Congestion Detection, Prevention and Avoidance ...

3 downloads 0 Views 975KB Size Report
J. Sarnecki, C. Vinodrai, A. Javed, P. O. Kelly, and K. Dick,. Microcell design principles. IEEE Communications Magazine 31, 76. (1993). 25. V. Chandrasekhar ...
Copyright © 2013 American Scientific Publishers All rights reserved Printed in the United States of America

Journal of Computational Intelligence and Electronic Systems Vol. 2, 1–19, 2013

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent, Energy and Spectrum Efficient Green Mobile Network Anwesha Mukherjee∗ and Debashis De Department of Computer Science and Engineering, West Bengal University of Technology, BF-142, Sector-I, Salt Lake City, Kolkata 700064, West Bengal, India

Keywords: Traffic Load, Blocking Probability, Congestion Control, Power Consumption, Cell Breathing, Path Loss.

1. INTRODUCTION Network traffic forecasting is a critical element of initial network deployment and dynamic optimization of network capacity. For better traffic forecasting better prediction of user movement pattern is required. For better prediction of user movement pattern, two kinds of information are recorded:1 i) Long term: contains slow changing information of the user which is concerned of only the long periods of time such as weeks, months. ii) Short/medium term: contains information which represents the recent behavior of the user (day hour). A mobile cell needs to make short-term predictions for all of its active users, including their movements. In Personal Communication Services (PCS) network, the service area is divided into a number of location areas where each location area (LA) consists of a number of cells. Each cell contains a base station (BS). A macrocell with radius approximately 1–10 km provides radio coverage served by a high power cellular BS ∗

Author to whom correspondence should be addressed.

J. Comput. Intell. Electron. Syst. 2013, Vol. 2, No. 1

which in turn causes poor coverage in indoor regions e.g., home, office, shopping malls etc. Previously, repeater was used for quick, cheap coverage inside buildings although it drains capacity from the macro network, distorts the cell, and creates interference. In such situation, picocell is introduced which adds capacity to the network avoiding cell distortion and interference. Picocell with radius approximately 20–200 m is a small cellular base station covering small area. Microcell is also developed to deal with the same problem. A microcell is larger than a picocell, though hardly distinguishable. A microcell contains a low power cellular base station, covering a limited area such as a hotel, shopping mall or a transportation hub. Microcell has radius of approximately 200 m–1 km. The latest development for reducing power consumption in a cellular network is Femtocell which is a low-power wireless access point that operates in licensed spectrum.2–4 Femtocell connects standard mobile devices to a mobile operator’s network using residential DSL or cable broadband connections.2–4 The Femto Access Point (FAP) provides cellular access in indoor environment and connects to the

2326-3008/2013/2/001/019

doi:10.1166/jcies.2013.1044

1

RESEARCH ARTICLE

This paper has proposed a congestion control approach through congestion detection, prevention and avoidance for an intelligent, energy and spectrum efficient green mobile network. The network is trained to make it an artificially intelligent one thus it can predict the number of users visiting each cluster and offered traffic load on the cluster in a location area automatically based on the mobility information of the users maintained in the dynamic location area list in the profile of each individual user. Depending on the number of users and the blocking probability determined by the intelligent network, congestion is detected, prevented or avoided. For congestion control, cell size in each cluster i.e. macrocell, microcell, picocell or femtocell to be allocated is decided based on the number of users and blocking probability considering both new and handoff call in the cluster. Power consumption model in such macrocell-microcell-picocell-femtocell based location area is proposed in this paper and compared to that of a macrocell based location area and cell breathing which is an existing congestion control method. Simulation results present that using macrocell-microcellpicocell-femtocell based network 35.29–40% and up to 25% reductions are achieved in power consumption and average call blocking probability respectively than only macrocell based network.

RESEARCH ARTICLE

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

operator’s network through the customer’s own Broadband connection to the Internet. Due to small size, low cost and high performance, femtocells are highly desirable for radio access in cellular network.3 Although femtocell has low cost, its functionality is almost same as that of typical macro BS. A femto BS can serve a small number of users at relatively low transmission power.5 Residential and enterprise femtocell can support up to 8 to 12 and 16 users respectively in its range. A femtocell has a cell radius of approximately 10 m. Femto BSs can be easily installed in a plug-and-play manner.5 Moreover femtocell is self-organized and portable. As it does not have to send the signals over a long distance, there will be reduction in the total transmission power compared to that of a picocell and macrocell, thus providing greener cellular network. Spectrum and energy efficiencies are among the most important venues for technological advances in mobile networks. Reduction of energy consumption by the base stations is essential to prevent environmental pollution and carbon emission. Spectrum efficiency is also important to prevent wastage of spectrum. To provide good QoS even during congestion is a challenge in mobile computing. Cell breathing is a well-known technique to deal with congestion in mobile network but causes high power consumption by base stations.6 Channel borrowing, priority based channel allocation, buffer maintenance, channel reservation, dynamic call admission control are other well-known approaches in the field of congestion control.7–11 But most of these methods cannot be applied if most of the adjacent cells of an overloaded cell also suffer from congestion. Moreover delay is introduced which degrades QoS and may cause call dropping. Thus a new congestion control scheme is required to be introduced that will achieve low power consumption, spectrum efficiency, delay minimization and will be independent of the load status of other cells. In such scenario, this paper has proposed a green network development strategy where spectrum and energy efficiency both can be achieved simultaneously as well as congestion can be controlled without increasing the power consumption. The cell size of a location area in the proposed network is decided based on the traffic load offered to each cluster of the location area. Each location area contains a number of clusters. We have first considered a macrocell based network where each cluster of a location area is covered by a macrocell. Then the number of users visiting each macrocell, the traffic load offered to the macrocell and the blocking probability (considering probability of blocking a new call and dropping a handoff call both) in each macrocell is predicted based on the users mobility information maintained in dynamic location area list in the profile of individual user. Depending on the number of users and the blocking probability, congestion is detected, prevented or avoided. For congestion control, cell size in each cluster is decided 2

Mukherjee and De

based on the number of users and the blocking probability. Hence macrocell, microcell, picocell and femtocell based location area is developed. An analytical model of power consumption in such macrocell, microcell, picocell and femtocell based network is calculated and compared to that of a macrocell based network and cell breathing. Based on users mobility information spectrum is allocated to reduce its wastage. The organization of the paper is as follows: Section 2 contains related works, Section 3 contains the proposed method for predicting number of users, traffic load and the blocking probability in a cell and spectrum allocation based on the predicted traffic load, Section 4 presents the proposed congestion control approach including congestion detection, prevention and avoidance methods, Section 5 presents power consumption model and path loss in proposed network, Section 6 presents results and discussions on the performance analysis of the proposed network, Section 7 contains conclusion.

2. RELATED WORKS Several approaches have been proposed on spectrum utilization and traffic forecasting in mobile network. To reduce the traffic due to mobility management by predicting user movement pattern, a profile is maintained for each user to record his most probable mobility patterns in Ref. [1]. An improvement of borrowing channel assignment (BCA) for patterned traffic load by using the short-term traffic prediction ability of cellular probabilistic self-organizing map (CPSOM) is proposed in Ref. [7]. Mobile motion prediction (MMP) algorithm based on movement circle, movement track and Markov chain model is proposed to determine the motion of mobile as well as to represent the daily route of a user in Ref. [12]. A mobility model in three-dimensional PCS environment is proposed based on user’s horizontal and vertical movement in floor and staircase region in Ref. [13]. A mobility model in three-dimensional indoor environment is proposed based on proper boundary conditions on each floor and users horizontal and vertical motions in Ref. [14]. Several traffic forecasting algorithms based on various traffic models that employ the periodicity, recent traffic history, and flow-related information are proposed and evaluated in Ref. [15]. An extensive study of a GSM network utilization based on traffic analysis is presented in Ref. [16]. Network traffic prediction is performed through the process of maturation in Ref. [17]. To predict traffic patterns considering random movement of users methods are evaluated in Ref. [18]. A spectrumaware mobility management scheme is proposed for cognitive radio (CR) cellular networks in Ref. [19]. The effects of dynamic spectrum allocation (DSA) in the third world countries are evaluated in Ref. [20]. To avoid the bottleneck of the air interface for the future broadband mobile and achieve spectrum efficiency, deployment of advanced J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Mukherjee and De

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

3. PROPOSED SPECTRUM ALLOCATION METHOD BASED ON USERS MOBILITY INFORMATION In this section we have proposed the spectrum allocation method based on users mobility information. 3.1. Parameters Used The parameters used are shown in Table I. 3.2. Dynamic Location Area List The network contains a number of location areas a1  a2  a3       ak . Each LA contains a number of clusters. Firstly we have considered a macrocell based network where each cluster is covered by a macrocell. Thus a location area ai contains a number of macrocells like celli 1 , celli 2 , celli 3      , celli n . The probability of the user located at a1  a2  a3       ak are 1  2  3       k  respectively and ki=1 i ≤ 1 based on Ref. [31]. The probability of the user located at celli 1 , celli 2 , celli 3      , celli n in a particular LA ai are i 1  i 2  i 3       i n  respectively and nj=1 i j = i .31 A dynamic LA list is J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Table I. Parameter Callm Hm i i j TS k n Chmacro Chmicro Chpico Chfemto r  Ptm Ptmi Ptp Ptf Prminm Prminmi Prminp Prminf Rm Rmi Rp Rf L Gtm Gtmi Gtp Gr G hte hre

Parameters used. Definition Average number of call requests for MT (Mm  Average call holding time of MT (Mm  Probability of locating a MT at LA ai Probability of locating a MT at celli j of LA ai Time instant Number of LAs in the network Number of cells in the LA ai Number of channels in a macrocell Number of channels in a microcell Number of channels in a picocell Number of channels in a femtocell Reduction in the radius of congested inner cell in cell breathing Wavelength given by C/fc where C is the speed of light and fc is the frequency of the carrier wave Transmitted power of a macro BS Transmitted power of a micro BS Transmitted power of a pico BS Transmitted power of a femto BS Minimum Received power by a Mobile Terminal (MT) in macrocell Minimum Received power by a MT in microcell Minimum Received power by a MT in picocell Minimum Received power by a MT in femtocell Radius of a macrocell Radius of a microcell Radius of a picocell Radius of a femtocell System loss not related to path loss Macro BS antenna gain Micro BS antenna gain Pico Base station antenna gain Mobile station (MS) antenna gain Femto BS antenna gain BS antenna height Mobile station antenna height

maintained in the profile of each individual user. It is a list of location area IDs (LAIDs) with probabilities of visiting them in descending order.31 Each LAID points to a cell list containing the cell IDs with probabilities of visiting them in descending order.31 This list is updated when a location update takes place or a call arrives for the MT.31 The location update as well as dynamic LA list update method described in Ref. [31] is followed. In this section the traffic generated in each macrocell is determined based on the mobility information of individual user. After evaluating the dynamic LA list stored in the profile of individual user, the most probable macrocell to locate the user at each time instant is determined. To predict the number of users and traffic in each macrocell at each time instant, timing information is attached to the dynamic LA list31 to maintain short and long term mobility information of the user as shown in Figure 1. In this new dynamic LA list, contained in the profile of each user, the entry time (E.T.) and leaving time (L.T.) of the user in each macrocell is maintained from which the visiting time and duration of the user in each macrocell can be determined. Using this information, 3

RESEARCH ARTICLE

technologies such as multi-band, multi-carrier aggregation is proposed in Ref. [21]. A spectrum sharing scheme between two cellular networks is proposed in Ref. [22]. A water-filling theorem based joint power and spectrum allocation method is proposed in Ref. [23]. On the other hand, energy efficient green network development is a challenging issue in the field of mobile computing. Macro base stations have a typical power output in tens of watts. Due to large coverage area, macro base station has very high transmission power. Hence calculating the power for a microcell, picocell and femtocell and to what extent they can be used in a cellular network to develop an energy efficient network is a critical issue. Microcell systems increase system capacity significantly, and are low power systems and can be easily deployed as stated in Ref. [24]. A femtocell network is enhancing the indoor received signal quality and coverage and enabling variety of user-specific applications.25–28 Through the deployment of portable femtocells service can also be provided to the outdoor users. Using transmission power control and time hopping coupled with antenna sectoring cross-tier interference control methods are developed in Ref. [29] for a two-tier CDMA network.30 In this paper we have considered an artificially intelligent network where depending on the number of users, traffic load and blocking probability, cells of different sizes are deployed in such a way that congestion can be detected, prevented and avoided. Analytical models of power consumption and spectrum allocation are also proposed in this paper.

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

Mukherjee and De

location update method of Ref. [31] is: The entry and leaving time in each cell is maintained in the dynamic LA list and updated accordingly. The network is trained become artificially intelligent so that it can update the LA list and the corresponding cell lists automatically in the user profile when a location or cell update takes place.

RESEARCH ARTICLE

3.3. Traffic Load, Number of User, Blocking Probability Determination In this section the status of a cell in a location area is predicted from the view point of congestion. Let each location area contains a number of clusters and each cluster is covered by a macrocell. Suppose there are M mobile terminals (M1  M2      MM ) in the network and t time instants (T1  T2      Tt ). With the help of dynamic LA list a particular cell within a particular LA is identified using its unique identity in the network. The ID of a cell is denoted by Celli j where i denotes the LA ai and j denotes the cell number in ai where 1 ≤ i ≤ k and 1 ≤ j ≤ n. Let Ncapacity i j denotes the maximum capacity of Celli j in terms of the number of users. From the new dynamic LA list and cell lists, the most probable cell and the location area to track each user at a particular time, is identified. Then a matrix TR is generated based on this information. TR contains the most probable Cell ID visited by each MT (Mm ) at each time instant (TS ) as follows where 1 ≤ m ≤ M and 1 ≤ S ≤ t:

M1 M2   Fig. 1.

Dynamic LA list.

the most frequently visited macrocell by each user at each time instant can be obtained and the most probable macrocell to locate the user at that time is predicted. If all the profiles are checked, the most probable macrocell locating each user at each time instant can be predicted. From this information the number of users in each macrocell at each time instant can be assumed. Let there are three location areas with IDs a1  a2  a3 and probabilities of visiting a1  a2  a3 are 1  2  3 respectively where 1 > 2 > 3 . The number of cells in a1  a2 and a3 are 3, 4, 7 respectively as shown in Figure 1. In Figure 1, a1 points to cell list consists of the IDs of the cells (Cell1 1 , Cell1 2 , Cell1 3 ) contained in a1 with probabilities of visiting these cells 1 1  1 2  1 3 respectively where 1 1 > 1 2 > 1 3 . This LA list and corresponding cell lists are updated when a location update takes place.31 When a cell update occurs within a LA, the corresponding cell list is updated.31 The only difference from the 4

T R = Mm   MM−1 MM

T1

T2



TS

Cell1 1 ⎢ Cell2 1 ⎢ ⎢  ⎢  ⎢ ⎢ Celli 1 ⎢ ⎢  ⎢  ⎢ ⎣ Cellk−1 n Cellk 1

Cell1 2 Cell2 1  

   

Cell1 3 Cell2 3  

Celli 2  

  

Celli j  

Cellk−1 n Cellk 2

 

Cellk−1 1 Cellk n−2



 Tt−1    Cell1 n−1    Cell2 n        Celli n−1      

Cellk−1 1 Cellk n−1

Tt ⎤ Cell1 n Cell2 n−1 ⎥ ⎥ ⎥  ⎥  ⎥ (1) Celli n ⎥ ⎥ ⎥  ⎥  ⎥ Cellk−1 2 ⎦ Cellk n

Then a counter (Count) is initialized to 0. If a particular Celli j 1 ≤ i ≤ k, 1 ≤ j ≤ n) appears in the matrix TR for each time TS , set Count = Count + 1. The final value of the counter (Count) denotes the number of users visiting Celli j at TS and assign it in Numi j S i.e., set Numi j S = Count where 1 ≤ S ≤ t. J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Mukherjee and De

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

Based on Numi j S , the average number of users visiting Celli j is calculated as follows,

3.5. Dropping Probability of Handoff Call

(2)

As handoff calls are treated in the same way as new calls as in Ref. [33], the handoff failure probability at TS is given by, pf  S = Pnblocking S (9)

The traffic intensity for a MT, Mm , at TS is given by Ref. [32], (3) T Ii j m S = Callm S Hm S

The call dropping probability for handoff calls in Cellij at TS is given by Ref. [33],

where CallmS is the number of call requests and HmS is the call holding time by a MT, Mm , at TS . Then the total traffic load generated by Numi j S users in Celli j at TS is determined as follows:

Where H is the number of handoff calls. Thus average call dropping probability in Celli j is calculated as follows:

Numi j =

t

NumijS

t



S=1



Numi j S

T Li j S =

TS

S=1



APdropping =

Numi j S

T Ii j m S =

m=1

Callm S Hm S

Thus the total traffic load offered to the cell, Cell i j , in total amount of time is given by, T Li j =

T Li j S =

S=1

i j S t Num

S=1

T Ii j m S

TS

(6)

Using above equation average traffic load in a cell is predicted. 3.4. Blocking Probability of New Call The blocking probability of new call in Celli j at TS is given by Refs. [32, 33], Ch Chmacro T Lci j S T Li jmacro S (7) Pnblocking S = c! Chmacro ! c=0

APnblocking =

Pnblocking S

S=1

=

3.6. Probability of Call Completion, Satisfaction and Blocking Considering New and Handoff Calls

Pcom = 1 − APdropping 1 − APnblocking 

S=1

Chmacro !

t



c=0

Psat = Pcom = 1 − APdropping 1 − APnblocking 

TS

APblocking = 1−1−APdropping 1−APnblocking    t      Chmacro /Chmacro ! 1− 1− T LijS = 1− 1− s=1



t H  

Chmacro



T LcijS /c!

c=0 t 



TS



S=1



Ch

macro T LijS /Chmacro !



t    TS /c!

Chmacro

T LcijS 

c=0

(14)

S=1

(8)

S=1

Using the above equation average blocking probability of new call is predicted. J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

(13)

Thus ultimately average probability of call blocking considering new call blocking and handoff call dropping probability, is given by,

s=1

c!

(12)

If the total number of calls is Ntot and number of accepted call without dropping in later stage is Nacp , then the probability of satisfaction is given by Ref. [7], Psat = Nacp /Ntot . But Nacp = Pcom · Ntot . Hence we can conclude that,

·

TS

Ch

Chmacro t T Lci j S T Li jmacro S

S=1

(11)

Using the above equation average handoff call dropping probability is predicted.

Hence the average blocking probability of new call in Celli j is given by, t



TS

S=1

The call completion probability is given by Ref. [33],

S=1

t

t



In this way the blocking probability at each time instant, average blocking probability considering new and handoff calls in a cell, number of users at each time instant and average number of users visiting each cell can be predicted by the system. 5

RESEARCH ARTICLE

t



Pdropping S

(10)

(5)

m=1

The average traffic load in the cell, Celli j ,is obtained by dividing the total traffic load by the total time instants i.e., sum of all time instants as follows: AT Li j m = T Li j

t S=1

(4)

m=1

t

Pdropping S = 1 − 1 − pf  S H

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

3.7. Spectrum Allocation Based on Assumed Traffic Load The allocated spectrum bandwidth is determined by the system based on the number of users visiting each cell and each location area predicted based on proposed number of users and traffic load determination methods. Let, the maximum number of users supported on each radio channel in TDMA is maxus, thus the number of channel required in Cellij at time instant TS is given by, NumijS Nch = maxus

(15)

The number of TDMA channel slots is given by based on Refs. [32, 33], NchS = maxusBS −2Bguard /Bc  Where Bc is the carrier channel bandwidth, Bguard is the guard bandwidth and BS is the total allocated bandwidth at TS . Number of channels in TDMA is given by, Nch =

BS −2Bguard  NchS = maxus Bc

(16)

Comparing Eq. (15) and (16) the total allocated bandwidth is obtained as follows,

RESEARCH ARTICLE

NumijS BS −2Bguard  Bc NumijS = => BS = +2Bguard maxus Bc maxus Thus the average allocated spectrum considering all time instants is given by, t t    BS Baloc = S=1 Bc NumijS /maxus = t S=1 TS S=1 +2Bguard

t 

TS



(17)

S=1

With the help of Eq. (17) the average allocated spectrum in a cell can be determined. If TDMA is used, total number of slots required by the MT is the traffic intensity of the MT (Mm ) and given by, Nslotm = Callm Hm

(18)

Where Hm is the average call holding time and Callm is the average number of call requests for the MT (Mm ). Hence, the amount of channel bandwidth utilized by the MT (Mm ) is given by, Butm =

Nslotm ×Bc Callm Hm ×Bc = maxus maxus

(19)

where the maximum number of users supported on each radio channel in TDMA is maxus. Total utilized bandwidth at time instant TS is given by,

NumijS

BtotutiS =

m=1

6

Callm Hm ×Bc maxus m=1

NumijS

Butm =

(20)

Mukherjee and De

where NumijS is the number of users visiting the Cellij at time TS . Thus the average utilized bandwidth in the cell is determined as follows, NumijS Butm (21) tm=1 T S=1 S  t NumijS  Callm Hm ×Bc /maxus m=1 S=1 = t S=1 TS t

Butavg =

S=1

The average utilized spectrum can be determined by the network using Eq. (21) and compared to that of average allocated spectrum calculated using Eq. (17). Comparing these two equations it is observed that the average utilized spectrum is much less than that of allocated spectrum. The amount of bandwidth utilized by the users at each time instant (TS where 1 ≤ S ≤ t) is calculated using Eq. (20) and stored in a set as follows: But = Btotuti1  Btotuti2 Btotutit

This set contains the utilized bandwidth in the cell at each time instant. Take the maximum utilized bandwidth over all the time instants as follows: Butmax = max Btotuti1  Btotuti2 Btotutit

(22)

Using Eq. (22), the maximum utilized bandwidth in the cell (Cellij ) can be determined and allocated to the cell (Cellij ) i.e., the utilized bandwidth at busy time instant is determined and allocated to Cellij . This in turn reduces wastage of spectrum and thus spectrum efficiency can be achieved. So we can call this strategy as busy time spectrum allocation method. As the spectrum can be reused, it can be allocated to different location area without causing interference. Hence we can conclude that applying busy time strategy efficient spectrum allocation can be done in the network.

4. PROPOSED CONGESTION CONTROL APPROACH In this section we have proposed and discussed congestion detection, prevention and avoidance methods based on the call blocking and dropping probabilities in the following three subsections respectively. 4.1. Congestion Detection and Control in Overcrowded Cell The blocking probability at each time instant, average blocking probability (APblocking ) of a cell, number of users at each time instant and average number of users (Numij ) visiting each cell can be predicted by the network as discussed in the previous section. J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Mukherjee and De

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

If APblocking ≈ 1 and Numij > Ncapacityij in the macrocell, Cellij , the cell is called overcrowded cell and it denotes that most of the time this cell is suffered from congestion i.e., congestion is detected. To deal with congestion, replace the macrocell, Cellij , by picocells. The number of picocells can cover the area of the existing macrocell, Cellij , is calculated as follows:  √  √  Np = 3 3/2R2m 3 3/2R2p (23) √ 2 where √ 3 2 3/2Rm is the area of a macrocell and 3 3/2Rp is the area of a picocell. Let the number of channels allocated to picocell is Chpico . The traffic intensity for a MT, Mm , at TS is, T ImS = CallmS HmS where CallmS is the average number of call requests and HmS is the average call holding time by a MT (Mm  at TS .

4.1.2. Dropping Probability of Handoff Call

4.1.1. Blocking Probability of New Call

4.1.3. Probability of Blocking Considering New and Handoff Call

Let the picocell, Cellpj , contains NuspjS number of users at TS . Then the traffic load offered to Cellpj at TS is given by, T LpjS = NuspjS CallmS HmS . The new call blocking probability in Cellpj at TS is given by Refs. [32, 33], pico Chpico pi T LpjS T LpjS

Ch

PrnpjS =

Chpi =0

Chpi !

(24)

Np

PrnpjS

(25)

p=1

Thus the average new call blocking probability in this cluster considering total number of users (NumijS ) at TS  Np is given by, AvgPrnNp S = PrnNp S p=1 NuspjS . Thus the average blocking probability in this cluster for average number of users (Numij ) considering all time instants is given by, t

AvgPrnNp S

S=1



=

t 

TS

S=1

N Chpico Ch Chpi p pico t T LpjS T LpjS

Chpico ! Chpi =0 Chpi ! Np t

   NuspjS TS S=1

p=1

 =

t

 Np 

S=1





1− 1−

pico Chpico pi T LpjS T LpjS

NuspjS

Ch

Chpico !

p=1 Np 

Ch

  t

p=1

Chpi =0

H 

Chpi !

 TS

(27)

S=1

Using the above equation average handoff call dropping probability is predicted.

The call blocking probability considering the new and handoff calls in this cluster for average number of users (Numij ) considering all time instants is given by,

p=1

(26)

S=1

Using the above equation average blocking probability of new call is predicted. J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

APblockingcon = 1−1−AvgPrnNp 1−AvgPrHNp     N Chpico  Ch Chpi  p pico t T LpjS T LpjS = 1− 1− Chpico ! Chpi =0 Chpi ! S=1 p=1  Np

 NuspjS

t 

p=1



Chpico T LpjS

Chpico !





TS

 1−

S=1

t

 Np 

S=1

Chpi H   Chpico T LpjS Chpi =0

Chpi ! 



t 

TS

 1− 1−

p=1

 Np NuspjS

p=1



(28)

S=1

The satisfaction probability is given by, APsatcon = 1−AvgPrnNp 1−AvgPrHNp  = 1−APblockingcon

(29)

If picocells are not used then the average blocking probability in the macrocell, Cellij , for average number of users (Numij ) considering all time instants is calculated using Eq. (14). As a number of picocells are used replacing the macrocell, the total number of channels in the area is increased as compared to the macrocell and hence blocking probability is reduced. Moreover applying busy time spectrum allocation method discussed in Section 3, spectrum wastage in each cell can be minimized by the network. Thus probability of call blocking is minimized by increasing total number of channels without wastage of 7

RESEARCH ARTICLE

In proposed LA, the congested macrocell, Cellij , is replaced by picocells i.e., the congested cluster now contains Np number of picocells instead of one macrocell. The number of picocells, Np , required to cover this area is calculated using Eq. (23). Total number of users in the cluster containing Np picNp ocells is given by, NumijS = p=1 NuspjS . For Np cells the total new call blocking probability at TS is given by,

AvgPrnNp =

AvgPrHNp

Ch

Chpico !

PrnNp S =

The average call dropping probability for handoff calls in this cluster for average number of users (Numij ) considering all time instants is given by,

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

spectrum by the network. In Section 6, the performance of proposed picocell based congestion detection and control method is analyzed and compared to the macrocell based location area to show that proposed congestion control method minimizes the blocking probability such that congestion can be reduced. The network is trained to become artificially intelligent to detect the number of users and the blocking probability in the clusters of a location area itself, based on which the congestion detection and control is done.

Mukherjee and De

Nmi NusmijS . The average AvgPrnNmi S = PrnNmi S / mi=1 blocking probability in this cluster for average number of users (Numij ) considering all time instants is given by, AvgPrnNmi t t    = AvgPrnNmi S TS S=1

=



Nmi t S=1



mi=1

The blocking probability at each time instant, average blocking probability (APblocking ) in a cell, number of users at each time instant and average number of users (Numij ) visiting each cell can be predicted by the network as discussed Section 3. If 08 ≤ APblocking < 1 and Numij ≈ Ncapacityij in the macrocell, Cellij , the cell is called fully loaded cell and it denotes that this cell is congestion prone and thus congestion prevention is needed. As congestion is likely to be occurred, then to deal with it, replace the macrocell, Cellij , by microcells. The number of microcells can cover the area of the existing macrocell, Cellij , is calculated as follows:   √   √ Nmi = 3 3/2R2m 3 3/2R2mi (30) √ 2 where √ 3 2 3/2Rm is the area of a macrocell and 3 3/2Rmi is the area of a microcell.

 Chmicro Chmi T LmijS

Chmicro !  Nmi

4.2. Congestion Prevention in Fully Loaded Cell

RESEARCH ARTICLE

S=1

Chmicro T LmijS

Chmi !  t 

Chmi =0

NusmijS

mi=1

 (33)

TS

S=1

Using the above equation average blocking probability of new call is predicted. 4.2.2. Dropping Probability of Handoff Call The average call dropping probability in this cluster for average number of users (Numij ) considering all time instants is given by,  t  Nmi    Chmicro Chmicro T LmijS 1− 1− AvgPrHNmi = Chmicro ! Chmi =0 S=1 mi=1 Ch

mi T LmijS

H  Nmi

Chmi !

mi=1



NusmijS

t 

TS



S=1

(34) Using the above equation average handoff call dropping probability is predicted.

4.2.1. Blocking Probability of New Call

4.2.3. Probability of Blocking Considering New and Handoff Call

Let the microcell, Cellmij , contains NusmijS number of users at TS . Then the traffic load offered to Cellmij at TS is given by, T LmijS = NusmijS CallmS HmS . The call blocking probability in Cellmij at TS is given by Refs. [32, 33],

The call blocking probability considering the new and handoff calls in this cluster for average number of users (Numij ) considering all time instants is given by,

Ch

PrnmijS =

micro T LmijS

 Chmicro Chmi T LmijS

Chmicro !

Chmi =0

Chmi !

APblockingF L (31)

In proposed LA the fully loaded macrocell, Cellij , is replaced by microcells i.e., the fully loaded cluster now contains Nmi number of microcells instead of one macrocell. The number of microcells, Nmi , required to cover this area is calculated using Eq. (30). Total number of users in thecluster containing Nmi microcells is given by, Nmi NumijS = mi=1 NusmijS . For Nmi cells the total blocking probability at TS will be given by, PrnNmi S =

Nmi

PrnmijS

(32)

mi=1

The average blocking probability in this cluster for total number of users (NumijS ) at TS is given by, 8

= 1−1−AvgPrnNmi 1−AvgPrHNmi     Chmicro Chmicro Nmi  t T LmijS = 1− 1− Ch ! micro S=1 mi=1 Chmi =0    t Chmi  Nmi  T LmijS  NusmijS TS Chmi ! mi=1 S=1      Chmicro

Nmi  t T LmijS 1− 1− 1− Chmicro ! mi=1 S=1  Chmi H  Nmi Chmicro T LmijS  NusmijS mi=1 Chmi =0 Chmi !  t

  TS

(35)

S=1

J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Mukherjee and De

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

4.3.1. Blocking Probability of New Call

The satisfaction probability is given by, APsatF L = 1−AvgPrnNmi 1−AvgPrHNmi  = 1−APblockingF L

(36)

If microcells are not used then the average blocking probability in the macrocell, Cellij , for average number of users (Numij ) considering all time instants is calculated using Eq. (14). As a number of microcells are used replacing the macrocell, the total number of channels in the area is increased as compared to the macrocell and hence blocking probability is reduced. Moreover applying busy time spectrum allocation method discussed in Section 3, spectrum wastage in each cell can be minimized by the network. Thus probability of call blocking is minimized by increasing total number of channels without wastage of spectrum. In Section 6, the performance of proposed congestion prevention method is analyzed and compared to the macrocell based network to show that proposed microcell based congestion prevention method minimizes the blocking probability such that congestion can be prevented. The congestion prevention is done by detecting blocking probability and number of users by the trained artificially intelligent network. 4.3. Congestion Avoidance in Temporary Congested Cell

Then calculate the required number of femtocells to be deployed in Cellij during congestion as follows:   Nexcessij Nf ij = (38) Nfemuser where Nfemuser denotes maximum user supported by a femtocell. Deploy and activate Nf ij femtocells within Cellij until NumijS < Ncapacityij and PblockingS < 08. When NumijS < Ncapacityij and PblockingS < 08, the deployed femtocells are deactivated i.e., turned off. In this way by deploying femtocells within the macrocell congestion can be avoided. J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Ch

PrfemjS =

femto T LfemjS

Chfem  Chfemto T LfemjS

Chfemto !

Chfem =0

(39)

Chfem !

In proposed LA the temporary congested macrocell Cellij contains femtocells depending on the excess number of users. Let Nf ij number of femtocells are allocated to Cellij . For Nf ij cells the total new call blocking probability at TS is given by,

Nf ij

PrNf ij S =

PrfemjS

(40)

fem=1

Now, the macrocell, Cellij , contains the number of users equals to its capacity. Let NusmjS is the number of users in Cellij at TS . Then the traffic load offered to Cellij at Ts is given by, T LmjS = NusmjS CallmS HmS . The new call blocking probability in the macrocell Cellij at TS is given by Refs. [32, 33], Chmacro Chmacro Chm T LmjS T LmjS (41) PrmjS = Chmacro ! Chm =0 Chm ! Thus the total new call blocking probability at TS in this temporary congested cluster is given by, Prtcijs = PrmjS +PrNf ij S

(42)

The average new call blocking probability for total number of users (NumijS ) at TS in this temporary congested cluster is given by,   AvgPrtcijs = PrmjS +PrNf ij S 





Nf ij

NusmjS +

NusfemjS

(43)

fem=1

Nf ij Where NumijS = NusmjS + fem=1 NusfemjS . When the cell is congested only then the femtocells are activated. Thus at each time instant the femtocells may not be activated and depending on the number of excess users Nf ij also varies. When the macrocell is not congested Nf ij = 0 and NusfemjS = 0. The average new call blocking probability in this cluster for average number of users (Numij ) considering all time instants is given by, AvgPrtcij =

t S=1

AvgPrtcijS

t



TS

S=1

9

RESEARCH ARTICLE

The blocking probability at each time instant, average blocking probability (APblocking ) of a cell, number of users at each time instant and average number of users (Numij ) visiting each cell can be predicted by the network as presented Section 3. If, APblocking < 08 and Numij < Ncapacityij , in the macrocell, Cellij , but the macrocell is temporarily suffered from congestion at a particular time instant, TS , then the macrocell, Cellij , falls into the category of temporary congested cell. To avoid congestion during peak period, femtocells are deployed within that congested cell and activated until the congestion is removed. Calculate the number of excess users during congestion in Cellij as follows: Nexcessij = NumijS −Ncapacityij (37)

Let the femtocell, Cellfemj , contains NusfemjS number of users at TS . Then the traffic load offered to Cellfemj at TS is given by, T LfemjS = NusfemjS CallmS HmS . The call blocking probability in Cellfemj at TS is given by Refs. [32, 33],

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

=



S=1



Nf ij

+



Ch

 



Nf ij

NusmjS +

Nusfemjs



t

S  



TS

(44) +

 Nf ij 



Ch

1− 1−

RESEARCH ARTICLE

Ch

macro T LmjS /Chmacro ! Chmacro Chm Chm =0 T LmjS /Chm !

H 

H 

Ch

femto T LfemjS /Chfemto !

Chfemto

Chfem Chfem =0 T LfemjS /Chfem !

Nf ij fem=1

Nusfemjs

t  

 TS

(45)

S=1

Using the above equation average handoff call dropping probability is predicted. 4.3.3. Probability of Blocking Considering New and Handoff Call The call blocking probability considering the new and handoff calls in this cluster for average number of users (Numij ) considering all time instants is given by, APblockingtcon = 1−1−AvgPrtcij 1−AvgPrH tcij     Chmacro t T LmjS /Chmacro ! = 1− 1− Chmacro Chm S=1 Chm =0 T LmjS /Chm ! Chfemto  Nf ij  T LfemjS /Chfemto ! + Chfemto Chfem fem=1 Chfem=0 T LfemjS /Chfem ! 10







Nf ij

NusmjS +

Nusfemjs

fem=1

= 1−APblockingtcon

1− 1−

NusmjS +



t 

 

TS

(46)

S=1

APsattcon = 1−AvgPrtcij 1−AvgPrHtcij 

fem=1



H 

Ch

femto T LfemjS /Chfemto !

The satisfaction probability is given by,

t=1



H 

Chfem Chfem =0 T LfemjS /Chfem !

AvgPrH tcij  S   1− 1− =



1− 1−

Chfemto

The average call dropping probability in this cluster for average number of users (Numij ) considering all time instants is given by,

+



fem=1

4.3.2. Dropping Probability of Handoff Call



TS

macro T LmjS /Chmacro ! Chmacro Chm Chm =0 T LmjS /Chm !



 Nf ij 





S=1

fem=1



Using the above equation average blocking probability of new call is predicted.



Nusfemjs

t

t=1

S=1

fem=1

NusmjS +

· 1−

Chfem Chfem=0 T LfemjS /Chfem !

 



Nf ij



Chfemto

fem=1





macro T LmjS /Chmacro !  Chmacro Chm Chm =0 T LmjS /Chm ! Chfemto  T LfemjS /Chfemto !

t  

Mukherjee and De

(47)

If femtocells are not used then the average blocking probability in the macrocell, Cellij , for average number of users (Numij ) considering all time instants is calculated using Eq. (14). As femtocells are deployed within the temporary congested macrocell according to the number of excess users, the total number of channels in the area is increased as compared to the macrocell and hence blocking probability is reduced. Moreover applying busy time spectrum allocation method discussed in Section 3, spectrum wastage in each cell can be minimized by the network. Thus probability of call blocking is minimized by increasing total number of channels without wastage of spectrum. In Section 6, the performance of proposed congestion avoidance method is analyzed and compared to the macrocell based location area to show that proposed congestion avoidance method minimizes the blocking probability such that congestion can be avoided. The network is trained artificially so that the congestion avoidance is done by detecting blocking probability and number of users by the intelligent network itself. 4.4. Under Loaded Cell The blocking probability at each time instant, average blocking probability (APblocking ) of a cell, number of users at each time instant and average number of users (Numij ) visiting each cell can be predicted from Section 3. If, APblocking < 08 and Numij < Ncapacityij , over all the time instant TS where 1 ≤ S ≤ t in the macrocell, Cellij , then the cell falls into the category of under loaded cell i.e., it never suffers from congestion. As it never suffers from congestion no modification is done i.e., the macrocell is not replaced. J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Mukherjee and De

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

4.5. Explanation of Proposed Method

For example, location area a1 contains seven cells: cell1 , cell2 , cell3 , cell4 , cell5 , cell6 , cell7 as shown in Figure 2. Among these cells, let cell3 and cell4 are overcrowded macrocells, cell2 and cell6 are fully loaded macrocells, cell1 and cell7 suffer from temporary congestion and cell5 is under loaded based on the predicted number of users, traffic load and blocking probability as presented in Figure 3. Hence according to the proposed congestion detection, prevention and avoidance methods method replacing cell3 and cell4 picocells are allocated and replacing cell2 and cell6 microcells are deployed as shown in Figure 4. Although cell1 and cell7 are under loaded according to the average number of users and blocking probability, each of them suffers from congestion over certain J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Fig. 2. A 7-cluster location area where each cluster contains a macrocell.

amount of time. Thus according to need femtocells are deployed within them temporarily as shown in Figure 4. No modification is done to cell5 as it is always under loaded. If the proposed congestion detection, prevention and avoidance methods are applied to each location area of the network, then a microcell, macrocell, picocell and femtocells based network can be obtained where the cell size will be decided based on user density, traffic load and blocking probability as shown in Figure 4. As the network is trained to detect blocking probability and number of users to achieve congestion detection, prevention and avoidance, hence the network is artificially intelligent.

5. PROPOSED POWER CONSUMPTION MODEL In this section we have proposed the power consumption model of a macrocell-microcell-picocell-femtocell based heterogeneous network where the cell sizes are decided based on the call blocking and dropping probabilities to deal with congestion as discussed in the previous section. The power consumption in macrocell-microcellpicocell-femtocell based network is compared to the power

Fig. 3. A 7-cluster location area where each cluster contains a macrocell where cell3 and cell4 are overcrowded cells, cell2 and cell6 are fully loaded cells, cell1 and cell7 are temporary congested cells and cell5 is under loaded cell.

11

RESEARCH ARTICLE

The congestion detection, prevention and avoidance methods can be summarized as follows: i) Based on the dynamic LA list generate matrix TR as presented in Section 3. ii) Based on the matrix TR the probable number of users visiting each macrocell at each time instant is determined. iii) The average numbers of calls generated or received and average call holding time by each user from user profile are determined. iv) Using this information the traffic intensity per user and blocking probability considering probabilities of blocking new call and dropping handoff call both due to congestion at each time instant are calculated. v) The average numbers of users visiting each macrocell, the average traffic load offered to the macrocell and the average blocking probability in the macrocell over an amount of time are determined. vi) If the average number of users exceeds the capacity of the macrocell and blocking probability is approximately equals to 1, then congestion is detected and hence the macrocell is called overcrowded. To remove congestion the overcrowded macrocell is replaced by a number of picocells. vii) Else if, the average number of users is approximately equals to the capacity of the macrocell and blocking probability ranges from 0.8 to 1, the macrocell is called fully loaded. Then the fully loaded macrocell is replaced by a number of microcells to prevent congestion. viii) Else if the average number of users in a macrocell is less than its capacity and blocking probability is less than 0.8 but suffers from congestion at a certain amount of time, then to avoid congestion a number of femtocells are deployed and activated within this macrocell. When the macrocell becomes under loaded, the femtocells contained in it are deactivated. xi) If a macrocell has blocking probability less than 0.8 and average number of users is less than its capacity as well as never suffers from congestion, then it is under loaded cell. Thus it is not replaced i.e., no modification is done.

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

Fig. 4. Microcell, picocell, femtocell and macrocell based network where the cell size will be decided based on the proposed congestion detection, prevention and avoidance methods.

consumption involved in cell breathing which is a popular congestion control method.

Mukherjee and De

The maximum transmitted power by a femto BS is given by Ref. [34], W 4 R2f (54) Ptf = DG Where W is the minimum received power density (Watt/m2 ), D is the normalized radiation pattern in the direction    i.e., D is equal to unity in the direction of maximum radiation, G is the femto BS√antenna gain. Area of a femtocell having radius Rf is 3 3/2R2f . The minimum received power by the MT in a femtocell is given by, √ Pr minf = W√·3 3/2R2f which implies, W= 2 Pr minf /3  √ 3/2Rf . Replacing power density W by Pr minf / 3 3/2R2f we obtain,34 Ptf =

5.1. Power Consumption in Proposed Macrocell-Microcell-Picocell-Femtocell Based Network The received power by a mobile terminal at a distance d from the base station of a cell is given by Refs. [32, 34],

RESEARCH ARTICLE

P G G 2 Pr d = t t 2 r 2 4  d L

(48)

Where Gt is the base station antenna gain. When a MT is situated at the border region of the cell, the received power by the MT is the minimum received power by that MT in that cell. For the MT situated at the border region of a cell of radius R, the received power is given by Ref. [34], Pr =

P t Gt Gr 2 4 2 R2 L

(49)

Hence the maximum transmitted power by a BS in a cell of radius R is given by, Pt =

Pr 4 2 R2 L G t G r 2

(50)

Let Pr minm is the minimum received power of a MT to accept or make a call in macrocell. Thus the maximum transmitted power by a macro BS is given by, 4 2 R2m L P Ptm = r minm (51) Gtm Gr 2 Similarly, the maximum transmitted power by a micro BS is given by, 4 2 R2mi L P Ptmi = r minmi (52) Gtmi Gr 2 Subsequently, the maximum transmitted power by a pico BS is given by, Ptp = 12

Pr minp 4 2 R2p L Gtp Gr 2

(53)

Pr minf 4 √ 3 3/2DG

(55)

Let the total number of macrocells in the location area before any cell replacement is Nm . The power consumption by these macrocells is given by, Ptotm =

Pr minm 4 2 R2m L ×Nm Gtm Gr 2

(56)

Let the number of existing macrocell replaced by the picocells is Nmrp and the number of existing macrocell replaced by the microcells is Nmrmi . Thus the remaining number of macrocell is Nm −Nmrp +Nmrmi . The power consumption by these remaining macrocells is given by, Ptotmnew =

Pr minm 4 2 R2m L ×Nm −Nmrp +Nmrmi  (57) Gtm Gr 2

The total number of picocells which replace overcrowded macrocells is given by the product of the number of picocells used to replace a macrocell calculated using Eq. (23) and the number (Nmrp) √ replacedmacrocell √  of  as follows: Ntp = Nmrp × 3 3/2R2m  3 3/2R2p  . The power consumption by these picocells is given by, √ Pr minp 4 2 R2p L 3 3/2R2m Ptotp = ×Nmrp × √ (58) Gtp Gr 2 3 3/2R2p The total number of microcells which replace the fully loaded macrocells is given by the product of the number of microcells used to replace a macrocell calculated using Eq. (30) and the number of replaced macrocells √  √ (Nmrmi) as follows: Ntmi = Nmrmi × 3 3/2R2m  3 3/2R2mi  . Thus the power consumption by these microcells is given by, √ Pr minmi 4 2 R2mi L 3 3/2R2m Ptotmi = ×Nmrmi × √ (59) Gtmi Gr 2 3 3/2R2mi Let the number of femtocells deployed in a temporary congested macrocell (Cellij ), is Nfconij and the number J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Mukherjee and De

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

temporary congested macrocell is Nmtcon . As the femtocells are activated when congestion occurs, we have considered the duty cycle (TDqij ) of a femtocell (F Cellq ) as follows,34  TDqij = Ponqij  Ponqij +Poff qij  where Ponqij and Poff qij represent the clock cycles during which the femtocell is powered on or off respectively. The power transmitted by a femtocell (F Cellq ) is given by, Ptfd = TDqij ×Ptf = TDqij ×

Pr minf 4 √ 3 3/2DG

(60)

The total power consumption by Nfconij femtocells in the macrocell (Cellij ) is obtained by calculating the sum of the power consumption by all the femtocells contained in Cellij as follows: Ptotf ij =

Nfconij 



q=1

Pr minf 4  TDqij × √ 3 3/2DG

(61)

Thus the total power consumption by all the femtocells contained in all of the temporary congested macrocells is given by the sum of the power consumption by all the femtocells contained in each congested macrocell as follows: Ptotf =



Nmtcon

Ptotf ij =

j=1

TDqij ×



j=1

q=1

Pr minf 4  √ 3 3/2DG

(62)

Ptotnew = Ptotp +Ptotmi +Ptotf +Ptotmnew

√ 3 3/2R2m = ×Nmrp × √ Gtp Gr 2 3 3/2R2p √ Pr minmi 4 2 R2mi L 3 3/2R2m + ×Nmrmi × √ Gtmi Gr 2 3 3/2R2mi Pr minp 4 2 R2p L

Nmtcon Nfconij 





j=1

q=1

TDqij ×

Pr minm 4 2 R2m L ×Nmrp +Nmrmi  Gtm Gr 2 Pr minp 4 2 R2m L − ×Nmrp Gtp Gr 2 =

+

Nmtcon Nfconij Pr minmi 4 2 R2m L ×N + mrmi 2 Gtmi Gr  q=1 j=1   Pr minf 4 TDqij × √ (64) 3 3/2DG

As, Pr minm > Pr minmi > Pr minp > Pr minf , and, TDqij ≤ 1, Pr minm 4 2 R2m L ×N mrp +N mrmi  Gtm Gr 2 Pr minp 4 2 R2m L Pr minmi 4 2 R2m L > ×N + mrp Gtp Gr 2 Gtmi Gr 2 Nmtcon Nfconij  Pr minf 4  ×Nmrmi + TDqij × √ 3 3/2DG j=1 q=1 which in turn implies that,

Thus the total power consumption by all base stations in macrocell-microcell-picocell-femtocell based location area is given by,

+

Ptotm −Ptotnew = Ptotm −Ptotp +Ptotmi +Ptotf +Ptotmnew 

Pr minf 4  √ 3 3/2DG

Ptotm > Ptotnew

(65)

Thus employing microcells, picocells, macrocell and femtocells based on traffic load of each cluster in each location area, low power consumption by the base stations can be achieved. Hence it can be concluded that using microcell, picocell, macrocell and femtocell based network, congestion can be controlled at low power as well as energy consumption and thus an energy efficient green mobile network can be achieved. In Section 6 the power consumption by the base stations in a macrocell based location area is compared to that of a microcellpicocell-macrocell-femtocell based location area to show that the proposed microcell-picocell-macrocell-femtocell based network achieves low power consumption.

+

Pr minm 4 2 R2m L ×Nm −Nmrp +Nmrmi  Gtm Gr 2

5.2. Comparison with Existing Congestion Control Approach Cell Breathing

=

Pr minp 4 2 R2m L Pr minmi 4 2 R2m L ×N + mrp Gtp Gr 2 Gtmi Gr 2

Cell breathing is a very well known congestion control approach in mobile network. Let again consider a 7-cluster location area where each cluster is covered by a macrocell and for congestion control, cell breathing will be applied in this location area. In cell breathing, the congested cell reduces its coverage area and the six under loaded adjacent cells increase their coverage area to capture the customers situated at the border of the overloaded cell.6 The difficulty

× Nmrmi + +

Nmtcon Nfconij 





j=1

q=1

TDqij ×

Pr minf 4  √ 3 3/2DG

2

Pr minm 4  R2m L ×Nm −Nmrp +Nmrmi  (63) Gtm Gr 2

J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

13

RESEARCH ARTICLE





Nmtcon Nfconij

Subtracting Eq. (63) from Eq. (56) and simplifying it we obtain,

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

of this scheme is if most of the adjacent cells are fully loaded or congested, then cell breathing cannot work. In cell breathing the new transmitted power of the congested inner macrocell after reduction in its coverage area is given by Ref. [6], Ptinner =

Pr minm 4 2 Rm −r2 L Gtm Gr 2

(66)

where r is the reduction in radius of the congested inner macrocell and Rm −r is the new radius of the congested inner macrocell after reducing its coverage area. Similarly the transmitted power of each adjacent macrocell base station after increase in the coverage area due to cell breathing is given by Ref. [6], Ptadj =

Pr minm 4 2 Rm +r2 L Gtm Gr 2

(67)

Hence the total transmitted power by the six adjacent macrocell base stations is given by Ref. [6], Ptotadj = 6×

Pr minm 4 2 Rm +r2 L Gtm Gr 2

(68)

Hence the total transmitted power in that location area is given by Ref. [6],

RESEARCH ARTICLE

PtotCB = Ptinner +Ptotadj = + 6×

Pr minm 4 2 Rm −r2 L Gtm Gr 2

5.3. Path Loss Cost-231 Hata model is used for path loss calculation.32 The path loss for a mobile terminal residing at a distance d from the base station in a cell is given by, PLm = 6955+2616logfc −1382loghte −ahre +449−655loghte logd

(70)

Where ahre  is the correction factor for effective mobile antenna height which is a function of the size of the coverage area and ahre  = 11logfc −07hre −156logfc − 08 The maximum path loss for that mobile station will occur when it is situated at the border of the cell. Thus the maximum path loss for a mobile station in a macrocell is given by, PLmaxm = 6955+2616logfc −1382loghte −ahre +449−655loghte logRm

PLmaxmi = 6955+2616logfc −1382loghte

2

(69)

Comparing Eqs. (69) and (56), it is observed that the total transmitted power by the base stations contained in a macrocell based location area after cell breathing is higher than that of before cell breathing i.e., PtotCB > Ptotm . This is because in cell breathing the transmitted power by the inner base station is although reduced because of reduction in its coverage area but causes increase in the transmitted power by the six adjacent base stations because of increase in their coverage area and thus increase total power consumption as transmitted power by a base station is proportional to its coverage area. As from Eq. (65) it is observed that Ptotm > Ptotnew then obviously PtotCB > Ptotnew which implies that our proposed macrocellmicrocell-picocell-femtocell based network for congestion control is more energy efficient than cell breathing. Thus we can conclude that by employing macrocell-microcellpicocell-femtocell based network for congestion control, power as well as energy consumption can be minimized. Hence the proposed congestion control approach is better than cell breathing from the view point of power consumption. In section six the power consumption by the base stations in a macrocell based location area after 14

cell breathing is compared to that of a microcell-picocellmacrocell-femtocell based location area to show that the proposed microcell-picocell-macrocell-femtocell network based congestion control achieves low power consumption than cell breathing.

(71)

Similarly, the maximum path loss for a mobile station in a microcell is given by,

Pr minm 4 2 Rm +r2 L Gtm Gr 2

P 4  L = r minm 7R2m +10Rm r +2r 2  Gtm Gr 2

Mukherjee and De

−ahre +449−655loghte logRmi

(72)

Subsequently, the maximum path loss for a mobile station in a picocell is given by, PLmaxp = 6955+2616logfc −1382loghte −ahre +449−655loghte logRp

(73)

As the radius of femtocell is very small, path loss in femtocell is negligible. As, Rm > Rmi > Rp , thus PLmaxm > PLmaxmi > PLmaxp which in turn indicates that the path loss can be minimized using microcell, picocell, femtocell and macrocell based network according to need instead of using only macrocell based network.

6. RESULTS AND DISCUSSIONS In this section we have first calculated the number of users visiting each macrocell in a 7-cluster LA considering the profiles of approximately 10000 users. Then the traffic load offered to each cluster covered by a macrocell is determined. According to congestion detection, prevention and avoidance methods, the average number of users visiting each cluster and average blocking probability considering both new and handoff call in each cluster J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Mukherjee and De

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

Table II. Parameter values. Parameter

Value

Prm Prmi Prp Prf Rm Rmi Rp Rf Gtm Gtmi Gtp Gr G r Nfemuser Ncapacityij

15–20 mW 10–15 mW 5–10 mW 1–5 mW 1–6 km 0.2–1 km 0.05–0.2 km 0.01–0.02 km 16–18 dBi 7 dBi 5 dBi 1 dBi 2 dBi 0.25–0.4 km 12 992 as 124 channels are allocated to each cell in GSM and number of slots per channel is considered 8

J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

observed that cell5 is under loaded in most of the times and never suffers from congestion. Figure 6 presents the allocated bandwidth calculated using Eq. (17) in each macrocell of a 7-cluster location area without considering the user mobility based traffic load and the allocated bandwidth calculated using Eq. (22) i.e., maximum utilized bandwidth in each macrocell of a 7-cluster location area based on busy time spectrum allocation method considering the user mobility based traffic load. Figure 6 also presents the average utilized bandwidth calculated using Eq. (21) in each macrocell of a 7-cluster location area. From Figure 6 it is observed that allocating the spectrum using busy time spectrum allocation method considering user mobility based traffic load, reduction in the wastage of bandwidth can be achieved than spectrum allocation without considering the user mobility based traffic load. Generally, if user mobility based traffic load is not considered, the amount of bandwidth allocated to each macrocell is approximately equal as shown in Figure 6. Figure 6 presents that spectrum allocation without considering the user mobility based traffic load, causes approximately 94% wastage of bandwidth whereas spectrum allocation using busy time spectrum allocation method considering user mobility based traffic load causes approximately 58% wastage of bandwidth. Hence it is observed that approximately 36% reduction in bandwidth wastage can be achieved by allocating spectrum

Fig. 6. Macrocell ID versus allocated and utilized bandwidth in each cell in a 7-cluster macrocell based LA.

15

RESEARCH ARTICLE

are determined. If the average number of users exceeds its capacity and average blocking probability is approximately equals to 1, then congestion is detected and to control it the cluster is covered by picocells replacing the macrocell. Else if, the average number of users approximately equals to its capacity and average blocking probability ranges from 0.8 to 1, then to prevent congestion the cluster is covered by microcells replacing the macrocell. Then track the cluster having average blocking probability less than 0.8 and average number of users is less than its capacity but suffers from congestion at a certain amount of time. To avoid congestion deploy and activate the femtocells along with the existing macrocell within such cluster during congestion according to the need. When this temporary congested cluster becomes under loaded, the femtocells contained in it are deactivated. If a cluster has average blocking probability less than 0.8 and average number of users is less than its capacity as well as never suffers from congestion, no replacement is done i.e., it is remained covered by macrocell. We have compared the power consumption and blocking probability in this macrocell, microcell, picocell and femtocell based LA to that of a macrocell based LA. The values of the parameters assumed323435 are presented in Table II. Figure 5 presents the number of users visiting each cell in a 7-cluster LA predicted based on the matrix TR determined using Eq. (1). Each cluster of this LA is covered by a macrocell. Threshold in this case is the maximum capacity (Ncapacityij ) of each cell i.e., 992. From Figure 5 it is presented that cell3 and cell4 suffer from congestion most of the time period. From Figure 5 it is also presented that cell2 and cell6 also suffer from congestion over certain amount of time and most of the times reach their capacity. Figure 5 also demonstrates that cell1 and cell7 becomes congested for certain amount of time although most of the time they remain under loaded. From Figure 5 it is

Fig. 5. Time (1 hour interval) versus number of users in each cell in 7-cluster macrocell based LA.

RESEARCH ARTICLE

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

using busy time spectrum allocation method considering user mobility based traffic load and hence spectrum efficiency can be achieved. Figure 7 presents the average number of users visiting each macrocell in a 7-cluster LA calculated using Eq. (2). Threshold in this case is the maximum capacity (Ncapacityij ) of each cell i.e., 992. From Figure 7 it is presented that cell3 and cell4 suffer from congestion because their average load exceeds their maximum capacity. From Figure 7 it is presented that cell2 and cell6 are fully loaded because their average load approximately equals to their capacity. Figure 7 also presents that other three cells are under loaded as their average load is less than their capacity. Figure 8 presents the average blocking probability considering both new and handoff call in each macrocell in a 7-cluster LA determined using Eq. (14). Threshold in this case is considered 1. From Figure 8 it is presented that cell3 and cell4 have average blocking probability approximately equals to 1. From Figure 8 it is also presented that cell2 and cell6 have average blocking probability ranges from 0.8 and 1. Figure 8 also demonstrates that other three cells have average blocking probability less than 0.8. From Figures 7 and 8 it is observed that cell3 and cell4 , both of them have average blocking probability approximately equals to 1 and average number users more than their capacity. Thus congestion is detected in these two cells and hence these two cells are overcrowded. To deal with congestion, these two cells are replaced by picocells. Subsequently, from Figures 7 and 8 it is observed that cell2 and cell6 , both of them have average blocking probability ranges from 0.8 to 1 and average number users approximately equals to their capacity. Hence cell2 and cell6 are fully loaded cells and thus congestion prone. To prevent congestion, these two cells are replaced by microcells. From Figures 5, 7 and 8 it is observed that cell1 and cell7 becomes congested for certain amount of time although their average blocking probability is less than 0.8 and average number users less than their capacity. Hence cell1 and cell7 are temporary congested cells and to avoid congestion

Fig. 7. Macrocell ID versus average number of users in each cell in macrocell based LA.

16

Mukherjee and De

Fig. 8. Cluster ID versus average blocking probability in each cell in macrocell based LA.

in these two cells femtocells are deployed and activated according to requirement. From Figures 5, 7 and 8 it is observed that cell5 has average blocking probability less than 0.8 and average number users less than its capacity and cell5 never suffers from congestion. This implies cell5 is under loaded cell and thus no modification is needed. The power consumption in the proposed macrocell, microcell, picocell and femtocell based LA is determined using Eq. (63) and compared to the power consumption of a macrocell based 7-cluster LA calculated using Eq. (56). The power consumption in the proposed macrocell, microcell, picocell and femtocell based LA is also compared to the power consumption of a macrocell based 7-cluster LA after applying cell breathing calculated using Eq. (69) to deal with congestion. The power consumption in this proposed macrocell, microcell, picocell and femtocell based LA and in a macrocell based LA are presented in Figure 9. The area of the 7-cluster location area is obtained by multiplying the area covered by√a macrocell by 7. The area covered by a macrocell is 3 3/2R2m where Rm is the radius of the macrocell. As transmitted power by the BSs is directly proportional to the coverage area, in Figure 9 total transmission power by the BSs are increased with area. Figure 9 presents that using macrocellmicrocell-picocell-femtocell based LA 35.29–40% of total power consumption by the base stations can be minimized as compared to the only macrocell based LA. Figure 9 also presents that using proposed LA 50.91–57.74% minimization in total power consumption by the base stations can be achieved as compared to the power consumption involved in cell breathing applied to the macrocell based LA. From Figure 9 it is observed that the total transmission power by the base stations contained in a location area after cell breathing is approximately 24.57–29.56% higher than that of before cell breathing. This is because in cell breathing the transmission power of the inner base station is although reduced but causes increase in the transmission power of the six adjacent base stations, thus increase total power consumption. As it is observed that using proposed macrocell-microcell-picocellfemtocell based location area total power consumption by all base stations can be minimized, thus we can conclude that by J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Mukherjee and De

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

7-cluster LA calculated using Eq. (13) as shown in Figure 11. Figure 11 demonstrates that using macrocellmicrocell-picocell-femtocell based network as in the proposed congestion control method, satisfaction probability can be increased approximately 37.8% than previous channel borrowing method7 in a macrocell based network with respect to the average traffic load calculated using Eq. (6). Thus it is observed that using proposed congestion control scheme satisfaction probability can be increased than channel borrowing scheme in a congested mobile network. As the blocking probability increases with traffic load, thus satisfaction probability will decrease with traffic load as shown in Figure 11. Thus it is observed that in other congestion control approaches such as cell breathing and channel borrowing scheme either the power consumption is increased or the adjacent cells are required to be under loaded. Moreover maintaining buffer to hold the call requests in a congested cell introduces delay and sometimes call dropping. Hence

Fig. 10. Cluster ID versus average blocking probability in each cluster in macrocell based LA and macrocell, microcell, picocell and femtocell based LA.

Fig. 9. Area covered by all the clusters in a LA versus total transmitted power by all of the base stations contained in the LA in cell breathing, in macrocell based LA and in proposed macrocell, microcell, picocell and femtocell based LA.

J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Fig. 11. Traffic load offered to the network versus satisfaction probability.

17

RESEARCH ARTICLE

employing macro-micro-pico-femtocell based network for congestion control, power as well as energy consumption can be minimized. The average blocking probability considering both new and handoff call in each cluster of the proposed picocell, microcell, femtocell and macrocell based LA are determined using Eqs. (14), (28), (35), (46) depending on type of cells used and compared to the average blocking probability considering both new and handoff call in each cluster of a only macrocell based 7-cluster LA calculated using Eq. (14) as shown in Figure 10. The average blocking probabilities in each cluster in the proposed LA and a macrocell based 7-cluster LA are presented in Figure 10. Threshold in this case is considered 1. As microcells are used in cluster 2 and cluster 6 replacing the macrocells, their average blocking probabilities are reduced. Subsequently, as picocells are used in cluster 3 and cluster 4 replacing the macrocells, their average blocking probabilities are reduced. As cluster 1 and cluster 7 suffered from temporary congestion, femtocells are allocated in cluster 1 and cluster 7 within the existing macrocell to avoid congestion, hence probabilities in these two clusters are reduced at a small amount. As cluster 5 never suffered from congestion, no modification is done to the existing macrocell, hence probability in this cluster has not been changed. Figure 10 demonstrates that using proposed picocell, microcell, femtocell and macrocell based LA average blocking probability can be minimized up to 25% than the only macrocell based LA. The satisfaction probability in each cluster of the proposed picocell, microcell, femtocell and macrocell based LA are determined using Eqs. (13) (29), (36), (47) depending on type of cells used and compared to the satisfaction probability in each cluster of a only macrocell based

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

if cell sizes are decided by assuming the traffic load and channels are allocated based on cell size, the total number of channels in a cluster can be increased without causing wastage of spectrum. As blocking probability is reduced in the proposed congestion control approach, hence the QoS can be improved. Moreover, the congestion can be controlled even if the adjacent cells are also overloaded i.e., the proposed congestion control approach is independent of the load status of the adjacent cells. Thus we can conclude that assuming traffic load based on users mobility information, a spectrum and energy efficient network can be developed with improved QoS.

RESEARCH ARTICLE

7. CONCLUSIONS Spectrum and energy efficiency are two important challenges in the area of mobile computing. Energy consumption minimization is required to reduce environmental pollution and spectrum efficiency is important to prevent wastage of spectrum without compromising with QoS even during congestion. Channel borrowing, priority based channel allocation, buffer maintenance, channel reservation, dynamic call admission control or cell breathing are well-known approaches in the field of congestion control. But some of these methods cannot be applied if most of the adjacent cells of an overloaded cell are also suffer from congestion and in some methods delay is introduced through buffer maintenance which degrades QoS and may cause call dropping. Thus a new congestion control scheme is introduced in this paper that achieves low power consumption, spectrum efficiency and is independent of the load status of other cells. In this paper a green network is developed where the cell size of a location area is decided based on the traffic load offered to each cluster in the location area. We have first considered a macrocell based network where each cluster of location area is covered by a macrocell. Then the number of users visiting each macrocell, the traffic load offered to the macrocell and the blocking probability considering both new and handoff call in each macrocell are predicted by the network based on users mobility information maintained in dynamic location area list in the profile of individual user. Based on traffic load spectrum allocation is done in this paper. The network is trained to become artificially intelligent so that it can assume the traffic load itself on the network. Depending on the traffic load assumed by the network congestion is detected, prevented or avoided. If the average number of users exceeds the capacity of the macrocell and the average blocking probability is approximately equals to 1, then congestion is detected and to control it, the macrocell is replaced by a number of picocells. If the average number of users approximately reaches the capacity of the macrocell and the average blocking probability ranges from 0.8 to 1, then to prevent congestion, the macrocell is replaced 18

Mukherjee and De

by a number of microcells. If average number of users in a macrocell is less than its capacity and the average blocking probability is less than 0.8 and but the macrocell suffers from congestion at a certain amount of time, then for avoiding congestion, femtocells are deployed and activated within this macrocell during congestion. The number of femtocells will be decided based on the number of excess users during congestion. When the macrocell becomes under loaded the femtocells contained in it are deactivated. If a macrocell has blocking probability less than 0.8 and average number of users is less than its capacity as well never suffers from congestion, it is not replaced i.e., no modification is done to the existing macrocell. Applying the proposed congestion detection, prevention and avoidance methods in each cluster of each location area contained in the network, the congestion can be controlled. The power consumption by the base stations in such a macrocell, microcell, picocell and femtocell based location area is calculated and compared to that of a macrocell based location area. Simulation results present that the macrocell, microcell, picocell and femtocell based location area achieves 35.29–40% less power consumption by the base stations than the only macrocell based location area. Simulation results also present that using macrocell, microcell, picocell and femtocell based LA as in proposed congestion control scheme approximately 50.91–57.74% minimization in total power consumption by the base stations can be achieved as compared to the power consumption involved in the existing congestion control scheme, cell breathing, applied to a only macrocell based LA. The average blocking probability in each cluster of such macrocell, microcell, picocell and femtocell based location area is calculated and compared to that of a macrocell based location area. Simulation results present that the average blocking probability in the macrocell, microcell, picocell and femtocell based location area is up to 25% less than the only macrocell based location area. Path loss can also be minimized using such macrocell, microcell, picocell and femtocell based location area as compared to a macrocell based location area because each of a microcell, picocell or femtocell has smaller coverage area than that of a macrocell. As the blocking probability and path loss can be reduced using macrocellmicrocell-picocell-femtocell based location area, improved QoS can be achieved. Thus it is observed that deploying cells of different size in a location area according to the traffic load macrocell-microcell-picocell-femtocell based network can be developed and low power consumption, less path loss and reduction in blocking probability can be achieved. Moreover allocating spectrum based on busy time spectrum allocation method considering user mobility based traffic load, wastage of spectrum can be reduced approximately 36% and hence spectrum efficiency can be achieved. Using proposed congestion control scheme satisfaction probability can be increased approximately J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Mukherjee and De

Congestion Detection, Prevention and Avoidance Strategies for an Intelligent

37.8% than channel borrowing based congestion control scheme in a congested network. Hence we can conclude that predicting traffic load based on users mobility information, a spectrum and energy efficient green mobile network can be achieved. Acknowledgments: Authors are grateful to Department of Science and Technology (DST) for sanctioning a research Project entitled “Dynamic Optimization of Green Mobile Networks: Algorithm, Architecture and Applications” under Fast Track Young Scientist scheme reference no.: SERB/F/5044/2012-2013 under which this paper has been completed.

References

J. Comput. Intell. Electron. Syst. 2, 1–19, 2013

Received: 12 April 2012. Accepted: 22 April 2012. 19

RESEARCH ARTICLE

1. S. Tabbane, An alternative strategy for location tracking. IEEE Journal in Selected Areas in Communications 13, 880 (1995). 2. http://www.femtoforum.org/fem2/about-femtocells.php. 3. http://www.femtoforum.org/femto/Files/File/Interference Management in UMTS Femtocells.pdf. 4. K. Han, Y. Choi, D. Kim, M. Na, and S. Choi, Optimization of femtocell network configuration under interference constraints. International Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks (2009), pp. 1–7. 5. S. Y. Lien, C. C. Tseng, K. C. Chen, and C. W. Su, Cognitive radio resource management for QoS guarantees in autonomous femtocell networks, IEEE International Conference on Communications p. 1 (2010). 6. A. Mukherjee and D. De, An inverse cell breathing based power management for congested mobile network, International Conference on Innovative Techno-management Solutions for Social Sector (IEMCON), (2012), pp. 378. 7. S. Wu, T. Chow, K. T. Ng, and K. F. Tsang, Improvement of borrowing channel assignment for patterned traffic load by online cellular probabilistic self-organizing map. Neural Computing and Applications 15, 298 (2006). 8. Y. Zhang and E. Salari, A hybrid channel allocation algorithm with priority to handoff calls in mobile cellular networks. Elsevier Computer Communications 32, 880 (2009). 9. G. S. Huang, Z. G. Chen, Q. H. Li, M. Zhao, and Z. Guo, A novel dynamic call admission control policy for wireless network. Journal of Central South University of Technology, Springer, vol. 17 (2010), pp. 110–116. 10. J. Hou and S. Papavassiliou, A dynamic reservation-based call admission control algorithm for wireless networks using the concept of influence curve. Telecommunication Systems 22, 299 (2003). 11. Y. M. Chen and C. L. Su, Meeting QoS requirements of mobile computing by dual-level congestion control, Proceedings of International Conference on Advances in Grid and Pervasive Computing (2007), pp. 241–251. 12. G. Y. Liu and G. Q. Maguire, A predictive mobility management algorithm for wireless mobile computing and communications, IEEE International Conference on Universal Personal Communications (1995), pp. 268. 13. T. S. Kim, M. Y. Chung, and D. K. Sung, Mobility and traffic analyses in three-dimensional PCS environments. IEEE Transactions on Vehicular Technology 47, 537 (1998). 14. T. S. Kim, M. Y. Chung, D. K. Sung, and M. Sengoku, Mobility and traffic analyses in three-dimensional indoor environments. IEEE Transactions on Vehicular Technology 47, 546 (1998).

15. M. Boulmalf, J. Abrache, T. Aouam, and H. Harroud, Traffic analysis for GSM networks, Proceedings of IEEE (2009), pp. 498–503. 16. M. Papadopouli, E. Raftopoulos, and H. Shen, Evaluation of shortterm traffic forecasting algorithms in wireless networks. Proceedings of IEEE (2006), pp. 102–109. 17. P. Darwood, I. Oppermann, S. Jakas, and W. Linton, Mobile network traffic forecasting, IEEE Conference on Vehicular Technology 6, 2932 (2000). 18. K. Zhang and L. Cuthber, Traffic congestion forecasting in WCDMA networks, International Conference on Communications and Networking, p. 992 (2008). 19. W. Y. Lee and I. F. Akyildiz, Spectrum-aware mobility management in cognitive radio cellular networks. IEEE Transactions on Mobile Computing 11, 529 (2012). 20. A. A. Adesegun, The impact of dynamic spectrum access network on third world countries: Spectrum allocation issues, network and economic growth (The African Tale), IEEE International Symposium on New Frontiers in Dynamic Spectrum Access Networks (2005), pp. 66–72. 21. S. Lixin, J. Jian, and L. Yanbin, Flexible spectrum usage for the future broadband mobile and fixed convergence. URSIGA, p. 1 (2011). 22. Y. Zhang and W. Li, A spectrum sharing scheme in two cellular wireless networks. LNICST, 48, 439 (2010). 23. Y. Liu, X. Tan, and A. A. Anghuwo, Joint power and spectrum allocation algorithm in cognitive radio networks. Journal of Systems Engineering and Electronics 22, 691 (2011). 24. J. Sarnecki, C. Vinodrai, A. Javed, P. O. Kelly, and K. Dick, Microcell design principles. IEEE Communications Magazine 31, 76 (1993). 25. V. Chandrasekhar, T. Muharemovic, Z. Shen, J. G. Andrews, and A. Gatherer, Power control in two-tier femtocell networks. IEEE Transactions on Wireless Communications 8, 4316 (2008). 26. A. Golaup, M. Mustapha, and L. B. Patanapongpibul, Femtocell access control strategy in UMTS and LTE. IEEE Communication Magazine 47, 117 (2009). 27. R. Y. Kim and J. S. Kwak, WiMAX femtocell: Requirements, challenges, and solutions. IEEE Communication Magazine 47, 84 (2009). 28. F. Pantisano, M. Bennis, R. Verdone, and M. Latva-aho, Interference management in femtocell networks using distributed opportunistic cooperation, IEEE Vehicular Technology Conference 1 (2011). 29. J. S. Wu, J. K. Chung, and Y. C. Yang, Performance study for a microcell hot spot embedded in CDMA macrocell systems. IEEE Transactions on Vehicular Technology 48, 47 (1992). 30. H. S. Jo, C. Mun, J. Moon, and J. G. Yook, Interference mitigation using uplink power control for two-tier femtocell networks. IEEE Transactions on Wireless Communications 8, 4906 (2009). 31. A. Mukherjee and D. De, DAS: An intelligent three dimensional cost effective movement prediction of active users in heterogeneous mobile network. Journal of Computational Intelligence and Electronic Systems, American Scientific Publishers, 1, 31 (2012). 32. T. S. Rappaport, Wireless Communications. PHI, Englewood Cliffs, NJ (1996). 33. M. Ghaderi and R. Boutaba, Call admission control in mobile cellular networks: A Comprehensive Survey. Wireless Communications and Mobile Computing Wiley 6, 69–93 (2006). 34. A. Mukherjee, S. Bhattacherjee, S. Pal, and D. De, Femtocell based green power consumption methods for mobile network. Elsevier Computer Networks 57, 162 (2012).