Methods for Moment Manipulation of Various Traffic Sessions in Modern Small Cell Environment Paramita Bhattacharya1, Samya Bhattacharya1,2 1 1 2 M/S Untamed Spectrum , Post Box 119 Espoo 02320, Finland ; School of Electronic and Electrical Engineering , University of 2 2 Leeds , LS2 9JT, United Kingdom 1 1 2 [email protected] ; [email protected] , [email protected] ABSTRACT In this paper, we exercise some important derivation involving moment manipulations to estimate various traffic sessions in modern small cell environment. In the light of some previous work on cellular environment, we derive accurate expressions for a proposed Semi-Markov model that accounts for the smaller population size of the users in modern small cells. The small cells are generally powered by access points (WiFi/mini base stations) and are typically used as mobile clouds, hotspots. We validate the analysis for small cell model with an existing cellular traffic model by increasing the population size of the users to a large value (typical for a macro/micro-cellular environment). Regardless of the use of communication technology and nature of user sessions (voice, video, data etc.), the generic analysis can provide a good insight to the state-of-the-art traffic dimensioning for small cells. CONCLUSION We developed some traffic estimation techniques using various moment manipulations for Semi-Markov model applicable in a small cell scenario. The analysis accounts for the limited population size of the users in modern small cells. It is observed that the distributions of the traffic streams largely vary from the large population scenario in traditional cells. These affect the quality of service (QoS) parameters such as congestions both for fresh session requests and hand-off requests. The deviation between the small cell model and large population model increases with the user mobility while the same decreases with the increase in population size. Thus, it can be concluded that as the cell size diminishes, small cell model becomes more appropriate for dimensioning. Further, if the cell size increases and population size becomes comparable with the large cell case, both small cell models and R-T models yield same results because population size shapes the fresh traffic to be Poisson nature. Thus, for accurate traffic dimensioning in small cells, it is essential to adopt proper modelling and estimation technique. Access this article online Quick Response Code Website: www.ijrws.org ISSN Citation Paramita Bhattacharya, Samya Bhattacharya, “Methods for Moment Manipulation of Various Traffic Sessions in Modern Small Cell Environment,” International Journal of Research in Wireless Systems (IJRWS), Vol. 2, No. 3, pp. 23–34, October, 2013 ISSN: 2320 - 3617 International Journal of Research in Wireless Systems (IJRWS), Volume 2, Issue 3, October (2013) 23 Methods for Moment Manipulation of Various Traffic Traffic Sessions in Modern Small Cell Environment I. INTRODUCTION With the rapid growth of mobile users, deployment of smaller cells powered by miniature base stations (BSs), such as access points (APs)) or WiFi clouds have become increasingly popular cost-effective effective choice for the mobile operators around the world. Supporting multimedia sessions in such cells require accurate traffic dimensioning to maintain various quality of service (QoS) requirements. In addition, to maintain seamless connectivity on the move, these small cells require better session hand-off off management. Therefore, hand-off hand traffic plays an important role in state-of-the-art state traffic analysis. The researchers in the past have extensively extensivel studied the above in the context of cellular networks. Early traffic analyses for cellular networks were carried out with M/M/C loss systems and single-moment moment approach where fresh (i.e. new) and hand-off off calls were assumed to be Poisson arrival processes [1]. In the context of micro-cellular cellular systems where a user undergoes frequent hand-offs, offs, it was shown [4] that the hand-off off process does not remain Poisson distributed. As a result, even if the fresh sh call arrival process can be represented as a Poisson distribution, the aggregated traffic stream is more realistically represented by a General distribution. It was also proposed [7] that in case of General distributed traffic, traff the two-moment moment representation of traffic (using mean and variance) is a better approach than a single-moment single representation. Rajaratnam and Takawira [7] in their studies pointed out that the hand-off off traffic is the main distinguishing dis features between the PSTN and mobile cellular networks by analysing the traffic using General distributed hand-off hand traffic. These analyses were applicable for densely populated macromacro micro cells. With diminishing cell sizes and thus user population tion in a cell, fresh call arrivals, or a new session request in the context of modern mobile environment, follow Engset distribution [10] instead of Poisson distribution. The corresponding effects on the traffic streams and user us sessions in small cell environment have not been thoroughly investigated in the literature to the best of our knowledge. In [11][12], authors suggested to follow the models developed in [7] for a less populated area. But the use of the above models leads to difficulties, when the fresh call arrival process is Engset distributed. Firstly, the estimation of traffic offered to the virtual cell in their model needs a Semi-Markov Semi analysis with Engset distribution. Secondly, the suggested use of Binomial-Poisson-Pascal (BPP) method in [7] yields inaccurate estimates of the carried traffic in a cell from the traffic offered to that cell when the peakedness akedness factor (or peakedness) of the offered traffic falls below 0.5 (common in presence of an Engset distributed traffic stream) [13].. Also, the suggested use of Girard's method [9] to estimate the he carried traffic in a cell from the traffic offered to that cell is limited to Pure Chance Type-I (PCT-I) I) traffic; making it inapplicable in the case of Engset distributed fresh call arrival process i.e. Pure Chance Type-II (PCT-II) traffic [10].. Thus, a thorough Semi Semi-Markov analysis for Engset distributed fresh call traffic and the corresponding traffic estimates in modern mobile environment merits investigation. Therefore, our contributions in this paper are two two-fold. Firstly, we propose a Semi-Markov Markov analysis for the Engset distributed fresh traffic sessions and General distributed handoff sessions in a small cell environment. Secondly, we derive two moments (mean and variance) of fresh and handoff traffic sessions and use tthose estimates to compute corresponding congestion parameters in a small cell. In Section II,, we derive various traffic estimates for the models developed in [7] for micro-cellular cellular environment. This forms the background for the proposed work in Section III. Section IV presents some numeric results related to the model verification and validation. Finally, we conclude in Section Error! Reference source not found. found.. II. RELATED WORK In this section we study the various traffic sessions related to traditional cellular networks. In cellular networks, a channel was considered to be a single call carrying resource [7]. Fixed Channel Allocation (FCA) was considered where the number of simultaneous ongoing calls or sessions is fixed in a cell. A traffic stream was represented by two moments (( ܯǡܸ) denoting the corresponding mean and variance. The hand hand-off traffic was obtained when a fraction of users from a cell hand handoff to an adjacent cell in their two two two-cell model (Fig. 1). Fig. 1. The two-cell model with a virtual ual cell [7]. Initially, it was assumed that the originating cell receives no hand-off off traffic. Later, the restriction was removed through an iterative process. To estimate the offered traffic to the target cell, the carried traffic affic estimate in the previous (originating) cell is needed. It has been shown that the variance of the carried traffic in the originating cell does not match with that of the traffic offered to the target cell, though the corresponding means match. This iis because the offered traffic used to be stochastically different from the carried traffic. Thus, a virtual (hypothetical) cell of infinite capacity has been introduced between the two cells. The carried traffic in the virtual cell (also known as freed car carried traffic [14]) will now be equivalent to the offered traffic in the target cell. Later, the two-cell cell model was generalised to a multi multi-cell International Journal of Research in Wireless Systems (IJRWS), Volume 2, Issue 3, October ber (201 (2013) 24 Methods for Moment Manipulation of Various Traffic Sessions in Modern Small Cell Environment model where the estimated hand-off traffic using the two-cell model was aggregated with fresh call offered traffic to form a combined traffic stream offered to a cell. Using various moment manipulation techniques presented in [13],[14], we in this section, present an alternative approach to [7] to formulate and derive the moments of the corresponding traffic streams in their two-cell model. These form the background of our work in the next section involving small cells. A session initiated by a mobile user may suffer blocking at two stages - (a) at session initiation, known as fresh blocking or fresh congestion, and (b) at the cell boundary, which is called hand-off blocking or hand-off congestion. Let us consider the ݅௧ cell ܮ (݅= 1, 2) in Fig. 1. Let ܥbe the number of channels assigned to the cell; ߣ = ߣ be the mean arrival rate of fresh requests to the cell, which is Poisson distributed. A mobile user vacates a channel in a cell due to either session termination or session hand-off to the other cell. Both processes are assumed to be Poisson distributed with respective mean rate ߤ௧ and ߤ. Thus, the session departure process from a cell is also Poisson distributed with mean rate ߤௗ = ߤ௧ + ߤ. It is imperative to state that the channel holding time is Negative Exponential distributed with mean 1/ߤௗ. It is noted that some studies modelled the channel holding time as Gamma distribution [15] or Hyper-Erlang distribution [16] or Lognormal distribution [17] or a combination of constant and Negative Exponential distribution [8]. However, considering the urban scenario [18], the channel holding time is modelled as Negative Exponential distribution in this work. Assuming that the originating cell ܮଵ initially receives no hand-off traffic, the offered traffic to that cell is Poisson distributed (fresh requests only) with mean ܣ = ߣ /ߤௗଵ. The carried traffic ( ܯభ ,ܸభ ) in the cell is obtained as below. A. Expression for the Carried Traffic in ࡸ Let there be ݇ ongoing sessions in ܮଵ. Then the probability ( ) that ܮଵ is in state [13], is given by ܣ ܣ ݇! = = ݇! ܣ ݁ ൫ܣ൯ ∑ୀ ݅! (1ܽ) ܣ = !ܥ = ܤ൫ܣ, ܥ൯(1ܾ) ܣ ∑ୀ ݅! where݁ ൫ܣ൯= ∑ୀ is the incomplete exponential ! function and is the Erlang loss function. Since the carried traffic has only the fresh component, it is sufficient to use Probability Generating Function (PGF) [13] alone, which is defined as G(z) = ∑ஶୀ ݖ . Using Eqn. 1(a), we obtain = )ݖ(ܩ Therefore, and ܩ ᇱ()ݖ = ܩ′′(= )ݖ ∑ୀ ݖ ܣ ݇! ܣ ∑ୀ ݅! ಲೖ ೖష భ ∑ ೖసభ ௭ ಲ ∑ సబ ! (2) (3) ೖ! ಲೖ ೖష మ ∑ ೖసమ (ିଵ)௭ ಲ ∑ స బ ! (4) ೖ! We express the mean ( ܯభ ) of the carried traffic from [13] as ܯభ = ܩᇱ(1) = = ಲೖ ∑ ೖసభ [Using Eqn. (3)] ೖ! ಲ ∑ సబ ! ಲೖ ∑ ೖసభ(ೖషభ)! ಲ ∑ సబ ! ಲ షభ ∑సబ ! ಲ ∑ సబ ! ಲ ಲ ∑ ି! సబ ! ಲ ∑ సబ ! ಲ =ܣ =ܣ = ܣ ቌ1 − ! ಲ ∑ సబ ! ቍ = ܣൣ1 − ܤ൫ܣ, ܥ൯൧[Using Eqn. (1b)] (5) We express the variance (ܸభ ) of the carried traffic from [13] as ଶ ܸభ = ܩᇱᇱ(1) + ܩᇱ(1) − ൫ܩᇱ(1)൯ ଶ = ܯభ − ቂ൫ܩᇱ(1)൯ − ܩᇱᇱ(1)ቃ[Using Eqn. (5)] (6) Using Eqn. (1b), (3), (4), and (5), we get ଶ ቂ൫ܩᇱ(1)൯ − ܩᇱᇱ(1)ቃ= ܣଶ(1 − )ଶ − ܣ (݇ − 2)! ܣ ∑ୀ ݅! ∑ୀଶ ܣ ݅! = ܣଶ(1 − )ଶ − ܣଶ ܣ ∑ୀ ݅! ܣ ܣିଵ ܣ ∑ୀ − − ݅! ܥ( !ܥ− 1)! = ܣଶ(1 − )ଶ − ܣଶ ܣ ∑ୀ ݅! ∑ିଶ ୀ International Journal of Research in Wireless Systems (IJRWS), Volume 2, Issue 3, October (2013) 25 Methods for Moment Manipulation of Various Traffic Sessions in Modern Small Cell Environment moment of the ݏ௧ traffic stream, where = ݏ1 denotes the overflow stream and = ݏ2 denotes the freed carried traffic stream [14]. In our notation, the mean ( ܯଵ) of the overflow traffic [14] is written as ܣ ܣ ܥ !ܥ൪ = ܣଶ(1 − )ଶ − ܣଶ ൦1 − !ܥ − ܣ ܣ ܣ ∑ୀ ∑ୀ ݅! ݅! ܥ = [Using Eqn. (1b)] ܣ = ܣଶ − 2ܣଶ + ܣଶଶ − ܣଶ + ܣଶ + ܣܥ = ܣܥ − ܣଶ + ܣଶଶ = ܣ ൣ ܥ− ܣ(1 − )൧ = ܣ ൣ ܥ− ܯభ ൧ ܣଶ(1 − )ଶ − ܣଶ ቈ1 − − 1 1 ݈! ܥ = = ቀ ቁ ݈ ( ܯାଵ) ܯଵ ܯଵ,(ଵ) ୀ Therefore, ܯଵ = = = ܣܤ൫ܣ, ܥ൯(7) = Substituting Eqn. (7) in Eqn. (6), we obtain ܸభ = ܯభ − ܣܤ൫ܣ, ܥ൯൫ ܥ− ܯభ ൯(8) ( ܯ) ିଵ 1 = ෑ ߤௗ ()ܣܧ ୀଵ ݆ߤ݆( ∗ܣௗ ) ݇ ∈ (ܫ9) 1 − ߤ݆( ∗ܣௗ ) where ߤௗ is the mean service rate (Poisson distributed), ܣܧis the mean inter-arrival time. If )ݏ( ∗ܣdenotes the L.S.T. of )ݐ( ∗ܣ, then for Poisson distributed offered traffic, we have = )ݏ( ∗ܣ Further, from [14], we get = ܣܧ−( ∗ܣ0) = − ߣ ݏ+ ߣ ଶอ ൫ݏ+ ߣ ൯ ௦ୀ = 1 ߣ Substituting ܣܧfrom Eqn. (11) in Eqn. (9) we obtain ( ܯ) = ଵ ఓ = భ ഊ ∏ିଵ ୀଵ ఒ ఓ ఒ = ቀ ቁ where = ܯ ఒ ఓ ఓ =ܯ ഊ ೕഋ శ ഊ ఓ (12) is the first factorial moment of the Poisson Let ܯ௦,() ( = ݏ1, 2 and݇ = 1, 2, …) be the ݇௧ factorial ! ಾ ! ∑ సబ [Using Eqn. (1b)] (13) From [13], the variance (ܸଵ) of the overflow traffic can be expressed in terms of the mean ( ܯଵ) and the second factorial moment ( ܯଵ,(ଶ) ) of the overflow traffic as ܸଵ = ܯଵ,(ଶ) − ܯଵ( ܯଵ − 1) (14) The second factorial moment ( ܯଵ,(ଶ) ) of the overflow traffic from [14] is written as 1 Therefore, ܯଵ,(ଶ) ܸଵ + ܯଵ( ܯଵ − 1) = ݈( ܥ+ 1)! = ቀ ቁ ݈ ( ܯାଶ) ୀ 1 ଵ !(శభ)! !(ష)!ಾ శమ ∑ స బ ଵ ! ಾ ష ∑ (ାଵ)(ష)! ಾ శమ స బ ಾ ெ మ ! ∑ సబ(ାଵି) = (ାଵ) ∑ ೖసబ = [Using Eqn. (14)] ݈( ܥ+ 1)! ∑ୀ ቀ ቁ ݈ ( ܯାଶ) = distributed offered traffic. C. Expression for the Overflow Traffic in ࡸ ಾ ! ಾ ష ∑ సబ(ష)! ಾ = ܯ(ܤ ܯ, )ܥ = ఒ !ಾ ష (ష)! = ೕഊ ೕഋ శഊ ଵି ∏ିଵ ୀଵ (11) [Using Eqn. (12)] ∑ స బ =ܯ (10) −ߣ ଵ ! ∑ స బቀቁಾ శ భ ெ =ܯ B. Factorial Moments for the Offered Traffic in ࡸ Let the inter-arrival time of the offered traffic be denoted as )ݐ(ܣ. From [14], the ݇௧ factorial moments ൫( ܯ) ൯ of the offered traffic is expressed as ଵ ! ∑ సబቀቁಾ ( శభ) ಾೖ ೖ! ಾ ெ మ ! ೕ ಾೖ భಾ ିெ ∑ష ೕసబ ೕ! ೖ! ெమ ಾ ! ಾೖ ∑ ೖసబ ೖ! ಾೕ ಾ ∑ ೕస బ ೕ! ష ! (ାଵ)ିெ ൦ ಾೖ ∑ ೖస బ ೖ! International Journal of Research in Wireless Systems (IJRWS), Volume 2, Issue 3, October (2013) ൪ 26 Methods for Moment Manipulation of Various Traffic Sessions in Modern Small Cell Environment ெ మ(ெ ,) = Therefore, ାଵିெ [ଵି(ெ ,)] ெ మெ భ = ାଵିெ ାெ భ భ ெభ + ( ܯଵ − 1) = = [Using Eqn. (13)] = ெ ାଵିெ ାெ భ ܯ ܯଵ ܸଵ = ܯଵ − ܯଵଶ + ܥ+ 1 − ܯ+ ܯଵ (15) D. Expression for the Freed Carried Traffic in ࡸࢂ From the conservation of flow, the mean ( ܯ ) of the freed carried traffic can be expressed as ܯ = ܯଶ = ܯ− ܯଵ=[1 − ܯ(ܤ, [ ])ܥUsing Eqn. (13)] (16) which is equal to the mean carried traffic. The second factorial moment ( ܯଶ,(ଶ) ) of the freed traffic from [14] is written as ܯଶ,(ଶ) = ܯଶ ( ܯଶ) − ܯଵ[ܸଵ + ܯଵ( ܯଵ − 1)] ( ܯଵ) ݉ ( ܯ ) ݈! ܥ × ቀ ቁ ቈ + 1(17) ݈ ( ܯାଵ) ( ܯ ାଵ) ୀଵ ୀଵ Using the relation between ordinary, central and factorial moments of a distribution [13], we rewrite Eqn. (17) as ܸଶ + ܯଶ( ܯଶ − 1) = ܯଶ × ∑ୀଵ ቀ ቁ ! ெ (శభ) ( ܯଶ) − ܯଵ[ܸଵ + ܯଵ( ܯଵ − 1)] ( ܯଵ) ெ ( ) ∑ୀଵ ெ ( శ భ) + 1൨ = ܯଶ ܯ− ܯଵ[ܸଵ + ܯଵ( ܯଵ − 1)] ∑ୀଵ = ܯଶ ܯ− ܯଵ[ܸଵ + ܯଵ( ܯଵ − 1)] ∑ୀଵ = ܯଶ ܯ− ܯଵ[ܸଵ + ܯଵ( ܯଵ − 1)] × ∑ୀଵ !ெ ష(శభ) (ି)! + ∑ୀଵ !ெ ష(శభ) (ି)! !ெ ష(శభ) (ି)! !ெ ష(శమ) (ାଵ) ଶ(ି)! ൨ ெ (ାଵ) ଶெ ቃ !ெ ష(ష ೖశభ) (ି) ିଵ = ܯଶ ܯ− ܯଵ[ܸଵ + ܯଵ( ܯଵ − 1)] ∑ ୀ ᇣᇧᇧᇧᇧᇧᇧᇤ ᇧᇧᇧᇧᇧᇧᇥ + !ெ ష (షೖశ మ) (ି)(ିାଵ) ିଵ ∑ ୀ ᇣᇧᇧᇧᇧᇧᇧᇧᇧᇤ ᇧᇧᇧᇧᇧᇧᇧᇧᇥ ଶ! ୀ(௦௬) Now, ∑ = ܣିଵ ୀ = !ெ ష(షೖశ భ) (ି) = ∑ !ܥܥିଵ ୀ ! ெ శ భ ! ୀ(௦௬) ∑ିଵ ୀ ! !ெ ష(ష ೖశభ) ெೖ ! − ! ! ெ − ∑ !ܥିଵ ୀଵ ∑ିଶ ୀ ெೕ ! = (18) ቂ∑ୀ ಾೖ ∑ೖసబ ೖ! ಾ ! ெ ଵ (ିଵ)! ! 1 − ெభ − =1+ and, ெభ ெభ − ∑ = ܤିଵ ୀ = = = = = ! ቃ− ಾ ! ಾೖ ∑ ೖసబ ೖ! ! ெ ൩− − ଵ ଵ (ெ ,) ቂ∑ୀ ಾೕ ೕ! ಾ ! ∑ ೕసబ ଵ (ெ ,) +1+ ெ ெೕ ! − 1 − ெ ! ெ ష భ − (ିଵ)!ቃ ಾ ! ಾೖ ∑ ೖసబ ೖ! − ቂ1 − ܯ(ܤ, )ܥ− ெ ெ ಾ ! ಾೖ ∑ ೖసబ ೖ! ൩ ܯ(ܤ, )ܥቃ [Using Eqn. (13)] (19) !ெ ష(ష ೖశమ) (ି)(ିାଵ) ! ∑ୀ ଶெ శ మ ! ∑ୀ ଶெ శ మ ! ெೖ ! ெೖ ! ଶ! ( ܥ− ݇)( ܥ− ݇ + 1) [ܥଶ + ܥ− 2 ݇ܥ+ ݇(݇ − 1)] ቂ(ܥଶ + ∑ )ܥୀ ଶெ శ మ ! ቂ(ܥଶ + ∑ )ܥୀ ଶெ శ మ ெೕ ! ಾೖ ೖ! ಾ ଶெ మ ! ಾ ! ଶ ಾೖ ∑ ೖసబ ೖ! ଵ ெ (ெ ,) ܯଶ ∑ୀ = − [1 − ܯ(ܤ, ]ܥ− ∑ ೖసబ ଵ ଶெ మ(ெ ,) ଶ −ܯଶ ெ ! ெೖ ! ெೖ ! ெೖ − 2∑ ܯܥୀ ெ ష భ − ܯଶ (ିଵ)!ቃ ܥଶ + ܥ− 2 ܯܥ+ 2ܯܥ − ܯܥ ಾ ! ಾೖ ∑ ೖసబ ೖ! ெೖ − 2∑ ܥୀଵ (ିଵ)! + ∑ୀଶ (ିଶ)!ቃ ൩ ெೕ ! + 2ܯܥ ಾ ! ಾೖ ∑ ೖసబ ೖ! ெ ! + +ܯଶ− [ܥଶ + ܥ− 2 ܯܥ+ 2 ܯ(ܤ ܯܥ, )ܥ+ ܯଶ − ܯ(ܤ ܯ, )ܥ− ܯ(ܤ ܯܥ, ])ܥ = ଵ ଶெ ெ భ [ܥଶ + ܥ− 2 ܯܥ+ ܯܥଵ + ܯଶ − ܯ ܯଵ] [Using Eqn. (13)] (20) Substituting ܣfrom Eqn. (19) and ܤfrom Eqn. (20) in Eqn. (18), we obtain ܸଶ + ܯଶ( ܯଶ − 1) = ܯଶ ܯ− ܯଵ[ܸଵ + ܯଵ( ܯଵ − 1)] ܥଶ + ܥ− 2 ܯܥ+ ܯܥଵ + ܯଶ ൬ ൰൨ ெభ (ெ ,) ଶெ ெ భ − ܯ ܯଵ ଵ = ܯଶ ܯ− [ܸଵ + ܯଵ( ܯଵ − 1)] ቂ ܯଵ + ܥ− ܯ+ (ܥଶ + ܥ− × 1 + − ଵ + ଵ 2 ܯܥ+ ܯܥଵ + ܯଶ − ܯ ܯଵ)ቃ = ܯଶ ܯ− ெ ష (షೖశభ) ெೖ ெ (ெ ,) (ெ ,) ܯ ∑ ୀଵ ቀ + 1ቁ ቂ݈+ = ! ெ శ భ = ܯଶ ܯ− = ܯଶ ܯ− Therefore, భାெ భ(ெ భିଵ) ଶெ ெ భିெ భమା ெభ ଶெ [Using Eqn. (13)] [ܥଶ − ܯଶ + ܥ+ ܯܥଵ + ܯ ܯଵ] ಾಾభ ାெ భ(ெ భିଵ) శ భషಾ శಾ భ ଶ(ାଵିெ మ) ܥଶ − ܯଶ + ܥ+ ܯܥଵ ൨ ଶெ + ܯ ܯଵ [Using Eqn. (15)] × [ܥଶ + ܥ+ ܯܥ− ܯܥଶ + ܯ ܯଶ] International Journal of Research in Wireless Systems (IJRWS), Volume 2, Issue 3, October (2013) [Using Eqn. (16)] 27 Methods for Moment Manipulation of Various Traffic Traffic Sessions in Modern Small Cell Environment మ ெమ ൌ ͳെ ܯଶ ܯെ = 1− = 1− ெభ ଶெ మ ଶெ మమ ቂ ெభ − ெ (ெ ǡ) (ெ ǡ) ଶ[ଵି(ெ ǡ)] =1− ଶ[ଵି(ெ ǡ)] =1− =1− ଶெ ெ మ ெభ ଶெ ൫ଵି(ெ ǡ)൯ =1− =1− ܥ(ܥ ͳ − ܯଶ) ൨ ଶெ మ(ାଵିெ మ) ( ܯ ( ܥ− ܯଶ) (ெ ǡ) ெభ ெభ ቂ ܥെ ʹ ܯଶ + Therefore, ܸ ൌ ܲଶܸ ܲଶ ܯଶ ܲ ܯ െ ܲଶ ܯ െ ܯଶ ൌ ܲଶ ܯ ܯଶ ܯ െ ܲଶ ܯ − ܯଶ [Using Eqn. (26a)] ൌ ܯ ܲଶ(ܸ െ ܯ ) ெ [Using Eqn. (26a)] ൌ ܯ ܲ (ܸ െ ܯ ) ெ (ିெ మ) ܥ ଶெ మ(ெ మିெ ) ቂ ܸ ܯଶ ൌ ܲଶ(ܸ ܯଶ) + (ܲ െ ܲଶ) ܯ ቃ ାଵିெ మ ܥ ெ ((ି ିெ మ) ାଵ ଵିெ మ ቃ ெೇ [Using Eqn. Eqn (13)] ெ (ିெ మ) ାଵିெ మ ቃ ൌ ܯ ቂͳ [Using Eqn. Eqn (16)] ቂ ܥെ ܯଶ െ ቀ ܯଶ − ெ ((ିெ ெ మ) ାଵିெ ெమ ቁቃ ܯ(ܤǡ)ܥ ܥ( ܯെ ܯଶ) ቈ ܥെ ܯଶ ܯଵ − ቆ ቆ ܯ− ቇ 2[ͳ െ ܯ(ܤǡ])ܥ ܥ ͳെ ܯଶ [Using Eqn. Eqn (16)] (ெ ǡ) ଶ[ଵି(ெ ǡ)] (ெ ǡ) ଶ[ଵି(ெ ǡ)] ெ ାெ ିெ ெ ெ మିெ ାெ ெ మ ቂ ܥെ ܯଶ ܯଵ െ ቀ ቂ ܥെ ܯଶ ܯଵ െ ቀ ெ ା ାଵିெ మ ାଵିெ మ ቁቃ ቁ (21) ቁቃ In the present case, ܯൌ ܣ, ܯଵ ൌ ܣܤ൫ ൫ܣǡܥ൯, ܯଶ ൌ ܯ and ܸଶ ൌ ܸ . Substituting the above ve quantities in Eqn. (21), we obtain ܸ ൌ ܯ ቈͳ െ ܤ൫ܣǡܥ൯ ൫ ܥെ ܯభ + ܣܤሺܣǡܥ൯ ʹൣͳ െ ܤ൫ܣǡܥ൯൧ ܣ (22) − ܥെ ܯభ + 1 E. Expression for the Hand-off Offered Traffic affic From [13], the first two ordinary moments (ߙ ( ଵǡߙଶ) of the aggregated offered traffic and the ݅௧ (݅ൌ 1, 2, …) individual traffic stream (ߙଵǡߙଶ) are related as ߙଵ ൌ ܲߙଵ (23a) ߙଶ ൌ ܲଶߙଵ (23b) where ܲ is the probability of a session belonging to t stream ݅. In the present case, there are two streams, viz. (i) the stream containing sessions that are completing within a cell and (ii) the stream containing sessions that are handed off to another cell. For the second stream (ܲ) is expressed as ܲ = ఓ ఓ ାఓ ൌ ߞ (24) The parameter ߞ is called the mobility [1]. Using the relation between the ordinary moments, mean and variance of a distribution [13], we write ߙଵ ൌ ܯ ߙଶ ൌ ܸ + ܯଶ ߙଵ ൌ ܯ ߙଶ ൌ ܸ + ܯଶ Using Eqn. (23), (24) and (25), we obtain ܯ ൌ ܲ ܯ = ఓ ఓ ାఓ ܯ ೇ ൌ ܯ ቂͳ ܲ ቀ (25a) (25b) (26a) ெೇ ఓ െ ͳቁቃ ೇ ቀ െ ͳቁቃ ቃ ఓ ାఓ ெ ೇ (26b) III. CASE OF FINITE POPULATION SIZE IN SMALL CELL WIFI / MOBILE CLOUD ENVIRONMENT As mentioned earlier, rlier, the fresh traffic in a small cell environment follows Engset distribution [10]. Thus, cell ܮଵ in the two-cell cell model receives Engset distributed fresh requests in our case instead of Poisson distributed fresh requests in case of [7]. Thus, the user population in ܮଵ is now finite. Let ܰ be the number of users in ܮଵ, ߣ ൌ ߣᇱ be the mean arrival rate of fresh sessions per idle user and ߤௗ be the mean departure rate from ܮଵ. We develop a joint Semi Semi-Markov model to obtain the variance of the offered traffic to ܮଶ. The Semi-Markov Markov model does not have a closed form solution and needs complex manipulation using Probability Generating Function (PGF) and Binomial Moment Generating Function (BMGF). A. Carried Traffic in ࡸ for the Engset type arrival The state transition diagram for the Engset distribution is shown in Fig. 2. Let be the probability of ܮଵin state ݅. Fig. 2.. State transitions of the Engset distribution. The cut equation in Fig. 2 yields ) ᇱିଵͲ ݅ ܥ ݅ߤௗ = (ܰ െ ݅ ͳ)ߣ ǡ݅ = ఒᇲ ఓ (ܰ െ ݅+ 1)ିଵ (27) The expression for the mean (( ܯ ) of the carried traffic is written as ܯ = ∑ୀ ݅ = ∑ୀ ఒᇲ ఓ = ∑ିଵ ୀ = ఒᇲ ఓ ఒᇲ ఓ (ܰ െ ݅ ͳ)ିଵ (ܰ െ ݅) ܰ ∑ିଵ ୀ − ఒᇲ ఓ ∑ିଵ ୀ ݅ International Journal of Research in Wireless Systems (IJRWS), Volume 2, Issue 3, October ber (201 (2013) 28 Methods for Moment Manipulation of Various Traffic Traffic Sessions in Modern Small Cell Environment ఒᇲ = ఓ ఒᇲ = ఓ ܰ (ͳ െ ) − ܰ (ͳ െ ) − ǡ ܯ ൬ͳ ܯ = ఒᇲ ఓ ఒᇲ ఓ ఒᇲ ఓ (∑ୀ ݅ െ ܥ ) since ∑ୀ = 1 ൫ ܯ െ ܥ ൯ ൰ൌ ఒᇲ ఓ [ܰ െ (ܰ െ ))ܥ ] ܰ ߣᇱ ܰെܥ ͳ െ ൨ ߣᇱ ߤௗ ܰ (28) The expression for the variance (ܸ ) of the carried traffic is written as ܸ = ∑ୀ(݅െ ܯ )ଶ = ∑ୀ ݅ଶ െ ʹ ܯ ∑ୀ ݅ ܯଶ ∑ୀ = ∑ୀ ݅ଶ െ ʹ ܯଶ ܯଶ = ∑ୀ ݅ଶ െ ܯଶ Now, let ܺ ൌ ∑ୀ ݅ଶ. (29) ǡ ܺ ൌ ∑ୀ ݅ή݅ ܩ (∑ = )ݖஶ ୀ ݉ ǡ ݖ ఒᇲ Therefore, ܩݖ (= )ݖ = ∑ୀ ݅ (ܰ െ ݅ ͳ)ିଵ = = ఓ ᇲ ିଵ ఒ ∑ୀ ݅ (݅ ͳ)(ܰ െ ݅) ఓ ఒᇲ ିଵ[ܰ݅ ∑ୀ െ ݅ଶ ܰ െ ఓ ఒᇲ ݅] ܰ ( ܯ െ ܥ ) − ∑ିଵ ( െ ) ୀ ݅ ܰ (ͳ ൨ ( െ ܥ ) −(ܯ ఒᇲ ఒᇲ ܰ ܰ ܯ − ܯ ܥଶ ǡ൬ͳ ൰ܺ ൌ ൨ ఓ ఓ െ ܰܥ ܥ െ ܰ = = = ఓ ఒᇲ ఓ ఒᇲ ఓ ∑ିଵ ୀ 1) ] Hence, ܸ = ߣᇱ ఒᇲ ାఓ [ܰ ܯ (ܰ െ ͳ) + ( ܥെ ܰ ) ( ܥ ߣᇱ [ܰ ܯ (ܰ െ ͳ) ߤௗ + ( ܥെ ܰ )( ܥ ͳ) ]] ܯଶ From Eqn. (32a), we obtain (32b) ܩᇱ(∑ = )ݖஶ ୀଵ ݉ ǡ ݖ ିଵ = ∑ஶ ୀ(݉ ͳ)ǡ ାଵݖ (32c) (32d) Further, the Binomial Moment Generating Function (BMGF) of the joint probability distribution is defined as ܨ ( )ݔൌ ܩ ( ݔ ͳ)) = ∑ஶ ୀ ߚǡ ݔ [ܰ ܯ (ܰ െ ͳ) + ( ܥെ ܰ )( ܥ ͳ) ) ] ǡ ܺ ൌ (32a) ∑ஶ ୀଵ ǡ ିଵݖ Therefore, ܩݖᇱ(∑ = )ݖஶ ୀଵ ݉ ǡ ݖ = ∑ஶ ୀ ݉ ǡ ݖ [ܰ ܯ (ܰ െ ͳ) ܥ ( ܥെ ܰ ) (ܥ ( െ ܰ )] ఒᇲ Fig. 3. Semi-Markov Markov model for the carried sessions in ܮଵ and offered sessions to ܮ . (30) (31) As mentioned before, mean of the offered traffic to ܮ is equal to the mean of the carried traffic in ܮଵ. Therefore, we need to find out variance of the traffic in ܮ . B. Semi-Markov Analysis to o estimate the variance of the traffic in ࡸࢂ The state space Markov diagram for the carried sessions in ܮଵ and ܮ is shown in Fig. 3. Let ǡ denotes the joint probability distribution, when there are ݊ carried sessions in cell ܮଵ and ݉ sessions are carried to ܮ . The PGF of the joint probability distribution is defined as ݍ ߚǡ = ∑ୀ ቀ ቁ ቁ ݉ (33a) (33b) ݇ = ∑ஶୀ(−1)ି ൬ ൰(33c) ݍ From the above Markov state diagram, the B B-D equations in terms of joint probability distribution are written as ൫ܰ ߣᇱ ݉ ߤௗଶ൯ǡ ൌ ߤௗଵଵǡ ିିଵ ( 1)ߤௗଶǡ ାଵ݊ ൌ Ͳ(34a) +(݉ ൣ(ܰ െ ݊)ߣᇱ ݊ߤௗଵ ݉ ߤௗଶ൧ǡ = (݉ ͳ)ߤௗଶǡ ାଵ ( + 1)ߤௗଵାଵǡ ିଵ +(ܰ െ ݊ ͳ)ߣᇱିଵǡ + (݊ Ͳ ൏ ݊ ൏ (ܥ34b) (݉ ߤௗଶ ߤܥௗଵ)ǡ = (݉ 1 1)ߤௗଶǡ ାଵ +(ܰ െ ܥ+ 1)ߣᇱିଵǡ ݊ ൌ (ܥ34c) For ݊ ൌ Ͳ,, we transform Eqn. (34a) in terms of PGF, using Eqn. (32), as International Journal of Research in Wireless Systems (IJRWS), Volume 2, Issue 3, October ber (201 (2013) 29 Methods for Moment Manipulation of Various Traffic Sessions in Modern Small Cell Environment ܰ ߣᇱܩ( )ݖ+ ߤௗଶܩݖᇱ(ߤ = )ݖௗଵܩݖଵ( )ݖ+ ߤௗଶܩᇱ()ݖ ⟹ ܰ ߣᇱܩ( )ݖ+ ߤௗଶ(ݖ− 1)ܩᇱ(ߤ = )ݖௗଵܩݖଵ()ݖ ⟹ ܰ ߣᇱܩ( ݔ+ 1) + ߤௗଶ(ܩ)ݔᇱ( ݔ+ 1) = ߤௗଵ( ݔ+ 1)ܩଵ( ݔ+ 1) Similarly, for 0 < ݊ < ܥ, Eqn. (34b) leads to ஶ (35a) (ܰ − ݊)ߣᇱܩ ( )ݖ+ ߤௗଵ݊ܩ ( )ݖ+ ߤௗଶܩݖᇱ()ݖ = ߤௗଶܩᇱ( )ݖ+ (ܰ − ݊ + 1)ߣᇱܩିଵ()ݖ + (݊ + 1)ߤௗଵܩݖାଵ()ݖ +(ܰ − ݊ + 1)ߣᇱܩିଵ( ݔ+ 1) Similarly, for ݊ = ܥ, Eqn. (34c) leads to ߤௗଶ ݔ ߚ, ݉ ݔ ିଵ + ߤܥௗଵ ߚ, ݔ ୀ ஶ ୀ = (ܰ − ݊ + 1)ߣᇱ ߚିଵ, ݔ for݊ = (ܥ37c) ୀ ߤௗଶߚ,ଵ + ܰ ߣᇱߚ,ଵ = ߤௗଵߚଵ, + ߤௗଵߚଵ,ଵ ⟹ ൫ܰ ߣᇱ + ߤௗଶ൯ߚ,ଵ = ߤௗଵߚଵ,ଵ + ߤௗଵଵᇱ where (35b) ߤௗଶܩݖᇱ()ݖ + ߤܥௗଵܩ ()ݖ = ߤௗଶܩݖᇱ( )ݖ+ (ܰ − ܥ+ 1)ߣᇱܩିଵ()ݖ ⟹ ߤௗଶ(ݖ− 1)ܩᇱ( )ݖ+ ߤܥௗଵܩ ( ܰ( = )ݖ− ܥ+ 1)ߣᇱܩିଵ()ݖ ⟹ ߤௗଶ(ܩ)ݔᇱ( ݔ+ 1) + ߤܥௗଵܩ ( ݔ+ 1) = (ܰ − ܥ+ 1)ߣᇱܩିଵ( ݔ+ 1) for 0 < ݊ < (ܥ37b) For ݊ = 0, equating coefficients of ݔfrom Eqn. (37a), we get ݊)ߣᇱ ⟹ ߤௗଶ(ݖ− 1)ܩᇱ( )ݖ+ ൣ(ܰ − + ݊ߤௗଵ൧ܩ ()ݖ = (݊ + 1)ߤௗଵܩݖାଵ()ݖ + (ܰ − ݊ + 1)ߣᇱܩିଵ()ݖ ᇱ(ݔ ⟹ ߤௗଶܩݔ + 1) + ൣ(ܰ − ݊)ߣᇱ + ݊ߤௗଵ൧ܩ ( ݔ+ 1) = (݊ + 1)ߤௗଵ( ݔ+ 1)ܩାଵ( ݔ+ 1) × ∑ஶ ୀ ߚିଵ, ݔ (35c) Transforming the set of Eqn. (35) in terms of BMGF using Eqn. (33), we obtain ߚ, = ᇱ = ᇲ (38a) ഊ ே ቀ ቁ൭ ൱ ഋ భ ᇲ ೖ ഊ ே ∑ ೖసబቀ ቁ൭ഋ ൱ భ ᇣᇧᇧᇧᇧᇧᇤ ᇧᇧᇧᇧᇧᇥ ౪౪ౌ౨ౘౘౢ౪౯౪ ుౝ౩౪ౚ౩౪౨ౘ౫౪[భబ] (38b) For 0 < ݊ < ܥ, equating coefficients of ݔfrom Eqn. (37b), we get ߤௗଶߚ,ଵ + ൣ(ܰ − ݊)ߣᇱ + ݊ߤௗଵ൧ߚ,ଵ = (݊ + 1)ߤௗଵ൫ߚାଵ, + ߚାଵ,ଵ൯+ (ܰ − ݊ + 1)ߣᇱߚିଵ,ଵ ൣ(ܰ − ݊)ߣᇱ + ߤௗଶ + ݊ߤௗଵ൧ߚ,ଵ ᇱ = (݊ + 1)ߤௗଵ൫ߚାଵ,ଵ + ାଵ ൯+ (ܰ − ݊ + 1)ߣᇱߚିଵ,ଵ (38c) For ݊ = ܥ, equating coefficients of ݔfrom Eqn. (37c), we get ߤௗଶߚ,ଵ + ߤܥௗଵߚ,ଵ = (ܰ − ܥ+ 1)ߣᇱߚିଵ,ଵ ⟹ (ߤௗଶ + ߤܥௗଵ)ߚ,ଵ = (ܰ − ܥ+ 1)ߣᇱߚିଵ,ଵ (38d) ߤௗଶ2ߚ,ଶ + ܰ ߣᇱߚ,ଶ = ߤௗଵߚଵ,ଵ + ߤௗଵߚଵ,ଶ (39a) ܰ ߣᇱܨ( )ݔ+ ߤௗଶܨݔᇱ(ߤ = )ݔௗଵ( ݔ+ 1)ܨଵ( )ݔfor݊ = 0 (36a) ߤௗଶܨᇱ()ݔ+ൣ(ܰ − ݊)ߣᇱ + ݊ߤௗଵ൧ܨ ( ݊( = )ݔ+ 1)ߤௗଵ( ݔ+ 1) For ݊ = 0, equating coefficients of ݔଶ from Eqn. (37a), we obtain ߤௗଶܨݔᇱ( )ݔ+ ߤܥௗଵܨ ( ܰ( = )ݔ− ݊ + 1)ߣᇱܨିଵ()ݔ For 0 < ݊ < ܥ, equating coefficients of ݔଶ from Eqn. (37b), we obtain × ܨାଵ( )ݔ+ (ܰ − ݊ + 1)ߣᇱܨିଵ( )ݔfor 0 < ݊ < (ܥ36b) for݊ = (ܥ36c) Expressing the set of Eqn. (36) in terms of ߚ, , we obtain ஶ ஶ ୀ ୀ ߤௗଶ ݔ ߚ, ݉ ݔ ିଵ + ܰ ߣᇱ ߚ, ݔ ஶ ஶ = ߤௗଵ( ݔ+ 1) ߚଵ, ݔ ୀ for݊ = 0 (37a) ஶ ߤௗଶ ݔ ߚ, ݉ ݔ ିଵ + ൣ(ܰ − ݊)ߣᇱ + ݊ߤௗଵ൧ ߚ, ݔ ୀ ஶ ୀ = (݊ + 1)ߤௗଵ( ݔ+ 1) ߚାଵ, ݔ + (ܰ − ݊ + 1)ߣᇱ ୀ ߤௗଶ2ߚ,ଶ + ൣ(ܰ − ݊)ߣᇱ + ݊ߤௗଵ൧ߚ,ଶ = (݊ + 1)ߤௗଵ × ൫ߚାଵ,ଵ + ߚାଵ,ଶ൯+ (ܰ − ݊ + 1)ߣᇱߚିଵ,ଶ (39b) For ݊ = ܥ, equating coefficients of ݔଶ from Eqn. (37c), we obtain ߤௗଶ2ߚ,ଶ + ߤܥௗଵߚ,ଶ = (ܰ − ܥ+ 1)ߣᇱߚିଵ,ଶ Adding Eqn. (39a), (39b) and (39c), we obtain (39c) ߤௗଶ൫2ߚ,ଶ + 2ߚଵ,ଶ + ⋯ + 2ߚ,ଶ൯ +ߤௗଵ൫ߚଵ,ଶ + 2ߚଶ,ଶ + ⋯ + ߚܥ,ଶ൯ + ߣᇱൣܰ ߚ,ଶ + (ܰ − 1)ߚଵ,ଶ + ⋯ + (ܰ − ܥ+ 1)ߚିଵ,ଶ] International Journal of Research in Wireless Systems (IJRWS), Volume 2, Issue 3, October (2013) 30 Methods for Moment Manipulation of Various Traffic Sessions in Modern Small Cell Environment = ߤௗଵ൫ߚଵ,ଵ + 2ߚଶ,ଵ + ⋯ + ߚܥ,ଵ൯ +ߤௗଵ൫ߚଵ,ଶ + 2ߚଶ,ଶ + ⋯ + 2ߚ,ଶ൯ +ߣᇱൣܰ ߚ,ଶ + (ܰ − 1)ߚଵ,ଶ + ⋯ + (ܰ − ܥ+ 1)ߚିଵ,ଶ൧ ୀ ୀ ⟹ ߤௗଶ2 ߚ,ଶ = ߤௗଵ ݊ߚ,ଵ ⟹ 2 ߚ,ଶ = ୀ ߤௗଵ ݊ߚ,ଵ ߤௗଶ ୀ (40) nd Where ∑େ୬ୀ β୬,ଶ is the complete 2 Binomial moments of the offered traffic to ܮ . The terms β୬,ଵ are the partial Binomial moments of the offered traffic to ܮ . From [13], using Eqn. (32), we express the variance of the offered traffic to L as ܸ = ܩᇱᇱ(1) + ܩᇱ(1) − [ܩᇱ(1)]ଶ = ܩᇱᇱ(1) − ܯଶ + ܯ (41) Using Eqn. (32) and (33), we obtain ܩ ( = )ݖ = ஶ ஶ ߚ, (ݖ− 1) ୀ ୀ ୀ ୀ ୀ (−1) + ݉ (ݖ−1) ିଵ = ݉ ଶ ൩ + ቀ ቁ( ݖ−1) ିଶ + ⋯ 2 ܩᇱ( = )ݖ ஶ 0 + ݉ (−1) ିଵ ߚ, ൨ +݉ (݉ − 1)(ݖ−1) ିଶ + ⋯ ୀ ୀ ஶ ܩᇱᇱ( = )ݖ ୀ ୀ Now = ݉ (݉ − 1)(݉ − 2)(݉ − 3) + ⋯ ] For݉ = 0 … 2 ܩᇱᇱ(1) = 2ߚ,ଶ ܩᇱᇱ(1) = ൣ2ߚ,ଶ + [(−3 ∙ 2) + (3 ∙ 2)]ߚ,ଷ൧ ܩᇱᇱ(1) 2ߚ,ଶ + [(−3 ∙ 2 + 3 ∙ 2)]ߚ,ଶ = ቈ +(4 ∙ 3 − 3 ∙ 2 + 4 ∙ 3 ∙ 2)ߚ,ସ ୀ = ݉ݎܨ0 … 3 ୀ = ݉ݎܨ0 … 4 ⋮ ୀ ⋮ ൨ (43) = 2ߚ,ଶ ୀ Finally, from Eqn. (41), we obtain ܸ = 2 ߚ,ଶ − ܯଶ + ܯ = ୀ ߤௗଵ ߤௗଶ ݊ߚ,ଵ − ܯଶ + ܯ ୀ [From Eqn. (40)] (44) We solve the equation set (38) for ߚ,ଵ as follows. Consider a × ܥ1 matrix ܺ that contains the ܥ+ 1 solutions in terms of ߚ,ଵ. Further, consider a ܥ × ܥmatrix ܣand a × ܥ1 matrix ܤ such that ܤ = ܺܣ. We obtain the elements of ܣand ܤfrom Eqn. (38) as ܣ, = −(ܰ − ݉ )ߣᇱ for 0 ≤ ݊ ≤ ܥ, ݉ = ݊ − 1 ⎧ ⎪(ܰ − ݊)ߣᇱ + ߤ + ݊ߤ ௗଵ ௗଶ for 0 ≤ ݊ ≤ ܥ, ݉ = ݊ (45a) for 0 ≤ ݊ ≤ ܥ, ݉ = ݊ + 1 −(݊ + 1)ߤௗଵ ⎨ ⎪ for݊ = ܥ, ݉ = ݊ ߤௗଵ + ߤܥௗଶ ⎩ (݊ + 1)ߤௗଵାଵ for 0 ≤ ݊ ≤ ܥ ܤ, = ቄ (45b) 0 otherwise IV. SMALL CELL MODEL VALIDATION AND DISCUSSION ିଶ 0 + 0 + ݉ (݉ − 1)(−1) ߚ, +݉ (݉ − 1)(݉ − 2) Once we compute ܺ, the variance ܸ can be easily computed from Eqn. (44). Finally, following the steps in Section (II -E), mean and variance of the hand-off offered traffic to ܮଶ can easily be computed. Since, various types of congestions are important QoS parameters in traffic dimensioning, we compare various types of congestions [12] estimated using our analysis for small cell model with that of [7] in the next section. ݉ ቀ ቁݖ (−1) ି ݇ Therefore, ܩᇱᇱ(1) (42) We validate the proposed model by considering a special case where it reduces to some already existing model. As mentioned in the literature, when the population size increases and theoretically tends to infinity, the Engset distribution behaves as Poisson distribution. In our small cell case, we observed that around ܰ ≥ 150, the corresponding fresh traffic behaves as Poisson distribution. Thus, we choose ᇱ ܰ , ߣ such that some equivalent fresh offered traffic (Poisson distributed) is injected into the proposed model. In such a condition, the congestions estimated using the proposed small cell model are compared with (a) the model proposed by Rajaratnam and Takawira (R-T Model) [7] and (b) the model proposed by Foschini et al. (Foschini’s Model) [1]. Table I presents the estimated congestion (fresh and handoff) for the above models. In the first case (Table I-a), the number of users in a small cell is varied from 150 to 430, while other parameters such as user mobility, session duration, session request rate, cell capacity (channels) are International Journal of Research in Wireless Systems (IJRWS), Volume 2, Issue 3, October (2013) 31 Methods for Moment Manipulation of Various Traffic Traffic Sessions in Modern Small Cell Environment (a) Varying number of users in a small cell. kept constant. In the second case (Table I-b) b) the user mobility in a small cell is varied from 0 (static) to 0:9 (highly mobile), while other parameters such as number of users, us session duration, session request rate, cell capacity (channels) are kept constant. Table I shows that all models are in good agreement. This validates the correctness of the small cell model. We further observe that in Table I-a, I both types of congestions increase with the number of users, which is expected. In Table I-b, we notice that the fresh congestion decreases with mobility. This is expected because the dropped hand-off off requests allow greater chances chan of acceptance to the fresh requests. However, hand-off hand congestion decreases with user mobility as well, which is unusual. This happens because of the insensitivity of the existing models towards population size. The existing models treat hand-off traffic ic as fresh traffic from the same cell of large population size. Hence, the phenomena nomena for fresh traffic repeat for hand-off off traffic. This reveals the key limitation of the existing models making them inapplicable for small cells. Figure 4 shows the percentage of deviation in congestion estimates using the Small Cell Model with that of the R-T R Model. Since the variance of the fresh requests, initiated from a finite population size, is smaller compared to that initiated from a large population opulation size, we observe that the R-T R Model largely over estimates fresh congestion. The hand-off hand congestion is also over estimated because a part of the fresh session requests hand-offs offs to the adjacent cell. We observe in (b) Varying user mobility in a small cell. Fig. 5. Percentage deviation in congestion (Small Cell Model versus R R-T Model). Fig. 6(a) that the deviation decrease decreases with the increase in number of users. This can be attributed to the fact that with increasing number of population the fresh request distribution (Engset) tends more towards Poisson distribution. Thus, the variance of the fresh offered traffic increases and the he difference with the case of large population size decreases. Since the fresh request is directly affected by the population size, the deviation is more in this case. However, with the increase in user mobility in Fig. 7(b), the deviation for both fresh requests and hand-off request increase drastically. This can be explained as follows. In the Small Cell Model, hand-off off congestion increases with the mobility, which is not the case for the R R-T Model. Thus, the estimates largely differs for the hand hand-off congestion. In case of fresh congestion, deviation occurs from two phenomena. Firstly, the small population size in the Small Cell Model limits the variance of the fresh session requests which, in turn, International Journal of Research in Wireless Systems (IJRWS), Volume 2, Issue 3, October ber (201 (2013) 32 Methods for Moment Manipulation of Various Traffic Sessions in Modern Small Cell Environment makes the fresh traffic more predictable. This reduces the fresh congestion. Secondly, the increasing hand-off congestion favours the fresh session requests and reduces the fresh congestion. R-T Model is insensitive to these phenomena. [9] M. Rajaratnam and F. Takawira, “Nonclassical traffic modeling and performance analysis of cellular mobile networks with and without channel reservation,” IEEE Trans. Veh. Technol., vol. 49, no. 3, pp. 817–834, May 2000. ACKNOWLEDGMENT [10] V. B. Iversen, Teletraffic Engineering Handbook, revised ed. Building 345v, DK-2800 Kgs. Lyngby: COM Center, Technical University of Denmark, May 2006, vol. ITU-D SG 2/16 & ITC Draft 2001-06-20. The authors acknowledge Prof. H. M. Gupta, Department of Electrical Engineering, Indian Institute of Technology Delhi, and Vivek Pal, Untamed Spectrum for their valuable comments and suggestions. REFERENCES [1] G. J. Foschini, B. Gopinath, and Z. Miljanic, “Channel cost of mobility,” IEEE Trans. Veh. Technol., vol. 42, no. 4, pp. 414–424, November 1993. [2] D. Hong and S. S. Rappaport, “Traffic model and performance analysis for cellular mobile radio telephone systems with prioritized and nonprioritized handoff procedures,” IEEE Trans. Veh. Technol., vol. VT-35, no. 3, pp. 77–92, August 1986. [3] D. Macmillan, “Traffic modeling and analysis for cellular mobile networks,” in Proc. 13th ITC, 1991, pp. 627–632. [4] E. Chlebus and W. Ludwin, “Is handoff traffic really poissonian?” in Proc. IEEE ICUPC ’95 Conf. Record, November 1995, pp. 348–353. [11] M. Rajaratnam and F. Takawira, “Hand-off traffic characterisation in cellular networks under engset new call arrivals and generalised service time,” in Proc. IEEE Conf. Wireless Commun. and Netw. WCNC’99, 21-24 September 1999, pp. 423–427. [12] M. Rajaratnam and F. Takawira, “Two moment analysis of channel reservation in cellular networks,” in Proc. AFRICON’99. Capetown, South Africa: IEEE, 28 Sept.-1 Oct. 1999,pp. 257–262. [13] A. Girard, Routing and Dimensioning in Circuit-Switched Networks. Addison-Wesley Publication, 1990. [14] A. Brandt and M. Brandt, “On the moments of overflow and freed carried traffic for the GI/M/C/0 system,” Methodol. Comput. Appl. Probab., vol. 4, pp. 69–82, 2002. [5] H. Zeng and I. Chlamtac, “Handoff traffic distribution in cellular networks,” in Proc. WCNC’99, New Orleans, September 1999, pp. 413–417. [15] Y. Fang and Y. Zhang, “Call admission control schemes and performance analysis in wireless mobile networks,” IEEE Trans. Veh. Technol., vol. 51, no. 2, pp. 371–381, March 2002. [6] D. R. Manfield and T. Downs, “Decomposition of traffic in loss systems with renewal inputs,” IEEE Trans. Commun., vol. COM-27, pp. 44–58, January 1979. [16] Y. Fang and I. Chlamtac, “Teletraffic analysis and mobility modeling of pcs networks,” IEEE Trans. Commun., vol. 47, no. 7, pp. 1062–1071, July 1999. [7] M. Rajaratnam and F. Takawira, “Hand-off traffic modelling in cellular networks,” in Proc. IEEE Conf. Global Telecommun. Phoenix, AZ: GLOBECOM’97, 3-8 November 1997, pp. 131–137. [17] A. E. Xhafa and O. K. Tonguz, “Does mixed lognormal channel holding time affect the handover performance of guard channel scheme?” in Proc. IEEE Conf. Global Telecommun. San Francisco, CA: GLOBECOM’03, 1-5 December 2003, pp. 3452–3456. [8] M. Rajaratnam and F. Takawira, “The two moment performance analysis of cellular mobile networks with and without channel reservation,” in Proc. IEEE ICUPC’98, Florence, Italy, October 1998, pp. 1157–1161. [18] R. Guerin, “Channel occupancy time distribution in a cellular radio system,” IEEE Trans. Veh. Technol., vol. VT35, no. 3, pp. 89–99, 1987. Paramita Bhattacharya received the B.Sc. degree in Science from University of Calcutta, India in 2002 and M.Sc. degree in Information Technology from Sikkim Manipal University, India in 2005. She worked as a project assistant at the Department of Electrical Engineering, Indian Institute of Technology Delhi in 2008. Currently, she is associated with M/S Untamed Spectrum Ltd., Finland. Her research interests include mobile networks, smart grid communication, next generation small cell and carrier WiFi networks. International Journal of Research in Wireless Systems (IJRWS), Volume 2, Issue 3, October (2013) 33 Methods for Moment Manipulation of Various Traffic Sessions in Modern Small Cell Environment Samya Bhattacharya received the B.E. degree in Electrical Engineering from Jadavpur University, Kolkata, India in 2000 and M.Tech. degree in Computer Science and Engineering from Vellore Institute of Technology (V.I.T. University), Vellore, India in 2002. From July 2002 to December 2002, he was attached to the Advanced Computing and Microelectronics Unit (ACMU), Indian Statistical Institute (I.S.I.), Kolkata, India as a project trainee. In 2008, he obtained the Ph.D. degree in Electrical Engineering from Indian Institute of Technology (I.I.T.) Delhi, New Delhi, India. Currently, he is a Research Fellow and Project Manager at the School of Electronic and Electrical Engineering, University of Leeds, United Kingdom and also is an advisory board member of M/S Untamed Spectrum Ltd., Finland. Dr. Bhattacharya is a member of technical program committee for IEEE Sarnoff Symposium, IEEE ICC 2010-14, IEEE GLOBECOM 2010-14, ICIT workshop and a number of conferences organised by the IEEE Communication Society. His current research interests include performance and energy efficient traffic engineering, network calculus, call admission control, mobility modeling and quality of service in next generation communication systems. International Journal of Research in Wireless Systems (IJRWS), Volume 2, Issue 3, October (2013) 34
© Copyright 2024 Paperzz