation is defined as:
• Let defines fraction of throughput demand of WSO w, then optimizing
= , ,…, by maximally equalizing ≈ ,∀ , ∈
results in fair allocation among heterogeneous WSOs.

Fairness function is defined as minimization function,

2. System throughput (ST) maximization depends upon multiple factors.


a) Contiguous Channel Allocation (CCA): CCA allows a network to have adaptive
channel widths that can increase ST by more than 60% compared to a fixed-width
configuration [20].
• Let be the channel j window time. Let = (0, ], then channel j allocation to WSO
w is defined using indicator function, , . CCA is a minimization function as,

Indicator function, , turns (16) to


zero if single channel or single block
of channels are allocated to WSO w. 36
Objective Functions Formulation
b) WSO Homogeneity: WSOs with same network technologies are encouraged to
share the channel to reduce control overhead.
– Let an indicator function , defines if WSO w and m share the channel j and ,
be defined as cost of sharing channel among them. The homogeneity optimization becomes
minimization function as,

(17)

c) SINR: let be a set of WSOs with maximal channel gain. Then, optimal
throughput of such WSOs is defined as,
(18)
– defines maximum throughput that is obtained by allocating TVWS among WSOs with
maximal channel gain.
3) WSO Satisfaction: defined as energy minimization function as:
(19)

37
Problem formulation

(20a)

(20b)

(20c)

(20d)
(20e)

• (20a) ensures that the total allocated occupancy time on channel j does not exceed the
channel window time Tj .
• (20b) ensures WSO w total allocated occupancy time does not exceed its desired
occupancy time,
• (20c) ensures each WSO gets some minimum channel occupancy on at least one channel,
• (20d), (20e) defines bounds on allocation variable.
38
Issue in Solving Problem in (22)
f2
• Pareto-optimal solution is required. Weak Pareto-optimal
– Objective functions in (20) contradict each
other. A single solution point optimizing all
objective functions is not possible. Feasible Region
• Let P be a feasible solution set defined on Pareto-optimal
the domain of the MOP in (20), Ω = 0,1 ,
the Pareto-optimality is then defined as,

f1

• The dominance concept can be defined as,

39
Solution Approach
• The problem in (20) is a nonconvex, non-linear multiobjective
optimization problem (MOP).
• No algorithm can converge to a global Pareto-optimal solution, at
least in a polynomial time [21].
• An evolutionary strategy (ES) based heuristic technique can
provide an approximate to Pareto-optimal solution quickly.
• Evolutionary algorithm performs ranking on solution points using
some fitness function, to find the optimal one.
• Defining fitness function to MOP in (20) requires us to normalize
the functions in (20) to the same scale, to have equal affect of each
objective functions in (20).
[21] R. Takapoui, et. al., “A simple effective heuristic for embedded mixed-integer quadratic programming,” arXiv
preprint, arXiv:1509.08416, 2015

40
Normalized MOP Definition

(21)

where (22)

41
41
Defining Fitness Function
• A binary epsilon indicator function is used to define fitness function.
• The function performs preference ordering on two sets of solution
points, by establishing Pareto-dominance on the corresponding
objective function vector.
• In simple words it measures the quality of sets of solution points
with respect to each other, defined as follows,
• Let be cluster of solution points, then a set of K
clusters is defined as,
(23)

• Let , 1,2, ⋯ , be the indices to C, then binary epsilon indicator


function [22] applied to MOP is defined on next slide.

[22] E. Zitzler, and S. Kunzli, “Indicator-based selection in multiobjective search,” Springer Lecture Notes in Computer Science, PPSN VIII, vol.
3242, pp. 832-842, 2004.
42
ε-indicator
Binary indicator epsilon [22]
• ( , )= Minimal translation to apply on the set A so that every solution in
set B is dominated by at least as one solution in A.
f2

Normalized space

I ε (B,A)
I ε (A,B)

Courtesy: Matthieu Basseur


f1
Angers, 10 June 2010 43
Defining Fitness Function
(24)

• The indicates denotes the minimum amount, ε, which is required


to improve each objective function

for each member of such that weakly dominates .


• The indicator function applied to MOP in (216) is redefined as max-
min optimization formulation as follows:

(25)

where

44
An Evolutionary algorithm for Coexistence
decision making in TVWS (EvCo)

45
EvCo(continued…)

46
Performance measurement
Simulation and comparison

47
Simulation setup
• Proposed algorithm is compared with FACT [17] and Share [15].
• Following comparative metrics are used:
 Fairness in TVWS sharing
 System throughput
 WSOs satisfaction in normalized rate achieved
 TVWS utility (spectral efficiency)
• Simulation parameters are defined as,
– Number of channels: J= 5,6, ⋯ , 16
– Number of networks: W = {8,16,24,32, , 64}
– Population: P= , ,⋯,
– Number of generations: M=300
– Optimal allocation: ( ∗ )= to be determined
• Algorithm is implemented in MATLAB.

48
Fairness comparison result

Fairness index value (FIV) calculated on normalized throughput achieved by


WSOs, calculated using Eq. (11). 49
Fairness result discussion
• EvCo achieves a higher fairness index value.
– This is due to maximally equalizing normalized throughputs of WSOs.
• The FACT scheme gives the least fairness index value
– It is due to its strict, discriminating allocation policy
– It allocates a channel bandwidth to a WSO unless it is satisfied.
• Share scheme, on other hand, gives better FIV than the FACT.
– It is because, it’s allocation policy is not strict.
– It improves fairness in allocation by sharing the channels among WSOs during
second and third phase of allocation.
• The gap between the proposed and the comparative schemes is huge
at low number of channels which shrinks as number of channels in
the system increases.

50
System throughput gain

The system throughput gain for all three ST   O w, j b log 2 1  SINRw, j  (26)
schemes is calculated as, jJ wW
51
System throughput result discussion
• The EvCo and FACT exhibit competitive behavior.
– It is due to optimizing homogeneity and contiguous channel
allocation .
– These optimizations result in lower scheduling delays and enable
WSOs have adaptive channel widths, thus improving system
throughput.
• Share gives lower ST gain.
– It is due to orthogonal channel allocation in first phase of allocation.
– If WSO with poor channel SINR happens to get channel in the first
phase of allocation, the condition of maintaining lexicographic order
throughputs, restricts other WSOs to share it subsequently. Thus,
reducing ST gain.

52
WSO satisfaction

WSOs satisfaction in terms of normalized rate 1 rw


achieved is calculated as,
S
W

wW d w
(27)
53
WSO satisfaction discussion
• The EvCo achieves the highest average WSO satisfaction as
compared to the comparative algorithms.
– It is due to optimizing the function in (19).
– The EvCo improves fraction of demand serve of WSOs by readjusting
the channel occupancy time of WSOs.
• FACT achieves the least WSO satisfaction from the allocation.
– Unequal channel slot allocation among WSOs results in higher
variation in WSO fraction of demand serve which leads to a
lower overall WSO satisfaction in the system.
• Share achieves better WSO satisfaction than FACT
– It improves fraction of demand serve of WSOs by sharing the
channels during the second and third phase of allocation.
54
TVWS Utility
According to ITU-R,
SE of radio comm. is
defined as,

M
 (28)
B  S T

where M is the
amount of
information
transferred over
distance S using
spectrum width B in
time T.

55
Algorithm complexity in system
execution time
• Graph shows effect
of different TVWS
size on the
algorithm
execution time.
• Different network
sizes are
considered for
measuring
performance under
different load on
the algorithm

56
Algorithm complexity discussion
• EvCo algorithm took few hundred of milliseconds to complete the
allocation.
– Measured time is on a small networks and large networks with small TVWS
• For large, highly dense network, it takes about few seconds.
• The reasons are:
– Evolutionary algorithm adopts large number of iterations unless it converges
or some objective function or any constraint tolerance is reached.
• The large number of iterations require comparatively more CPU time.
• The allocation time for large, highly dense networks is still
acceptable when compared to FACT and Share schemes
– For J channels, W WSOs and T be the number of time slots per channel, the
FACT requires high run-time complexity of ( 2 2 ) to compute the
weight of neurons.

57
Contribution Summary

[1] M. Asif Raza, Zafar Iqbal, Sang-Seon Byun, Hyunduk Kang, and Heung-No Lee, “A Versatile Coexistence Decision-
Making System for Efficient TV Whitespace Sharing among Whitespace Objects," Accepted for Publication in
Wireless Communications and Mobile Computing, Hindawi, Aug. 2017.
[2] M. Asif Raza, S. Park, and H.-N. Lee, “Evolutionary Channel Sharing Algorithm for Heterogeneous Unlicensed
Networks," IEEE Transactions on Wireless Communications, vol. 16, issue 7, pp. 4378-4389, Jul. 2017.
[3] M. Asif Raza, Zafar Iqbal, Seungchan Lee, Haeung Choi, and H.-N. Lee, “Issues and Resolution Efforts Pertaining to
TV Whitespace Usage," Korea Info. And Commun. Society Global Conference 2014, pp. 77-79, Korea, Jan. 22-24,
2014
58
FUTURE WORK

59
Traffic oriented channel allocation
• WSO QoS (throughput) demand and WSO accommodation in the
TVWS is covered in the thesis.
• The TVWS coexistence environment is however, anticipated to be
potentially more complicated due to mix of WSOs with variable
network technologies and varying traffic characteristics of host
applications.
• The proposed work needs to be extended to accommodate the traffic
characteristics in TVWS sharing problem formulation.
• The TVWS sharing problem in such a diverse, heterogeneous
coexistence environment can be modeled as mixed integer
programming (MIP) problem.
• In this case, the WSOs channel requirements necessary to fulfill can be
treated as binary allocation variables while relaxed requirements can be
treated as real valued variables.
• A channel allocation algorithm suitable for MIP problem then needs to
be defined.

60
Conclusion
• Developed a realistic, coexistence decision making (CDM) system for channel allocation
among WSOs operating on dissimilar technologies in TVWS.
1. The CDM system adapts the WSO demands in defining TVWS sharing problem and
make use of proportional fairness method:
– Maximizes in balance the fairness vs system throughput
– Fulfills the WSO channel demand satisfaction constraint in allocation.
2. The CDM system focuses on accommodating as many as WSOs in the scarce TVWS:
– The TVWS sharing problem is defined as multiobjective optimization problem (MOP) with
each objective function optimizes a critical coexistence requirement in TVWS sharing process.
– Transform the MOP into max-min optimization problem using binary epsilon indicator
function
– A better approximation of global minima of the TVWS sharing MOP is achieved.
• Performance results:
– The proposed algorithms runs in an acceptable amount of time.
– System throughput in proposed schemes is lower than the comparative algorithms for some
cases
– Higher fairness in allocation, WSO satisfaction and TVWS utility is achieved than that of the
comparative algorithms.
61
Thank You

62
APPENDICES
Appendix A: Proportional Fairness
max  log( )  

subject to ≤ ,∀ ∈

 Rates {xs} are proportionally fair if for any other feasible rate {ys}

max  ≤0

 This means that a vector of rates = ( ; s ∈ S) is proportionally fair if


it is feasible(i.e. ≥ 0 and ≤ C) and
 If for any other feasible vector , the aggregate of proportional
changes is zero or negative.
64
Appendix B: Linearization using
tangent plane approximation

65
Appendix C: Sugradient algorithm
in Detail
Step 0: a) Choose an initial λ 0 ,
b) Set parameters: = 2, min = 0.001,iter = 0, maxiter =5, k = 0,k max =10, F best = 0, h best =-, hupper =0, and Xk   0W  J .
Step 1: a) Increment : k  k + 1, iter = iter + 1.
b) Given λ, solve the relaxed problem and obtain X k .
Step 2: Adjust X k such that, set xwc , j = 0 if zwc , j  0.
Step 3: Perform frequency reuse as in Table 2 and get Xk .
Step 4: a) Use Xk to compute the value of the objective function F , and fairness index value H .
b) If F  F best , set F best  F , h upper  F best and X  Xk .
Step 5: Use Xk to compute:
 h 
 
 Subgradient vector of the dual objective function h with λ k as: h λ k   k , w , w, j
 w, j 
Value of the dual objective function h

step size: tk 

 hupper  h  λ k  
2
 
h λ k

   
 Dual variable: λ k 1  max λ k  tk h λ k , 0

   
Step 6: If hbest  h λ k , then hbest  h λ k elseif iter   max iter then   max  
2 
,  min and iter = 0.
Step 7: If tk  0.001 or k > k max stop; otherwise go to Step 1.
66
Appendix D: Subdifferential
method
• Gradient based method cannot be applied to the proposed dual
problem ℎ .
– It is due to discontinuity at adjacent pieces of linear objective function
• Therefore, subdifferential method is used to solve it.
• Subdifferential of Lagrangian ℎ at intersection of adjacent
linear pieces can be defined as,
– ℎ = ℎ ⁄ , ℎ ⁄ ,⋯, ℎ ⁄

where ℎ ⁄ = , − ,      = 1,2, ⋯ ,

• These subgradients are then solved to get feasible solution vector

67
Algorithm scalability test setup
• Algorithm performance is judged, in terms of time taken to
perform the channel allocation.
• Following parameters are used to judge the performance.
– Simulation Parameters:
• Number of channels: J={4,8,12, ⋯ , 48}
• Number of networks: W={8,16,32,64,128}
• Number of CMs: ={2,4,6, , 16}
– System configuration:
• Processor: Intel core i5-2500k CPU 3.30 GHz
• Total Memory: 8.00 GB
• Memory available = 3.36 GB
• MATLAB® tic-toc stopwatch timer is used to compute simulation
timespan.

68
Appendix E: Objective Functions
Formulation
• Let us calculate =T , , = and = , as in (2), (3) and
(4), respectively.
1. Fairness in allocation is defined as:

• Optimizing = , ,…, by maximally equalizing ≈


,∀ , ∈ results in fair allocation among heterogeneous WSOs.
• Fairness function is then defined as,

(15)

69
Appendix E: Objective Functions
Formulation
2. System throughput (ST) maximization depends upon multiple factors.
a) Contiguous Channel Allocation (CCA): CCA allows a network to
have adaptive channel widths that can increase ST by more than 60%
compared to a fixed-width configuration [20].
• Let be the channel j window time. Let = (0, ], then channel j
allocation to WSO w is defined using indicator function as,

• CCA then becomes energy minimization function defined as,

(16)

[20] R. Chandra, et. al., “A case for adapting channel width in wireless networks,” in proc ACM SIGCOMM, Seattle, USA, Aug. 2008.
70
Objective Functions Formulation
a) Contiguous Channel Allocation (CCA): continued
• The updated solution metric, , is defined by concatenating a
zero column on both, the leading and trailing end of the solution
matrix O, i.e.,

• The function , forces function in (18) to be zero if single


channel or a single block of contiguous channels allocated to WSO
w, defined as,

71
Appendix E: Objective Functions
Formulation
b) WSO Homogeneity: WSOs with same network technologies are
encouraged to share the channel to reduce control overhead.
• Let , be defined as cost of sharing channel between WSO w
and m, defined as,

where represents MAC technology of WSO w and be its control


overhead.
• The homogeneity optimization becomes minimization function as,

(17)

72
Appendix E: Objective Functions
Formulation
c) SINR: let a set of WSOs with maximal channel gain is defined as,

The optimal throughput is then defined as,

Throughput optimization then becomes minimization function as:


(18)

3) WSO Satisfaction: defined as energy minimization function as:

(19)

73
Appendix F:
An Evolutionary algorithm for Coexistence
decision making in TVWS (EvCo)

74
EvCo (Continued…)

75
EvCo (Continued…)

76
EvCo Calculations
• Updating WSO occupancy if (22a) is violated:

(35)

• Updating WSO occupancy if (22b) is violated:

(36)

• Updating WSO occupancy on channel j, if (22c) or (22d) violated:


O , = (37)
• Defining angular similarity among solution points:
(38)

where O and O are 1-D transformation of O and O , respectively.


77
Bibliography
[11] IEEE Standard for Information technology- Telecommunications and information exchange between systems- Local and
metropolitan area networks- Specic requirements- “Part 19: TV White Space Coexistence Methods," IEEE Std.
802.19.1, May 2014.
[12] F. Hessar and S. Roy, “Resource allocation techniques for cellular networks in TV white space spectrum,” in Dynamic
Spectrum Access Networks (DYSPAN), 2014 IEEE International Symposium on, April 2014, pp. 72–81.
[13] D. Zhang, Q. Liu, L. Chen, W. Xu, “Ecology-based Coexistence Mechanism in Heterogeneous Cognitive Radio
Networks,” IEEE GLOBECOM, 2015.
[14] K. Bian, J. M. Park and B. Gao, “Frequency reuse over a single TV white space channel,” USA, Elsevier, pp 77-100,
2014.
[15] T. Bansal, D. Li, and P. Sinha, "Opportunistic Channel Sharing in Cognitive Radio Networks," IEEE Trans. on Mobile
Computing., , vol. 13, no. 4, pp. 852-865, Apr. 2014
[17] B. Bahrak, and J.-M. J. Park, “Coexistence decision making for spectrum sharing among heterogeneous wireless
systems,” IEEE TWC, vol. 31, issue 3, pp. 1298-1307, Mar. 2014.
[18] L. M. O. Khanbary, and D. P. Vidyarthi, “Reliability-based channel allocation using genetic algorithm in mobile
computing,” IEEE Trans. Vehicular Tech., vol. 58, no. 8, pp. 4248-4256, Oct. 2009.
[19] Y. Jao, and I. Joe, “Energy-efficient resource allocation for heterogeneous cognitive radio network based on two-tier
crossover genetic algorithm,” Jr. Commun. Networks, vol. 18, no. 1, pp. 112-122, Feb. 2016.
[20] R. Chandra, et. al., “A case for adapting channel width in wireless networks,” in proc ACM SIGCOMM, Seattle, USA,
Aug. 2008.

78

Das könnte Ihnen auch gefallen