Professional Documents
Culture Documents
approximation
Abstract
This article proposes an access-point and channel selection method for Internet of Things environments. Recently, the
number of wireless nodes has increased with the growth of Internet of Things technologies. In order to accommodate
traffic generated by the wireless nodes, we need to utilize densely placed wireless access-points. This article introduces
a joint optimization problem of access-point and channel selection for such an environment. The proposed method deals
with the optimization problem, using Markov approximation which adapts to dynamic changes in network conditions.
Markov approximation is a distributed optimization framework, where a network is optimized by individual behavior of
users forming a time-reversible continuous-time Markov chain. The proposed method searches optimal solution for the
access-point and channel selection problem on the time-reversible continuous-time Markov chain. Simulation experi-
ments demonstrate the effectiveness of the proposed method.
Keywords
Internet of Things, access-point selection, channel selection, Markov approximation, optimization problem
Creative Commons CC BY: This article is distributed under the terms of the Creative Commons Attribution 4.0 License
(http://www.creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without
further permission provided the original work is attributed as specified on the SAGE and Open Access pages (http://www.uk.sagepub.com/aboutus/
openaccess.htm).
2 International Journal of Distributed Sensor Networks
RSSI-based selection causes another problem in terms proposed an online access-point selection method
of frequency resource utilization of access-points named SmartAssoc that aims at maximizing the mini-
because the RSSI values are independent of channels of mum throughput among users. SmartAssoc considers
access-points. In cases where neighboring access-points dynamic situations. When a new user arrives, the user
use the same channel, we should avoid simultaneously connects to an access-point based on p-norm of current
using the access-points even if those RSSIs are high. loads for access-points. Although these access-point
Frequency resources can be effectively used by selecting selection methods work under dynamic situations, they
an access-point that has a lower RSSI and uses a differ- do not consider a channel selection problem.
ent channel. In this article, we deal with a joint optimization
In the past, several access-point selection methods problem of access-point and channel selection for IoT
have been considered.7–9 Most of those methods use environments, considering dynamic situations. We first
techniques such as mathematical programming and introduce a mixed-integer programming (MIP) for the
heuristic algorithms to select the optimal access-point joint optimization problem of access-point and channel
based on the number of wireless nodes and RSSI levels selection. The MIP represents ideal performance, but it
at a given time.7 These methods must conduct the opti- is difficult to apply the MIP to large-scale problems in
mization process in a centralized manner each time the dynamic situations. We then propose a new access-
number of users changes or any RSSI value is updated. point and channel selection method using Markov
Hence, they are not suitable for access-point selection approximation in order to adapt to dynamic changes in
in situations where network conditions change dynami- network conditions. Markov approximation is a dis-
cally. In public spaces such as train stations and air- tributed optimization framework16 and provides
ports, network conditions frequently change. Typically, approximate solution for optimization problems. In
users stay in public spaces for a short period, and most Markov approximation, a network system is designed
users briefly connect to particular access-points. in such a way that it follows a time-reversible continu-
Especially, in IoT environments, namely, densely ous-time Markov chain, where its states correspond to
deployed wireless networks, network conditions more network conditions. The sojourn time in each state
frequently change because there are a lot of access- depends on the utility value of the network system. By
points and wireless nodes in a small area. Therefore, an doing so, the time average of the utility value
access-point and channel selection method that can approaches the optimal value. The proposed method
flexibly accommodate such dynamic situations is searches approximate solution for the joint optimiza-
needed for IoT environments. tion problem of access-point and channel selection on
To effectively select access-points and/or channels in the time-reversible continuous-time Markov chain,
IoT environments, some methods have been proposed aiming at maximizing the minimum throughput of
in the literature.10–15 Karimi et al.12 have proposed an wireless nodes. To do so, the proposed method repre-
access-point and channel selection method. In this sents combinations of access-points selected by wireless
method, multiple access-points collaborate with each nodes and channels selected by access-points as states
other in order to accommodate users’ traffic. The of the Markov chain.
authors have formulated the proposed method as a The motivation for this article is as follows. Markov
mathematical programming and maximized the approximation optimizes the network configurations
throughput of the users. However, this method is cen- by adaptively changing independent behaviors of users
tralized optimization and does not consider dynamic (i.e. wireless nodes) with time based on minimum neces-
situations. Wu et al.13,14 have proposed channel selec- sary information. It flexibly responds to dynamic char-
tion and access scheduling methods for wireless mesh acteristics of the network conditions caused by arrival
networks and sensor networks. These methods select a and departure of wireless nodes because the network
channel for each wireless node and schedule packet configurations are optimized by just the behaviors of
transmission among wireless nodes based on Latin the mobile users without network operators. Therefore,
squares. Although these methods enhance the through- it is suitable for situations where network conditions
put of wireless nodes, they also do not consider frequently change. In IoT environments, network con-
dynamic situations. Zhu et al.15 have proposed a chan- ditions often change in a given area. In such environ-
nel selection method for sensor networks in IoT envir- ments, as mentioned above, conventional optimization
onments, which utilizes the multi-armed bandit model. techniques used in access-point and channel selection
This method assumes dynamic situations, but does not methods do not work well. They are done in centralized
consider an access-point selection problem. Bai et al.11 control by resolving optimization problems according
have proposed an access-point selection for enterprise to a network condition at a certain time. In these meth-
wireless local area networks (WLANs), which collects ods, network operators collect information on the net-
information on networks and selects access-point based work, and they optimize and configure the whole
on stochastic dominance tools. Xu et al.10 have network, which has large overhead, every time the
Kimura et al. 3
Symbol Meaning
where pz is the percentage of time that strategy z 2 Z(t) With pz , the approximate value of equation (11) is
is selected, p(t) is the 1 3 jZ(t)j vector (p1 p2 pjZ(t)j ), given by
and Fz = mink2K(t) uk (z). X
Markov approximation obtains approximate solu- pz Fz ð14Þ
tion to an optimization problem using the following z2Z(t)
relation
As b increases, the accuracy of the approximation
improves, and the value of equation (14) approaches
maxfw1 , w2 , . . . , wn g
! the optimal value.
1 Xn
We now consider a time-reversible continuous-time
log exp(bwi )
b i=1
Markov chain, where states are elements in Z(t) and
each steady-state probability is equal to pz . In order to
1
maxfw1 , w2 , . . . , wn g + logn construct such a time-reversible continuous-time
b
Markov chain, for each pair of adjacent states
where b is a positive constant, and w1 , w2 , . . . , wn are z, z0 2 Z(t), the local balance equation should be satis-
non-negative variables. From this relation, we have the fied as follows16
log-sum-exponential (LSE) function as follows
pz qz, z0 = pz0 qz0 , z
!
1 Xn
where qz, z0 represents the transition rate from the state z
max(w1 , w2 , . . . , wn ) = lim log exp(bwi )
b!‘ b
i=1
to its adjacent state z0 . One way to satisfy the local bal-
ance equation is to set the transition rate qz, z0 to
When we define wi as
a(t)
qz, z0 = ð15Þ
wi = Fzi , zi 2 Z(t), i = 1, 2, . . . , jZ(t)j exp(bFz )
we have the following approximation where a(t) is a parameter controlling the transition rate.
! As we can see from equation (15), the transition rate
1 X depends on the utility value. Specifically, transition
max Fz ’ log exp(bFz ) ð12Þ rates from states with high utility values to other states
z2Z(t) b z2Z(t) are small, and thus, the sojourn time of the states
becomes large. However, the sojourn time of states with
From the above, we can see that the accuracy of the
low utility values is small. As a result, the time average
approximation depends on b, and we obtain the opti-
of the utility values approaches to the optimal value. In
mal value when b = ‘.
the literature,16 performance evaluations were con-
Owing to the fact that the LSE function is a closed
ducted under static situations with fixed a(t). In
and convex function, we obtain the following approxi-
dynamic situations, however, we should dynamically
mation for equation (11) from equation (12)16
adjust the value of a(t) according to system conditions
X 1 X because the transition rate qz, z0 depends on the number
max pz Fz pz logpz of users.
p(t)0
z2Z(t)
b z2Z(t)
X
subject to pz = 1 Proposed method
z2Z(t)
In our proposed method, each access-point selects a
This is a nonlinear programming problem, and thus, channel and each user selects an access-point according
we can obtain an optimal solution pz by solving the to a time-reversible continuous-time Markov chain on
Karush–Kuhn–Tucker conditions Z(t). To do so, each player follows the flow in Figure 2.
Each player first generates a counter r that is random
Fz b1 logpz b1 + h = 0, 8z 2 Z(t) value according to the exponential distribution with a
P
pz = 1 parameter l(t) depending on the current utility Fz (i.e.
z2Z(t)
minimum throughput of users). For access-points, l(t)
h0
is defined as
where h is a Lagrange multiplier. Solving the above
problem, pz (z 2 Z(t)) is given by a(t)(jCj 1)
l(t) = ð16Þ
exp(bFz )
exp(bFz )
pz = P ð13Þ and for users, l(t) is defined as
exp(bFz0 )
z0 2Z(t)
6 International Journal of Distributed Sensor Networks
Figure 6. Mean utility value: (a) homogeneous scenario and (b) heterogeneous scenario.
Figure 7. Mean number of iterations: (a) homogeneous scenario and (b) heterogeneous scenario.
our proposed method can select access-points effec- throughput. In contrast, in the heterogeneous scenario,
tively in static situations. The proposed method can the maximum throughput of users is different, and thus,
obtain the optimal value when b = ‘, which is the char- the throughput tends to decrease when a user changes
acteristic of LSE function. Therefore, the proposed an access-point. As a result, the number of iterations in
method works well when b is large. the heterogeneous scenario becomes large.
Figure 7 shows the mean number of iterations as a Figure 8 shows the normalized utility value as
function of b. In the homogeneous scenario, the mean a function of elapsed time t, where b = 5.
number of iterations is relatively small, whereas in the ‘‘Instantaneous’’ refers to the instantaneous utility
heterogeneous scenario the mean number of iterations value, and ‘‘Average’’ is the average utility value at
rapidly increases with b. These results imply that the each time. In both scenarios, strategy changes occur
utility value in the homogeneous scenario tends to be more frequently when jK(t)j = 50. The reason is that
larger than that in the heterogeneous scenario. In the strategies with small utility values are more likely to be
proposed method, strategies with high utility values selected when jK(t)j = 50 than when jK(t)j = 20.
remain for long periods because the transition rates Moreover, there is also a higher probability of selecting
from the strategies to other strategies are set to be low strategies with small utility values in the heterogeneous
values (see equations (16) and (17)). However, strategies scenario than in the homogeneous scenario. This results
with low utility values are frequently changed to other in frequent strategy changes in the heterogeneous
strategies. In the homogeneous scenario, the maximum scenario.
throughput that users can achieve is the same for each We then compare the proposed method with the
access-point (i.e. mk, a = 50). Therefore, even if a user RSSI-based method, the simulated annealing method
changes an access-point, it is not easy to decrease the and the throughput-based method. Figure 9 shows the
Kimura et al. 9
Figure 8. Changes in utility value with elapsed time t (static situation): (a) jK(t)j = 20 (homogeneous scenario), (b) jK(t)j = 50
(homogeneous scenario), (c) jK(t)j = 20 (heterogeneous scenario), and (d) jK(t)j = 50 (heterogeneous scenario).
mean utility value and the mean total throughput as proposed method is smaller than that of the RSSI-
a function of b, where jK(t)j = 50. In the figures, based method in the heterogeneous scenario. This
‘‘Proposal (Fairness),’’ ‘‘Proposal (Throughput),’’ stems from the fact that our proposed method ran-
‘‘Simulated Annealing,’’ ‘‘SmartAssoc,’’ and ‘‘RSSI’’ domly selects the access-point among all access-points,
refer to our proposed method, the throughput-based whereas the RSSI-based method selects the access-
method, the simulated annealing method, SmartAssoc point among the access-points with the highest RSSI
method, and the RSSI-based method, respectively. In value. This result implies that the performance of our
SmartAssoc, when new users arrive, they sequentially proposed method could be improved by restricting the
select access-points. We also plot the optimal values of selection of access-point to access-points that have high
the total throughput in the results. As we can see from RSSI value. We left the evaluation of the effectiveness
Figure 9(a) and (c), for large b, the mean utility value of the restriction for the future work.
of our proposed method is larger than that of other To demonstrate that our proposed method is effec-
methods. Specifically, our proposed method can tive under the dynamic situations where the number of
improve the throughput fairness of users. Moreover, users changes dynamically with time, we consider the
from Figure 8, we observe that the total throughput of following situation: jK(t)j = 50 for 0\t 20, 000,
our proposed method is comparable with that of the jK(t)j = 30 for 20, 000\t 40, 000, jK(t)j = 50 for
throughput-based method and the optimal value of the 40, 000\t 60, 000, jK(t)j = 20 for 60, 000\t
total throughput. We also observe that our proposed 80, 000, and jK(t)j = 40 for 80, 000\t 100, 000. We
method outperforms the RSSI-based method, also show the results of SmartAssoc and the simulated
SmartAssoc and the simulated annealing method in annealing method. As for the simulated annealing
terms of the total throughput. These results indicate method, the simulated annealing algorithm is con-
that the proposed method increases not only the ducted every time the number of users in the system is
throughput fairness of users but also the total through- changed. Note that in SmartAssoc, we sequentially add
put. However, for small b, the total throughput of our all the users to the system in random order. Moreover,
10 International Journal of Distributed Sensor Networks
Figure 9. Performance comparison (jK(t)j = 50): (a) mean utility (homogeneous scenario), (b) mean total throughput
(homogeneous scenario), (c) mean utility (heterogeneous scenario), and (d) mean total throughput (heterogeneous scenario).
Figure 10. Changes in utility with elapsed time t (dynamic situation): (a) homogeneous scenario and (b) heterogeneous scenario.
to fairly compare the performance, we modify the player is an access-point, it randomly selects the
SmartAssoc in such a way that players can dynamically channel among C.
change their strategies. Specifically, a player is ran- Figure 10 shows the normalized utility value as a
domly chosen among K(t) [ A according to a Poisson function of elapsed time t, where b = 5. From this fig-
process with rate 0.1 and then the player changes its ure, we observe that the normalized utility of
strategy. When the player is a user, it selects the access- SmartAssoc is stable because users periodically selects
point using SmartAssoc algorithm. By contrast, when access-points with small Lp norm load. We also observe
Kimura et al. 11
that immediately after the number of users jK(t)j 6. Balachandran A, Bahl P and Voelker GM. Hot-spot con-
changes, the utility value of our proposed method gestion relief in public-area wireless networks. In: Pro-
dropped significantly. As the time goes by, the mean ceedings of the 4th IEEE workshop on mobile computing
utility value of our proposed method becomes large systems and applications (WMCSA’02), Callicoon, NY,
and is larger than the utility values of SmartAssoc and 20–21 June 2002. New York: IEEE.
7. Akl RG and Park S. Optimal access point selection and
the simulated annealing method, because the selection
traffic allocation in IEEE 802.11 networks. In: Proceed-
of strategies with large utility values increases. This
ings of the 9th world multi-conference on systemics, cyber-
result indicates that our proposed method can adapt to netics and informatics (WMSCI’05), Orlando, FL, 10–19
sudden changes in the number of users and is an effec- July 2005.
tive method for selecting access-points. 8. Hamamoto R, Takano C, Obata H, et al. An access
point selection mechanism based on cooperation of
access points and users movement. In: Proceedings of the
Conclusion 2015 IFIP/IEEE international symposium on integrated
We have proposed a new method for selecting access- network management (IM’15), Ottawa, ON, Canada,
points and channels using Markov approximation. 11–15 May 2015. New York: IEEE.
Through simulation experiments, we showed that our 9. Koutsopoulos I and Tassiulas L. Joint optimal access
point selection and channel assignment in wireless net-
proposed method can obtain the high mean utility
works. IEEE ACM T Network 2007; 15: 521–532.
value under static situations. Therefore, our proposed
10. Xu F, Zhu X, Tan C, et al. SmartAssoc: decentralized
method is effective at improving the throughput fair- access point selection algorithm to improve throughput.
ness of users. We also showed that our proposed IEEE T Parall Distr 2013; 24: 2482–2491.
method adapts to dynamic changes in the number of 11. Bai Y, Vuran M, Batur D, et al. Stoop: stochastically-
users. In this article, we assumed instantaneous notifi- dominant access point selection in enterprise WLANs. In:
cation of strategy changes throughout the network. In Proceedings of the 2017 IEEE international conference on
practice, however, there is some latency in the exchange smart computing, Hong Kong, China, 29–31 May 2017.
of information. We thus will consider the effects of New York: IEEE.
latency in the future work. 12. Karimi OB, Liu J and Rexford J. Optimal collaborative
access point association in wireless networks. In: Pro-
ceedings of the 2014 IEEE INFOCOM, Toronto, ON,
Declaration of conflicting interests Canada, 27 April–2 May 2014. New York: IEEE.
The author(s) declared no potential conflicts of interest with 13. Wu D, Bao L, Regan AC, et al. Large-scale access sche-
respect to the research, authorship, and/or publication of this duling in wireless mesh networks using social centrality. J
article. Parallel Distr Com 2013; 73: 1049–1065.
14. Wu D, Bao L and Liu CH. Scalable channel allocation
and access scheduling for wireless Internet-of-Things.
Funding IEEE Sens J 2013; 13: 3596–3604.
The author(s) received no financial support for the research, 15. Zhu J, Song Y, Jiang D, et al. Multi-armed bandit chan-
authorship, and/or publication of this article. nel access scheme with cognitive radio technology in wire-
less sensor networks for the Internet of Things. IEEE
Access 2016; 4: 4609–4617.
References 16. Chen M, Liew SC, Shao Z, et al. Markov approximation
1. Al-Fuqaha A, Guizani M, Mohammadi M, et al. Inter- for combinatorial network optimization. IEEE T Inform
net of Things: a survey on enabling technologies, proto- Theory 2013; 59: 6301–6327.
cols, and applications. IEEE Commun Surv Tut 2015; 17: 17. Chen S, Gan X, Feng X, et al. Markov approximation
2347–2376. for Multi-RAT selection. In: Proceedings of the 2015
2. Gazis V. A survey of standards for Machine-to-Machine IEEE international conference on communications
(M2M) and the Internet of Things (IoT). IEEE Commun (ICC’15), London, 8–12 June 2015. New York: IEEE.
Surv Tut 2017; 19: 482–511. 18. Kimura T, Hirata K and Muraguchi M. Dynamic access-
3. Gubbi J, Buyya R, Marusic S, et al. Internet of Things point selection method using Markov approximation. In:
(IoT): a vision, architectural elements, and future direc- Proceedings of the 2017 international conference on infor-
tions. Future Gener Comp Sy 2013; 29: 1645–1660. mation networking (ICOIN’17), Da Nang, Vietnam, 11–
4. Whitmore A, Agarwal A and Xu LD. The Internet of 13 January 2017. New York: IEEE.
Things–a survey of topics and trends. Inform Syst Front 19. Van Laarhoven PJM and Aarts EHL. Simulated anneal-
2015; 17: 261–274. ing. In: Van Laarhoven PJM and Aarts EHL (eds) Simu-
5. Xu LD, He W and Li S. Internet of things in industries: lated annealing: theory and applications, vol. 37.
a survey. IEEE T Ind Inform 2014; 10: 2233–2243. Dordrecht: Springer, 1987, pp.7–15.