WLC25-2: Channel Allocation Algorithms for Three-tier ... - IEEE Xplore

2 downloads 0 Views 290KB Size Report
University of Alabama. Email: [email protected]. Xihui Zhang. Dept. of Management Information Systems,. University of Memphis. Email: xzhang2@memphis.
Channel Allocation Algorithms for Three-tier Wireless Local Loops Yang Xiao

Xihui Zhang

Dept. of Computer Science, University of Alabama Email: [email protected]

Dept. of Management Information Systems, University of Memphis Email: [email protected]

Xiaojiang Du

Jingyuan Zhang

Dept. of Computer Science, North Dakota State University Email: [email protected]

Dept. of Computer Science, University of Alabama Email: [email protected]

Abstract— A three-tier wireless local loop (WLL) extends the single-tier or two-tier WLL, and it is capable of accommodating more subscribers. This paper presents and analyzes channel allocation algorithms for a three-tier WLL. These algorithms include no repacking (NR), always repacking (AR), Repacking on Demand – Random (RoDR), Repacking on Demand – Least Load (RoDL), and Repacking on Demand – Subscriber Terminal (RoDST), depending on how repacking candidates are handled. Blocking probability and handoff probability are compared among all these channel assignment algorithms by simulations. It is shown that given the same set of simulation parameters, NR has the highest blocking probability, AR has the lowest blocking probability, and RoD has a blocking probability that falls in between. Compared with NR, both AR and RoD reduce the block probability at cost of a high handoff probability. Among RoDR, RoDL, and RoDST, RoDST has the lowest blocking probability but the highest handoff probability. Keywords-component; Channel allocation, Channel Repacking, Wireless Local Loop

I. INTRODUCTION In deployed infrastructure telecommunication networks, a wireless local loop (WLL) provides an important alternative to overcome the “last mile” problem for telephony services [2]. It uses radio signals to connect customer premise equipments (CPE) to the public switched telephone network (PSTN). In this way, a telephony service provider can deliver telephony services without relying on other providers’ existing proprietary wireline local loops. The WLL divides its service area into cells, each of which is served by a base station (BS). Many subscribers can locate within a cell. Each subscriber maintains its own telephony equipments, i.e., CPE, and each CPE connects to a co-located subscriber terminal (ST). An ST connects to a BS through wireless connections. A BS is connected to and is controlled by a base station controller (BSC) which, in turn, is connected to the PSTN. Cells within a WLL can be organized very differently. In a single-tier WLL, cells are laid out without fully overlapping with each other, and each ST can be reached by only one BS. In contrast, a two-tier WLL overlays a macrocell (cell with high power BS thus large radio coverage) with several

microcells (cells with low power BS thus small radio coverage). An ST in a microcell can be served either by the BS of the microcell or by the BS of the macrocell. A three-tier WLL takes one step further than a two-tier WLL by overlaying each microcell with several picocells. In this configuration, an ST in a picocell can be reached by the BS of the picocell, or the corresponding microcell or macrocell. Thus, when all the channels in the picocell are busy, future incoming or outgoing calls can be served by the corresponding microcell or macrocell. “Overflow” and “repacking” are two new operations for the multi-tier WLL [2]. In a two-tier WLL, for instance, “overflow” is the call handoff from the microcell to the macrocell, and “repacking” is the call handoff from the macrocell to the microcell. Repacking can only be exercised when there is at least one repacking candidate available. A repacking candidate is a call that has to meet the following two criteria: 1) the call occupies a macrocell channel, and 2) the microcell of this call has an idle channel. Channel allocation algorithms proposed for two-tier WLL can be either with or without repacking. In “no repacking” (NR) scheme described in [5] and re-described in [2], when a call arrives for an ST, WLL first allocates a channel in the microcell of the ST; if no channel is available in that microcell, the call overflows to the macrocell; if the macrocell has no idle channel, the call is blocked. Two repacking algorithms, always repacking (AR) and repacking on demand (RoD), are proposed [2] to mitigate the call blocking effect. In AR [6, 1, 4], the WLL always performs the repacking as soon as a channel is freed up in the microcell and this microcell is using a channel in the macrocell. AR keeps maximum number of idle channels in the macrocell at the cost of high handoff rate [3]. Unlike AR, RoD does not immediately perform repacking when a call in a microcell is completed. Instead, repacking is performed only when it is necessary. According to [2], there are several variants to handle the repacking candidates in RoD: 1) In BSC-based RoDR, the BSC randomly selects a repacking candidate for handoff; 2) In BSC-based RoDL, the BSC selects the repacking candidate whose microcell has the least traffic load; 3) In ST-based RoD (referred to as RoDST), repacking candidate selection and handoff decision are made by the STs upon a repacking request from the macrocell.

1-4244-0357-X/06/$20.00 ©2006 IEEE

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2006 proceedings.

In this paper, we study three-tier channel allocation algorithms. The channel allocation algorithms for three-tier WLLs are comparable to those schemes proposed for two-tier WLLs, with an extra tier of “overflow” and “repacking”, shown in Fig. 1. In this paper, we first apply a decision-tree approach to analyze channel allocation algorithms for three-tier WLLs. To evaluate performance of different schemes for threetier WLLs, we then build a simulation model based on our analysis. Call blocking performance and handoff performance are compared by simulation among NR, AR, RoDL, and RoDST, and RoDST.

Fig. 1: Overflow and Repacking in Three-tier WLLs

cannot locate an idle channel in the macrocell, the call is blocked, as shown in Fig.2. In AR, the WLL always moves a call in the macrocell to the corresponding microcell as soon as a channel is freed up at that microcell. Similarly, the WLL always moves a call in the microcell to the corresponding picocell as soon as a channel is freed up at that picocell. In this way, AR keeps maximum number of idle channels in the macrocell and microcells at the cost of high handoff rate. An interesting finding is that the decision-tree for AR scheme ends up the same with that for NR scheme. Recall that AR keeps maximum number of idle channels in the macrocell and microcells at the cost of high handoff rate, since when an overflowed call cannot find an idle channel at the corresponding microcell or macrocell, repacking won’t succeed because if there were repacking candidates available at the microcell or macrocell at that very moment, WLL itself shall have already performed the repacking. However, the performance of AR is different from that of NR. Intuitively, compared with WLL with a NR scheme, WLL with an AR scheme is expected to have a lower blocking rate at a cost of high handoffs. Call from/to picocell

Yes

Picocell has idle channel

No

Overflow to microcell

Yes

Microcell has idle channel

No

Overflow to macrocell

Yes

Macrocell has idle channel

Yes

Macrocell repacking

No

Macrocell has repacking candidates

Yes

Corresponding microcell has repacking candidates

Fig. 2: No-Repacking Scheme for Three-tier WLLs

II.

No

DECISION-TREE APPROACH

Microcell repacking

As noted in the Introduction, a three-tier WLL has an extra tier of “overflow” and “repacking” than a two-tier WLL does. In this section, we use a decision-tree approach to differentiate a series of scenarios pertaining to channel allocation schemes, e.g., NR, AR, and RoD for a three-tier WLL. A decision-tree approach is appropriate because basically a WLL has to make conditional decisions when performing call repacking or overflow. Not only can the generated decision-trees help us understand these channel allocation algorithms, but also serve as the starting points for us to construct the simulation model, draw the simulation flowchart, and implement the simulation program. With the NR scheme, we only need to consider “overflow.” When a call (either incoming or outgoing) for an ST arrives, the WLL first tries to allocate a channel in the picocell of the ST. If no channel is available in that picocell, then the call overflows to the corresponding microcell. If the microcell has any idle channels available, then one of the idle channels is allocated to the call. If the microcell has no idle channel either, then the call overflows to the macrocell. Finally, if the WLL

No

Yes

Other microcells with calls in macrocell have repacking candidates

No

Microcell repacking

Microcell repacking

Call served

Call blocked

Fig. 3: Repacking-on-Demand Scheme for Three-tier WLLs

Unlike AR, RoD does not immediately perform repacking when repacking candidates become available in macrocell or microcell. Instead, repacking is performed only when it is necessary, for instance, a new call will be blocked if repacking is not performed. Channel allocation in RoD is similar to that in NR, except that when a new call n arrives, if the picocell (of call n), the corresponding microcell and macrocell are all blocked, the WLL will perform repacking with the following steps. (1) The WLL checks whether there are repacking candidates in the macrocell. If yes, the macrocell picks one of the repacking candidates, for instance, call r, and moves call r to its microcell. The reclaimed macrocell channel is used to serve call n. (2) If the macrocell fails to locate a repacking candidate, the WLL checks whether there are repacking candidates in the corresponding microcell (of call n). If yes, the microcell (of call n) will exercise repacking, and the reclaimed

1-4244-0357-X/06/$20.00 ©2006 IEEE

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2006 proceedings.

microcell channel is used to serve call n. (3) If the corresponding microcell (of call n) fails to locate a repacking candidate, the WLL checks whether there are repacking candidates among other microcells that have call(s) being served by macrocell channel(s). If yes, the system performs repacking first at that microcell, then at the macrocell, and finally allocates the reclaimed macrocell channel to serve call n. (4) Otherwise, call n is blocked, shown in Fig. 3. The NR and AR schemes are straightforward. However, RoD can be BSC-based or ST-based depending on whether BSC or ST is in control of handling repacking candidates. In BSC-based RoD, repacking candidate selection and handoff decisions are made by the BSC. Two policies, RoDR and RoDL, can be used in BSC-based RoD. In RoDR, the BSC randomly selects one of the repacking candidates for handoff. In RoDL, the BSC selects a repacking candidate whose microcell or picocell has the least traffic load. In ST-based RoD (referred to as RoDST), repacking candidate selection and handoff decisions are made by the STs (Please refer [2] for detail). Note that more than one call may be handed off from the macrocell to the microcell or from the microcell to the picocell in RoDST. In this case, one of the released macrocell or microcell channels is chosen to serve the new call. III.

SIMULATION RESULTS

40 35 30

NR AR RoDR RoDL RoDST

Pb (%)

25 20 15 10 5 0 3

6

9

12

15

18

21

macro_max (Number of Macrocell Channels) (a)

7 6

Ph (%)

5

AR RoDR RoDL RoDST

4 3 2 1 0 3

6

9

12

15

18

21

number of radio channels in a microcell micro_max = 6, the maximum number of radio channels in a picocell pico_max = 6, the number of picocells in each microcell pico_in_micro = 6, the number of microcells in the macrocell micro_in_macro = 6, the call arrival rate arrivalRate = 0.08 per second, and the call holding time holdingTime = 100 seconds, which implies a traffic intensity ρ= 8 Erlangs. Similar conclusions can be drawn for various sets of channel numbers, cell numbers, and traffic instensities, and will not be presented in this paper. In each simulation run, 100,001 successful calls are executed to ensure that simulation results are stable. A. Effects of the Macrocell Channel Number macro_max Fig. 4 (a) and (b) plot Pb and Ph as a function of macro_max (Number of Macrocell Channels), respectively. In these figures, macro_max ranges from 3 to 21, and the traffic intensity ρ= 8 Erlangs. Fig. 4 (a) shows that Pb decreases as macro_max increases for all approaches. However, increasing macro_max doesn’t significantly improve Pb. Fig. 4 (b) shows that Ph decreases for RoDL and RoDST, although not much, as macro_max increases. However, Ph for AR is a very low constant, 0.2% - 0.3%; and Ph for RoDR has ups and downs because of its random nature of channel allocation. B. Effects of the Microcell Channel Number micro_max Fig. 5 (a) and (b) plot Pb and Ph as a function of micro_max (Number of Microcell Channels). In these figures, micro_max ranges from 3 to 21, and the traffic intensity ρ= 8 Erlangs. Fig. 5 (a) shows that Pb decreases as micro_max increases for all approaches. And increasing micro_max significantly improves Pb. Fig. 5 (b) shows that Ph decreases for RoDR, RoDL, and RoDST as micro_max increases. However, Ph for AR increases slightly as micro_max increases. C. Effects of the Picrocell Channel Number pico_max Fig. 6 (a) and (b) plot Pb and Ph as a function of pico_max (Number of Picocell Channels). In these figures, pico_max ranges from 3 to 21, and the traffic intensity ρ= 8 Erlangs. Fig. 6 (a) shows that Pb decreases as pico_max increases for all approaches. And increasing pico_max dramatically improves Pb. When pico_max is greater than 12, Pb approximates to 0 for all approaches. Fig. 6 (b) shows that Ph increases and then decreases for all approaches as pico_max increases. This nontrivial phenomenon is explained as follows. When pico_max is small, more calls will overflow to microcells and macrocell, and blocking is more likely to occur in macrocell. In this case, the on-demand handoffs are performed frequently. However, when pico_max is very large, calls can usually find idle channels in picocells; thus call overflow and handoff decrease dramatically.

macro_max (Number of Macrocell Channels) (b)

Fig. 4: Effects of Number of Macrocell Channels on Pb and Ph

In this section, we compare NR, AR, RoDR, RoDL, and RoDST in terms of the blocking probability Pb and the handoff probability Ph. In our simulations, each user-defined simulation parameter takes a constant value unless it is on the horizontal axis in the figure: the maximum number of radio channels in the macrocell macro_max = 6, the maximum 1-4244-0357-X/06/$20.00 ©2006 IEEE

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2006 proceedings.

50

6 NR AR RoDR RoDL RoDST

30 20 10 0

5 Ph (%)

Pb (%)

40

AR RoDR RoDL RoDST

4 3 2 1 0

3

6

9

12 15 18 21 micro_max (Number of Microcell Channels) (a)

3

6

9

12 15 18 21 pico_max (Number of Picocell Channels) (b)

AR RoDR RoDL RoDST

3

6

Pb (%)

Ph (%)

Fig. 6: Effects of Number of Picocell Channels on Pb and Ph

7 6 5 4 3 2 1 0 9

12 15 18 21 micro_max (Number of Microcell Channels) (b)

40 35 30 25 20 15 10 5 0

NR AR RoDR RoDL RoDST

3

6

9 12 15 18 21 micro_in_macro (Number of Microcells in Macrocell) (a)

Fig. 5: Effects of Number of Microcell Channels on Pb and Ph

12 10 Ph (% )

D. Effects of the Microcell Number micro_in_macro Fig. 7 (a) and (b) plot Pb and Ph as a function of micro_in_macro (Number of Microcells in Macrocell). In these figures, micro_in_macro ranges from 3 to 21, and the traffic intensity ρ= 8 Erlangs. Fig. 7 (a) shows that Pb increases slightly as micro_in_macro increases for all approaches. Fig. 7 (b) shows that Ph increases for all approaches as micro_in_macro increases.

AR RoDR RoDL RoDST

8 6 4 2 0

Pb (%)

3

70 60 50 40 30 20 10 0

NR AR RoDR RoDL RoDST

3

6

9

12 15 18 21 pico_max (Number of Picocell Channels) (a)

6

9 12 15 18 21 micro_in_macro (Number of Microcells in Macrocell) (b) Fig 7: Effects of Number of Microcells in Macrocell on Pb and Ph

E. Effects of the Picocell Number pico_in_macro Fig. 8 (a) and (b) plot Pb and Ph as a function of pico_in_micro (Number of Picocells in Microcell). In these figures, pico_in_micro ranges from 3 to 21, and the traffic intensity ρ= 8 Erlangs. Fig. 8 (a) shows that Pb increases at a decreasing rate as pico_in_micro increases for all approaches. Fig. 8 (b) shows that Ph increases slightly for AR and RoDL as pico_in_micro increases. RoDST has a jump in Ph from pico_in_micro = 3 to pico_in_micro = 6, then its increasing rate flats out to a comparable rate to that of other approaches. The Ph for RoDR has up-and-downs with a reason similar to that for the curves in Fig. 4 (b).

1-4244-0357-X/06/$20.00 ©2006 IEEE

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2006 proceedings.

50

7 NR AR RoDR RoDL RoDST

30 20 10

6 5 Ph (%)

Pb (%)

40

1 3

9 12 15 18 21 pico_in_micro (Number of Picocells in Microcell) (a)

Ph (%)

3 2

0 6

0 4

5

6

7 8 9 10 11 12 13 14 ρ(Traffic Intensity) (b)

Fig. 9: Effects of Traffic Intensity on Pb and Ph

8 7 6 5 4 3 2 1 0

AR RoDR RoDL RoDST

3

9 12 15 18 21 pico_in_micro (Number of Picocells in Microcell) (b)

6

G. Comparison of NR, AR, RoDR, RoDL, and RoDST In Figs. 4 (a), 5 (a), 6 (a), 7 (a), 8 (a), and 9 (a), NR has the highest Pb, AR has the lowest Pb, and the Pbs for RoDR, RoDL, and RoDST fall in between. Among RoDR, RoDL, and RoDST, RoDST has a lower Pb than RoDR and RoDL, whereas Pbs for RoDR and RoDL don’t have a fixed comparison result. In Figs. 4 (b), 5 (b), 6 (b), 7 (b), 8 (b), and 9 (b), AR has the lowest Ph, RoDST has the highest Ph, and the Phs for RoDR and RoDL fall in between. The Ph for RoDR can be lower or higher than that of RoDL. IV.

Fig. 8: Effects of Number of Picocells in Microcell on Pb and Ph

F. Effects of the Traffic Intensityρ Fig. 9 (a) and (b) plot Pb and Ph as a functions of ρ (Traffic Intensity). In these figures, ρ ranges from 4 to 14, and the average call holding time is 100 seconds. Fig. 9 (a) shows that Pb increases significantly as ρ increases for all approaches. Fig. 9 (b) shows that Ph increases and then decreases for RoDR, RoDL, and RoDST as ρ increases. The reason for this similar to that for the curves in Fig. 7 (b). The Ph for AR is flat and small.

60 NR AR RoDR RoDL RoDST

50 40 30 20

REFERENCES

[2]

[3] [4]

10 0 4

5

6 7 8 9 10 11 12 13 14 ρ(Traffic Intensity) (a)

CONCLUSIONS

In this paper, we have presented and analyzed channel allocation algorithms for a 3-tier WLL by employing a decision-tree approach. We investigated the blocking probability Pb and the handoff probability Ph of these algorithms by simulations. Our study showed that given the same user-defined parameters, NR has the highest Pb, AR has the lowest Pb, and the Pb of RoD falls in between. However, compared with NR, both AR and RoD reduce Pb at the cost of high handoff rate. Among RoDR, RoDL, and RoDST, RoDST has the lowest Pb but the highest Ph.

[1]

70

Pb (%)

AR RoDR RoDL RoDST

4

[5]

[6]

R. Beraldi, S. Marano, and C. Mastroianni, “A Reversible Hierarchical Scheme for Microcellular Systems with Overlaying Macroells,” Proc. of IEEE Infocom, 51-58, 1996. H.-N. Hung, Y.-B. Lin, B.-F. Peng, and H.-M. Tsai, “Repackng on Demand for Two-tier Wireless Local Loop,” IEEE Transactions on Wireless Communications, 3(3), 745-757, 2004. X. Lagrange, “Multitier Cell Design,” IEEE Communications Magazine, 35(8), 60-64, 1997. K. Maheshwari and A. Kumar, “Performance Analysis of Microcellization for Supporting Two Mobility Classes in Cellular Wireless Networks,” IEEE Tran. On Vehicular Tech., 49(2), 321-333, 2000. S. S. Rappaport and L.-R. Hu, “Microcellular Communication System with Hierarchical Macrocell Overlays: Traffic Performance Models and Analysis,” Proceedings of the IEEE, 82(9), 1383-1397, 1994. R. Steele, M. Nofal, and S. Eldolil, “Adaptive Algorithm for Variable Teletraffic Demand in Highway Microcells,” Electronics Letters, 26(14), 988-990, 1990

1-4244-0357-X/06/$20.00 ©2006 IEEE

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2006 proceedings.