Multi-Cluster based Dynamic Channel Assignment for Dense Femtocell Networks

Abstract

This paper proposes a novel channel assignment scheme called multi-cluster based dynamic channel assignment (MC-DCA) to improve system performance for the downlink of dense femtocell networks (DFNs) based on orthogonal frequency division multiple access (OFDMA) and frequency division duplexing (FDD). In order to dynamically assign channels for femtocell access points (FAPs), the MC-DCA scheme uses a heuristic method that consists of two steps: one is a multiple cluster assignment step to group FAPs using graph coloring algorithm with some extensions, while the other is a dynamic subchannel assignment step to allocate subchannels for maximizing the system capacity. Through simulations, we first find optimum parameters of the multiple FAP clustering to maximize the system capacity and then evaluate system performance in terms of the mean FAP capacity, unsatisfied femtocell user equipment (FUE) probability, and mean FAP power consumption for data transmission based on a given FUE traffic load. As a result, the MC-DCA scheme outperforms other schemes in two different DFN environments for commercial and office buildings.

1. Introduction

According to recent reports, mobile data traffic volume from smart phones, tablets, and so on is growing dramatically and more than 80% of them occur in indoor environments [1,2]. In order to solve these problems, the femtocell network has become a promising solution since femtocells improve both the system capacity and coverage with low cost and low energy consumption [3-6]. Therefore, the world’s major mobile network operators (MNOs) have been showing a great deal of attention to adopt femtocells for next generation mobile networks such as 3GPP LTE/LTE-Advanced [7-9] and IEEE 802.16m [10,11]. However, in spite of the advantages of femtocells, various technical challenges still remain to enhance system performance. Channel assignment considering interference mitigation is one of the main issues because femtocell access points (FAPs) use the licensed spectrum owned by the macrocell network, and thus have cross-tier and co-tier interference from macro base stations (MBSs) and neighbor FAPs, respectively [12,13].

In recent literature, several channel assignment schemes have been studied for femtocell networks. Early channel assignment schemes mostly aimed to mitigate cross-tier interference between macrocells and residential femtocell networks (RFNs) in which each detached house has one or more FAPs [14-21]. In [14-17], authors proposed channel assignment schemes based on frequency reuse (FR) or fractional frequency reuse (FFR) to assign different channels for macrocell and femtocell networks to improve system performance. However, from performance results, it is shown that even though cross-tier interference is remarkably attenuated, co-tier interference between FAPs significantly increases as the number of FAPs increases. Since then, in [22]-[24], some dynamic channel assignment (DCA) schemes have proposed using efficient heuristic algorithms for in-building dense femtocell networks (DFNs) because the channel assignment considering co-tier interference is a non-linear non-convex NP-Hard problem [25]-[26]. In [26] and [23], authors proposed DCA schemes using graph coloring algorithm (GCA). Each FAP is first included in one FAP cluster in both DCA schemes and subchannels are dynamically assigned to FAP clusters according to the order of maximum capacity of FAP clusters in [26], while using mathematical optimization techniques in [23]. However, even though FAPs have relatively good signal to interference plus noise ratios (SINRs), FAPs use subchannels assigned for one FAP cluster thus the system capacity is limited. On the other hand, in [24], authors proposed a multiple clustering based DCA scheme called graph-based dynamic frequency reuse (GBDFR). In the GB-DFR scheme, each FAP is first included in one FAP cluster by GCA and the same number of subchannels are assigned to FAP clusters. Then, in order to use more subchannels, FAPs find other FAP clusters in which no interfering FAPs are included. However, in the GB-DFR scheme, FAPs are members of as many FAP clusters as possible and it causes that co-tier interference between FAPs to increase significantly. As a result, some FAPs have better performance by using subchannels from more than one FAP cluster while others have worse performance with no additional subchannels and reduced SINRs of femtocell user equipments (FUEs).

In this paper, we propose a novel channel assignment scheme called multi-cluster based dynamic channel assignment (MC-DCA) to improve system performance for the downlink (DL) of DFNs based on orthogonal frequency-division multiple access (OFDMA) and frequency division duplexing (FDD). In order to dynamically assign channels for FAPs, the MC-DCA scheme uses a heuristic method that consists of two steps: one is a multiple KSII TRANSACTIONS ON INTERNET AND INFORMATION SYSTEMS VOL. 10, NO. 4, April 2016 1537 cluster assignment step to group FAPs using GCA with some extensions, while the other is a dynamic subchannel assignment step to allocate subchannels for maximizing the system capacity. Through simulations, we first find optimum parameters of the multiple FAP clustering to maximize the system capacity and then evaluate system performance in terms of the mean FAP capacity, unsatisfied FUE probability, and mean FAP power consumption for data transmission based on a given FUE traffic load. As a result, the MC-DCA scheme outperforms other schemes in two different DFN environments for commercial and office buildings.

The rest of this paper is organized as follows. Section II introduces the system model and problem formulation while Section III describes the proposed MCDCA scheme. Then, simulation results are presented and discussed in Section IV. Finally, Section V concludes this paper with future research direction.

 

2. System model

2.1 System topology and channel management

We consider a typical two-tier femtocell network architecture where femtocells are overlaped on the macrocell to analyze the system performance of DL DFNs based on OFDMA-FDD. Fig. 1 shows the system topology and channel management for MBSs and FAPs. There are M hexagonal macrocells and a set of MBSs, M = {1, 2, ..., M} (M = |M|), is installed at the center of each macrocell. We assume that M = 7 and the target macrocell is surrounded by six neighbor macrocells as shown in Fig. 1-(a). Further, an F-floor building is located in the center macrocell and a set of FAPs, V = {1, 2, ..., N} (N = |V|) composes DFNs in the building. Let dIS and dMB denote the inter-site distance between the center MBS (i.e., MBS 1) and the surrounding MBSs, and between the center MBS and the building with DFNs, respectively. Fig. 1-(b) shows an example of DFN topologies in which FAPs are uniformly deployed on each floor of the building and each FAP has not only the co-tier interference coming from the neighbor FAPs in the same floor, but also the co-tier interference coming from the floors above and below. A femtocell gateway (FGW) connected to the DFN controls all FAPs which support no handover request from FUEs (i.e., a centralized management system) while each FAP serves one FUE at a random location in the coverage of the serving FAP with the maximum radius, , in meters. In addition, the MBS uses a three-sectored antenna thus the macrocell coverage is divided into three cell sites, site 1, 2, and 3, while the FAP uses an omni-directional antenna. Therefore, MBSs divide total subchannels into three subchannel groups, g1, g2, and g3, to assign for macrocell user equipments (MUEs) in site 1, 2, and 3, respectively, as shown in Fig. 1-(c). On the other hand, in order to mitigate cross-tier interference from MBSs, the FGW assigns pairs of two subchannel groups, g2∪g3, g1∪g3, and g1∪g2, which are not used by MBSs but by FAPs in sites 1, 2, and 3, respectively. It is assumed that a set of subchannels, K= {1, 2, ..., K} (K = |K|), is assigned for FAPs in each site and all FAPs are with open access mode, in which every mobile device in the building can connect to FAPs for mobile services, to focus on analyzing the system performance of DFNs. Finally, we assume perfect knowledge of channel gains, which can be calculated using the propagation losses and shadowing statistics (but ignoring the short-term fading effects).

Fig. 1.The system topology and channel assignment: (a) the macrocell topology and location of a building with DFNs, (b) an example of DFN topologies with co-tier interference, (c) the channel assignment for MBSs and FAPs.

2.2 Propagation and SINR models

In order to calculate the SINR between the FUE and its serving FAP, we use the ITU indoor path loss model and the COST-231 Hata model (urban area) for indoor and outdoor propagation models, respectively [27][28]. Let and denote path losses of the FUE served by FAP n (n ∈ V) from FAP i (i ∈ V) and MBS m (m ∈ M) in dB, respectively. and can be expressed as

where fc is the carrier frequency in MHz, while din and dmn are distances from the FUE of FAP n to FAP i and to MBS m in meters, respectively. In (1), α and Lf(δ) are the path loss exponent and the floor loss penetration factor with the number of floors, (1 ≤ δ ≤ F), between the transmitter and receiver [27]. Lf(δ) can be expressed as

Further, in (2), ht and hr denote the antenna heights of MBSs and FUEs in meters, respectively, while Lr(hr) = (1.1log10(fc) - 0.7)hr - (1.56log10(fc) - 0.8) and Lw are the antenna height correction factor of receivers and attenuation loss of an outdoor wall in dB, respectively.

Through (1) and (2), the SINR of the FUE served by FAP n at subchannel k, γnk, can be expressed as

where and denote the transmission power of each subchannel for the FAP and MBS, respectively. Further, ωnk is an indicator variable in a binary subchannel assignment matrix, Ω = [ωnk]N×K (∀n∈V, ∀k∈K), ωnk = 1 if subchannel k is allocated for FAP n, and 0 otherwise. In addition, and pg are the white noise power and the azimuth antenna pattern between MBSs and FUEs in dB, respectively. pg can be expressed as

where βg and βmax are the maximum antenna gain and maximum attenuation in dB, respectively, while θ and θ3dB = 70° are the azimuth antenna pattern of MBSs and 3dB beamwidth, respectively [29].

Given a specific γnk in (4), the spectral efficiency for the FUE of FAP n at subchannel k, rnk, is obtained by

where SE(x) = lig2(1 + ηx) in bps/Hz and η = -1.5/ln(5Pe) with the target bit error rate Pe [30]. Further, γmin and γmax are the minimum and maximum SINRs in dB, respectively, while rmin = SE(γmin) = and rmax = SE(γmax) are the minimum and maximum spectral efficiencies in bps/Hz, respectively [29,31].

2.3 FAP capacity, unsatisfied FUE probability, and power consumption for data transmission

Through (6), the capacity of FAP n, Cn, can be expressed as

where W is the bandwidth of a subchannel in Hz. Further, let the unsatisfied FUE probability, Pus, to be the probability that FUEs have capacities less than a given FUE traffic load, ρ, in bps can be expressed as

In addition, the power consumption of FAP n, En, for data transmission in mW can be expressed as

 

3. Multi-cluster based dynamic channel assignment scheme

In this section, we propose the MC-DCA scheme using a heuristic method that consists of two steps, one is a multiple cluster assignment step to group FAPs using GCA with some extensions while the other is a dynamic subchannel assignment step to allocate subchannels.

3.1 Step 1: multiple cluster assignment

Fig. 2 shows an example of the multiple cluster assignment step which has two stages, the FGW first groups FAPs using GCA in stage 1 while adds FAPs to other FAP clusters considering the transmission rate and co-tier interference of FAPs in stage 2. Fig. 2-(a) is an example of DFN topologies in which nine FAPs (N = 9) are deployed in a three-floor building (F = 3) and some FAPs have co-tier interference with each other. Under the given DFN topology, in stage 1, the FGW first generates a matrix of ones, B = [bin]N×N (∀i,n ∈ V), to obtain a binary interference matrix, J = [jin]N×N (∀i,n ∈ V). Let Γn and Γth be the SINR of the FUE of FAP n calculated by the FGW using received signal strength indicator (RSSI) measurements from FUEs and a given target threshold of the SINR for FUEs (γmin ≤ Γth ≤ γmax) in dB, respectively. Γn can be obtained by

Fig. 2.An example of the multiple FAP clustering step in the MC-DCA scheme: (a) an example of DFN topologies in an F-floor building (F=3, N=9), (b) stage 1: an interference graph using GCA (Y=4), (c) stage 2: multiple FAP clustering based on the interference graph.

Then, the FGW finds an FAP, , which gives the strongest co-tier interference to the FUE of FAP n and sets to avoid co-tier interference if Γn < Γth until Γn ≥ Γth. can be obtained by

After generating B from (10) and (11), the FGW transforms B into J = ~ B in which “~” denotes a symbol to convert all elements in B from 1’s to 0’s and vice versa. Then, an interference graph G = (V, E) can be constructed by the FGW using GCA. For the interference graph, V is used for the vertex set while E is the edge set to denote co-tier interference between FAPs in J. Further, no two connected vertices in E have the same color, that is, the color means the FAP cluster and interfering FAPs do not become members of the same FAP cluster. For the GCA, we use DSATUR (Degree of Saturation) algorithm in which a predetermined order based on the number of different colors adjacent to the vertex, called the saturation degree of a vertex, is used to color the vertices [32]. Finally, from the interference graph, the FGW obtains a minimum number of colors, Y = |y|, y= {1, 2, ..., Y }, and generates a binary FAP cluster matrix, Z = [zny]N×Y (∀n ∈ V, ∀y ∈ y). zny can be obtained by

In addition, in order to add FAPs to other FAP clusters in stage 2, the FGW finds available FAP clusters for FUEs and generates a binary available FAP cluster matrix, A = [any]N×Y (∀n ∈ V, ∀y ∈ y), considering co-tier interference based on J. any can be obtained by

In Fig. 2-(b), for example, the FGW generates an interference graph with four different colors (Y =4) using GCA, and FAPs which have the same colors become members of the same FAP clusters. Then, the FGW finds available FAP clusters for each FAP considering co-tier interference in J, that is, FAP 1, 2, and 3 become members of FAP cluster 4, FAP 4 and 6 become members of FAP cluster 2, and so on.

Using Z and A, in stage 2, the FGW adds FAPs to other FAP clusters and FAPs that are members of additional FAP clusters use more subchannels. The FGW first finds an FAP, n*, which has available clusters in A with minimum SE(Γn) (∀n ∈ V), to give higher priority, to be added into additional FAP clusters. n* can be obtained by

Also, the FGW finds an available cluster, y*, which offers maximum SE(Γn*) for FAP n* in A. y* can be obtained by

In Fig. 2-(c), for example, the FGW adds some FAPs which have available FAP clusters and higher priority, to other FAP clusters. It is assumed that the order of priority to add FAP clusters obtained by (15) is from (i) to (ix). Therefore, FAP 2 is first added to FAP cluster 4 while FAP 1 and 3 have no chance to be added to FAP cluster 4 because of the co-tier interference with FAP 2. Then, both FAP 4 and 6 are added to FAP cluster 2, while FAP 5 has no available FAP clusters. Furthermore, FAP 7 is added to FAP cluster 3, while FAP 8 has no chance. Finally, FAP 9 is added to both FAP cluster 1 and 3. The procedure of the multiple FAP clustering step is described in Algorithm 1. In order to decide the addition of FAPs to other FAP clusters, the FGW first calculates the total spectral efficiency of FAPs in cluster y*, R1, in line 16 and then once again computes it, R2, after adding FAP n* in line 18. As a result, if R1 < R2 and Γn ≥ γmin (∀n ∈ y*), the FGW adds FAP n* to FAP cluster y* as well as setting aiy* = 0 to avoid co-tier interference in line 20. Otherwise, it does not add FAP n* by setting zn*y* = 0 in line 22.

3.2 Step 2: dynamic subchannel assignment

In step 2, the FGW dynamically assigns subchannels in K to FAPs based on ρ and Z obtained by step 1. In order to guarantee the minimum number of subchannels for each FAP, Kmin = |Kmin,y| (∀y ∈ y), the FGW first divides K into two subchannel groups named static subchannel group and dynamic subchannel group, KSG and KDG, as shown in Fig. 3. KSG and KDG can be obtained by

In Fig. 3, for example, the FGW has four FAP clusters (Y = 4) thus |KSG|= YKmin = 4Kmin. After assigning subchannels in KSG to FAPs, some FAPs included in one FAP cluster have Kmin subchannels while others in multiple FAP clusters have more than Kmin subchannels. Then, the FGW finds a cluster, y**, with maximum spectral efficiency of FAP clusters, to maximize the mean FAP capacity and dynamically assigns subchannel k (∀k ∈ KDG) to FAPs. y** can be obtained by

Fig. 3.An example of the dynamic subchannel assignment step with four FAP clusters (Y = 4) in the MC-DCA scheme.

Using (17) and (18), the FGW assigns subchannel k to FAP n by setting ωnk = 1 in Ω (∀n ∈ V, ∀k ∈ K). The procedure of dynamic subchannel assignment step is described in Algorithm 2. Meanwhile, some FAPs become satisfied with ρ before all subchannels in KDG are assigned. Therefore, in order to efficiently assign subchannels, the FGW dynamically sets Cn = ρ and zny = 0 (∀n ∈ V, ∀y ∈ y) in line 7 and 16 if Cn ≥ ρ.

 

4. Simulation results and discussions

In this section, we investigate the performance of the MC-DCA scheme in terms of the mean FAP capacity, unsatisfied FUE probability, and mean FAP power consumption for data transmission using a Monte Carlo simulation. In order to demonstrate superiority, we compare the MC-DCA scheme to four different schemes: dynamic clustering based subband allocation (DCSA) [22], GB-DFR [24], graph based static channel assignment (GB-SCA), and frequency reuse 1 (FR 1). In the GB-SCA scheme, the FGW first groups FAPs using GCA (i.e., stage 1 in subsection 3.1) and assigns subchannels for each FAP cluster, while in the FR 1 scheme every FAP uses all subchannels in K without considering co-tier interference. The system topology and channel assignment for MBSs and FAPs are as shown in Fig. 1. Further, it is assumed that the building with DFNs has five floors (F=5), thus, for example, 20 FAPs are randomly deployed on each floor when N=100. Log-normal shadow fading is considered with zero mean and standard deviation of 4dB and 10dB for macrocell and femtocell networks, respectively [28]. The system parameters are listed in Table 1.

Table 1.System parameters.

In addition, we consider two in-building DFN environments using α =22 and 30 for commercial and office buildings, respectively, since the system performance is greatly influenced by the indoor environments [23]. That is, the commercial building has more open space inside compared to that of the office building, thus FAPs have more serious co-tier interference in commercial buildings.

4.1 Commercial buildings

Fig. 4 describes the results of the mean FAP capacity in commercial buildings as Γth increases when N = 100 and ρ = 1Mbps. The MC-DCA, DCSA, GB-DFR, and GB-SCA schemes show convex graphs because the SINR of FUEs increases but the number of subchannels per FAP cluster decreases (since Y increases) as Γth increases. In other words, FAPs use more subchannels with lower SINRs of FUEs when Γth is low, and use less subchannels with higher SINRs of FUEs when Γth is high. Therefore, it is shown that the optimum values of Γth with maximum mean FAP capacities are 10, 0, 4, and -4dB for the MC-DCA, DCSA, GB-DFR, and GB-SCA schemes, respectively. As a result, based on the optimum values of Γth, the MC-DCA scheme outperforms others and is 13, 20, 50, and 147% better than the DCSA, GB-DFR, GB-SCA, and FR 1 schemes, respectively. Further, even though FAPs have one FAP cluster, the DCSA scheme has better performance than the GB-DFR scheme. This is because the DCSA scheme has lower interference between FAPs and dynamically assigns subchannels to FAP clusters according to the order of maximum capacity of FAP clusters. The GB-DFR scheme shows higher performance than the DCSA scheme using more subchannels with increased SINRs of FUEs when Γth≥ 3dB but the maximum mean FAP capacity is still lower. Then, the GB-DFR scheme has higher performance than the GB-SCA scheme since it assigns FAPs to multiple FAP clusters to use more subchannels. Finally, the FR 1 scheme has the worst performance with strong co-tier interference and is not affected by Γth thus the result is always the same at 0.32Mbps.

Fig. 4.Mean FUE capacity vs. Γth in commercial buildings.

Fig. 5 depicts the results of unsatisfied FUE probability in commercial buildings as Γth increases when N = 100 and ρ = 1Mbps. The MC-DCA, DCSA, and GBDFR schemes show concave graphs while the GB-SCA scheme becomes 1 when Γth≥ −4dB since all FAPs have an insufficient number of subchannels for each FAP cluster. The FR 1 scheme performs better than the GB-SCA scheme. Based on the optimum values of Γth obtained in Fig. 4, it is shown that the unsatisfied FUE probability of the MC-DCA scheme is 25, 33, 52, and 40% lower than the DCSA, GB-DFR, GB-SCA, and FR 1 schemes, respectively.

Fig. 5.Unsatisfied FUE probability vs. Γth in commercial buildings.

Fig. 6 shows the results of mean FAP power consumption for data transmission in commercial buildings as Γth increases when N = 100 and ρ = 1Mbps. The FR 1 scheme has approximately 0.88mW and is much higher than others while the MC-DCA, DCSA, GB-DFR, and GB-SCA schemes reduce as Γth increases. The MC-DCA and GB-DFR schemes assign FAPs to multiple FAP clusters to use more subchannels thus show higher power consumption than the DCSA and GBSCA schemes. However, based on the optimum values of Γth obtained in Fig. 4, the MC-DCA scheme outperforms others and reduces the power consumption by about 15, 25, 9, and 94% compared to the DCSA, GB-DFR, GB-SCA, and FR 1 schemes, respectively. Meanwhile, the GB-DFR scheme adds FAPs to available FAP clusters and thus shows a higher power consumption compared to the MCDCA scheme but the MC-DCA scheme becomes higher when Γth≥12dB. This is because the MC-DCA scheme assigns subchannels according to the order of maximum spectral efficiency of FAP clusters, thus more FAPs with higher SINRs of FUEs use subchannels while the GB-DFR scheme assigns the same number of subchannels per FAP cluster, thus FAPs with higher SINRs of FUEs remain subchannels in multiple FAP clusters but with lower SINRs of FUEs still need more subchannels.

Fig. 6.Mean FAP power consumption for data transmission vs. Γth in commercial buildings.

Fig. 7 describes the results of the mean FAP capacity in commercial buildings as N increases when ρ = 1 and 1.5Mbps (solid and dotted lines). We first found the optimum values of Γth according to different N and ρ as shown in Table 2 and then used them for performance evaluation. The MC-DCA, DCSA, GB-DFR, and GB-SCA schemes have almost the same performance when N=20 because of low co-tier interference. On the other hand, the MC-DCA scheme shows better performance than others in both ρ = 1 and 1.5Mbps when N > 20 and the gap of capacities between the MC-DCA and other schemes is increasingly bigger as N increases. As a result, FAPs are greatly influenced by co-tier interference from neighbor FAPs and the performance decreases significantly as N increases in commercial buildings.

Fig. 7.Mean FAP capacity vs. the number of FAPs in commercial buildings: ρ =1Mbps (solid line) and 1.5Mbps (dotted line).

4.2 Office buildings

Fig. 8 describes the results of the mean FAP capacity in office buildings as Γth increases when N = 100 and ρ = 1Mbps. The DCSA, GB-DFR, and GB-SCA schemes show convex graphs while the MC-DCA scheme does not decrease when Γth> 6dB. This is because the MC-DCA scheme uses more subchannels without strong co-tier interference in office buildings. Therefore, it is shown that the optimum values of Γth with maximum mean FAP capacities are 14, 4, 6, and 2dB for the MC-DCA, DCSA, GB-DFR, and GB-SCA schemes, respectively. As a result, based on the optimum values of Γth, the MC-DCA scheme outperforms others and is 0.2, 5, 4, and 90% better than the DCSA, GB-DFR, GB-SCA, and FR 1 schemes, respectively. Meanwhile, the FR 1 scheme consistently shows the same capacity at 0.52Mbps.

Fig. 8.Mean FUE capacity vs. Γth in office buildings.

Fig. 9 depicts the results of unsatisfied FUE probability in office buildings as Γth increases when N = 100 and ρ = 1Mbps. The DCSA, GB-DFR, and GBSCA schemes show concave graphs while the GB-SCA scheme becomes 1 when Γth≥ 10dB. Further, the MC-DCA scheme shows similar results when Γth ≥ 14dB while the FR 1 scheme continuously has the same Pus at approximately 0.61. As a result, based on the optimum values of Γth obtained in Fig. 8, the unsatisfied FUE probability of the MC-DCA scheme is 49, 84, 82, and 90% lower than the DCSA, GB-DFR, GB-SCA, and FR 1 schemes, respectively.

Fig. 9.Unsatisfied FUE probability vs. Γth in office buildings.

Fig. 10 shows the results of mean FAP power consumption for data transmission in office buildings as Γth increases when N = 100 and ρ = 1Mbps. The FR 1 scheme has approximately 0.74mW and is much higher than others while the MC-DCA, DCSA, GB-DFR, and GB-SCA schemes reduce as Γth increases. Based on the optimum values of Γth obtained in Fig. 8, the MC-DCA scheme reduces the power consumption by about 29, 40, 31, and 87% compared to the DCSA, GB-DFR, GB-SCA, and FR 1 schemes, respectively.

Fig. 10.Mean FAP power consumption for data transmission vs. Γth in office buildings.

Fig. 11 describes the results of the mean FAP capacity in office buildings as N increases when ρ = 1 and 1.5Mbps (solid and dotted lines). We use the optimum values of Γth as shown in Table 2 for performance evaluation. The MC-DCA, DCSA, GB-DFR, and GB-SCA schemes have almost the same performance when ρ = 1Mbps while the MC-DCA scheme outperforms others when ρ = 1.5Mbps. As a result, in office buildings, FAPs have less co-tier interference compared to commercial buildings thus the mean FAP capacity of the MC-DCA, DCSA, GBDFR, and GB-SCA schemes is close to 1Mbps until N ≤ 80 when ρ = 1Mbps, while is reduced from 40 ≤ N when ρ = 1.5Mbps. Meanwhile, the FR 1 scheme is much lower than other schemes and reduces from 20 ≤ N.

Fig. 11.Mean FAP capacity vs. the number of FAPs in office buildings: ρ =1Mbps (solid line) and 1.5Mbps (dotted line).

 

5. Conclusions

In this paper, we proposed a novel dynamic channel assignment scheme called MC-DCA to improve system performance for DL DFNs based on OFDMA and investigated the MC-DCA scheme compared to the DCSA, GB-DFR, GB-SCA, and FR 1 schemes. Further, we considered two different DFN environments for commercial and office buildings in which FAPs have different co-tier interference effects with each other. Through simulations, we first found the optimum values of Γth to maximize the system capacity and then evaluated system performance in terms of the mean FAP capacity, unsatisfied FUE probability, and mean FAP power consumption for data transmission according to different parameters, N and ρ. Simulation results showed that the MC-DCA scheme has better performance for not only the mean FAP capacity and unsatisfied FUE probability but also the FAP power consumption for data transmission. For future work, we are planning to study a multiple cluster based DCA scheme with adaptive power control for data transmission to improve system performance of DFNs.