Distributed Algorithm for Cooperative Coverage ... - IEEE Xplore

4 downloads 0 Views 172KB Size Report
Queen Mary University of London. London, UK john.bigham, [email protected]. Abstract— Recent developments in mobile networks have looked.
Distributed Algorithm for Cooperative Coverage Provisioning in Mobile Cellular Networks Peng Jiang

John Bigham, Jiayi Wu

Department of Electronic Engineering Queen Mary University of London London, UK [email protected]

Department of Electronic Engineering Queen Mary University of London London, UK john.bigham, [email protected]

Abstract— Recent developments in mobile networks have looked at placing more autonomy and intelligence at base stations to encourage cooperation between them using a distributed approach in order to try and develop a network more robust to attacks and failures. A real time fully distributed algorithm that would allow base stations in cellular networks to autonomously and cooperatively find the wireless coverage pattern to optimize network performance is described. The semi-smart antenna system is assumed at each base station. Simulations based on Mobile WiMAX technology show the effectiveness and robustness of the approach. Keywords-distributed algorithm; cooperative coverage;mobile celluar network;semi-smart antenna;Mobile WiMAX

I.

INTRODUCTION

Mobile cellular networks are by far the most common of wireless communication systems. As mobile cellular communication has grown rapidly and become an integral part of our everyday life communication, mobile cellular networks need to be flexible to heterogeneous traffic demands and robust to attacks and failures. Cooperative control for geographic load balancing is recognized as a new approach for traffic load balancing that provides dynamic load re-distribution in real time according to current geographical distribution of the traffic and its type. Several optimization algorithms [1-3] to create cooperative coverage have been developed and they show significant improvements in call blocking, call dropping and system capacity. The original work on the cooperative control is centralized, and it is assumed that the processing associated with the optimization would typically be located at, e.g. an Authentication and Service Authorization (ASA) server in WiMAX network or Radio Network Controller (RNC) in WCDMA network. Based on current technology trends and economic drivers, we address the possibility of more autonomy and intelligence being placed at base stations (BSs). The BSs would sense the local environment via various forms of measurements and consequently adapt their mode of operation, in order to optimize the whole network performance. This paper describes a completely decentralized approach where the intelligence is distributed to the BSs. If more

1-4244-2424-5/08/$20.00 ©2008 IEEE

autonomy and intelligence is placed at the BS, then cooperation between the BSs without the use of a central controller can be considered as a way to making the network more robust to failures and attacks. The calculations required by the optimization algorithms can also be separated by the distributed approach to make the optimization more efficient. Our work concentrates on one particular aspect that needs to be solved in order to implement a truly distributed approach. It develops a novel technique for coloring a cellular network so that every cell of the same color is not a neighbor of each other, by using a completely distributed algorithm. Therefore, the optimization algorithms can be run in parallel in cells with the same color. While this is not current practice, the extension of software defined radio concepts to base stations and considerations of retention of aspects of such behavior in emergency contexts motivates these developments. It also has potential in some sensor networks. The remainder of this paper is organized as follows. First, we review the related work on the cooperative control and describe the bubble oscillation algorithm (BOA) in Section II. Then Section III describes the distributed coloring algorithm and proposed distrusted bubble oscillation algorithm. The numerical results are presented in Section IV. Finally, we draw conclusions in Section V. II.

COOPERATIVE CONTROL FOR LOAD BALANCING

A. Cooperative Control for Radio coverage The cooperative real time control is a control mechanism to cooperative control the physical layer (such as the radiation pattern or transmit power) according to the traffic demand changed for the geographic load balancing. It can be used to improve the performance of a wireless system containing nonuniformly distributed traffic, especially for resolving the traffic hot spots. A central feature of the mechanism is the development of algorithms to coordinate the radiation patterns of antenna that can learn from the environment during operation, so that the radiation patterns generated provide coverage where and when it is needed, while also ensuring that no gaps in coverage are created. Cooperative control for geographic load balancing has been shown to be an effective approach to improving the QoS and capacity of a cellular network and relay based cellular network. Results based on semi-smart antennas were presented in [1][3]

1565

ICCS 2008

and for cooperative tilting in [2]. The key idea behind geographic load balancing is that call blocking and call dropping in a heavily loaded cell can be mitigated by reshaping the antenna patterns to ensure adequate coverage of the sources of peak traffic and expanding adjacent antenna patterns to fill in the coverage loss. If one cell is helping another, then a further cell can help the helper. This is achieved without leaving holes in the network wide coverage. Similarly if a base station fails neighboring base stations can mitigate the loss of coverage by extending their coverage. B. Bubble oscillation algorithm with semi-smart antenna Finding the optimal coverage for the whole wireless network is a complex optimization problem. Bubble oscillation algorithm which was presented by Du and Bigham [3] shows a nearly optimal performance. In the BOA, the local coverage scheme is treated as an air bubble, the local traffic load is treated as the air within the bubble, and the un-served traffic is treated as the vacuum between adjacent bubbles. The process of geographic load balancing is performed by emulating the behavior of bubbles when disturbed. Attempts to reallocate the un-served traffic are fulfilled by a process analogous to the oscillations caused by attraction forces from temporary vacuums or the pressure difference between adjacent cells. The process is more complex than the analogy as there are different kinds of traffic, user and terrain. The process is performed cooperatively by base stations, as local base stations have limited capabilities in resolving traffic hot spots independently. The real time algorithms have been benchmarked in simulations against non real time evolutionary algorithms, transformed to reduce the number of constraints and tuned to find the global solutions. The BOA works on geographic load balancing used “semismart” antennas [4] as the vehicle to provide the flexibility in the physical coverage. For definiteness, this paper also assumes that each base station has a semi-smart antenna system to perform pattern shaping. Semi-smart antennas combine shapes to provide coverage patterns over areas. They are much cheaper to construct and do not require extensive signal processing at the elements. The capability to shape patterns over a small area, rather than pin pointing individual mobile terminals as fully smart antenna systems do, is what makes the semi-smart antenna attractive. While fully smart antennas can be effective in certain applications they have great problems in handling the changing multi-paths for large amounts of moving traffic, managing the difference in uplink and downlink frequencies, and other complexities of real topography and scale. Importantly semi-smart antennas can be retrofitted into existing networks, without the other physical components needing replacement or even modification. More information on our work on geographic load balancing and semi-smart antenna systems can be found at [5]. However, the approach described applies to a wide range of antenna systems including smart and MIMO systems. III.

DISTRIBUTED COOPERTIVE CONTROL ALGORITHM

As mentioned above the computation of the BOA for a cell is dependent on information from its neighboring cells and responds to their operations on the radio coverage. The operation of the BOA is carried out in a sequential manner,

4 1

7 3

2

6 5

Figure 1. Conversion of wireless network to a planar directed acyclic graph

with each cell performing the local optimizations until convergence. Termination is based either all the un-served traffic being served, or the change over a set of iterations being below a threshold, or the predefined maximum number of iterations is reached. Therefore every cell cannot run the algorithm simultaneously. The main idea of distributed cooperative control algorithm is to apply a novel technique for coloring a cellular network so that every cell of the same color is not a neighbor of each other, by using a completely distributed coloring algorithm. This coloring of the network then allows the cooperative algorithms like BOA to be used in a distributed fashion. A. Distributed Colouring Algorithm Distributed algorithms can be defined as a procedure designed to simultaneously process a task on multiple computers without a central controller. The aim of the distributed coloring algorithm used in this approach is simple, distributed labeling the cells in the network in such a fashion that no two adjacent cells have the same color. Based on the four colors theory, when the graph is planar the minimum number of labeling colors is four. A distributed self stabilizing coloring algorithm for planar graphs [6] is used to label each cell with a different color. The algorithm used is based on graph theory and was designed for coloring the nodes of a planar graph. The algorithm is divided into two distinct parts – the first part describes the conversion of any planar graph into a directed acyclic graph (DAG) and the second part illustrates coloring of this directed acyclic version of the planar graph. In our case the graphs can reasonably taken to be planar. The cellular network has to be converted to a directed acyclic graph before the coloring algorithm can be applied. An illustration for the conversion of the cell network into a directed acyclic graph is shown in Figure.1. The communication edges in a cellular network are computed using the reference variable arbitrarily assigned to the cell at the beginning. The cell also has a unique immutable reference ID. This referencing and calculation of the direction of the edges of a cell assigns unambiguous “initial” directions to the network and converts the network from a planar graph representation to a directed acyclic graph. The directions are termed as “initial” because some cells may still have an out-degree greater than five. Therefore, an adjustment is required in the edge directions

1566

180

160

Four Colour

Central BOA

Five Colour

Four Colour

0.5

Ten Colour

Five Colour

140

120

0.4

Colour proportions

Calcultion Time (Seconds)

Ten Colour

100

80

60

0.3

0.2

40 0.1

20

0 0

10

20

30

40

50

60

70

80

90

100

0

0--25

Snapshot number

25--50

50--75

75--100

Number of oscillations to convergence

(a) Calculation time comparison for different colours (b) The Number of oscillation for different colours Figure 2 Results for the different colours efficacy test

(by adjusting the reference values) in such a fashion that eventually no cell has an out-degree greater than five. After the adjustment, a DAG can be established so that an ordering between nodes (here cell IDs) and the coloring assignment component that uses the DAG to traverse the network. The number of colors (4 or more) is a parameter in the algorithm. In our experiments, 4, 5, and 10 colors were used. Unless the network configuration changes, e.g. by the failure of a BS, or the infilling by a new BS, then once the coloring of the cells has been performed, the distributed coloring process is no longer needed. Note that the distributed coloring algorithm is described as self-stabilizing, that is, it will always converge to the end point regardless of the initial configuration of the network. While the two components of the algorithm would seem to need to run sequentially they can in fact run simultaneously. B. Distributed Bubble Oscillation Algorithm The distributed coloring algorithm colors the cell in a distributed manner using a minimum of four colors, with a condition that two cells of the same color cannot be neighbors to each other. This makes every cell in one color set independent of each other in terms of computation of the BOA, allowing the cells in a color set simultaneous computation. This leads to faster convergence times and better adaptability to geographical changes in traffic demands. The bubble oscillation algorithm has been adapted so that it can control the antenna pattern of the base station where it is, perhaps only notionally, located. It performs its assignment of mobile stations (MSs) to the BS where it is located on the basis of a ranking established from computing the utility of sending radiation power (in the downlink case) in each direction. The utility is computed on the assumption that all the neighboring cells of the BS have their coverage pattern temporarily fixed to their current pattern. The utility of a direction (a small sector) is the sum of a prescribed function (typically, but not always, the closer the higher the utility) of the distance from the BS to any MS along that direction (a small constant if no MS) augmented by the sum of the projections a prescribed value associated with covering each unassigned MS onto the direction in question.

A sketch of the distributed BOA algorithm is as follows: Label cells with colors, order the colors, and choose one of the colors to be active. First all cells of one color are active, then all cells of another, and the algorithm cycles around all the colors. Give each BS its default (e.g. circular) pattern at each active cell do { compute utilities of each direction (as described fully in [2]) find potential assignments of MSs to the BS consistent with BS constraints compute the chosen QoS metric (e.g. blocking rate) activate cells with next color in the set of colors (cycling if at end) } until (QoS measure at BS ceases to improve over successive iterations or iteration count exceeds threshold) } IV.

SIMULATION RESULTS

System-level simulations for a Mobile WiMAX cellular network were performed to test the effectiveness and the robustness of the distributed approach. A. Simulaton Assumption To evaluate the system performance, a 10×10 hexagon model Mobile WiMAX network was used in the simulation with the following simplifying assumptions: • • • • • • •

1567

Cell radius: 1Km Frequency:3.5 GHz Channel bandwidth: 10 MHz BS height 30m, MS height 1.5m TX Power: BS 43dBm (6 sector), MS 21dBm Noise Spectral Density: -174 dBm/Hz Path loss factor: 4.0

In this simulator, the Mobile WiMAX network is based on the IEEE802.16e standard with OFDMA architecture and TDD duplex technique [7]. At the beginning of the simulations, the MSs are uniformly distributed over the network, and the demand is such that all MSs can be served. At time passes, as measured by 100 snapshots, the MS cluster into hotspots. Each traffic snapshot contains 50 000 mobile users and 20% of traffic is gathering to 10 hot spots from the first snapshot to the last one.

0.35 Conventional Decreased order Increased order Random order

0.3

Call Blocking Rate

0.25

0.2

0.15

0.1

B. Efficacy Performance With the typical character of distributed algorithm, the BOA computation can be separated and run in parallel with different color components. The performance for different colors is shown in Figure.2(a). The x-axis represents the traffic snapshots, and the y-axis represents BOA calculation time. At the beginning of the simulation, when the traffic is nearly uniformly distributed, the computation requirement of the BOA is limited and calculation times of these approaches are nearly same. When there are more MSs in hotspots, i.e. as the traffic becomes more heterogeneous, the distributed approaches achieve a significant effectiveness improvement. With the DBOA, the number of cells working in parallel is M/N where M is the number of BSs and N (N•4) is the number of colors. The worst case number of iterations before the algorithm terminates is N*K, where K is the maximum number of iterations at a BS. BSs with the same color can run at same time, and more label colors mean less components run in parallel. This can potentially decrease the speed. However, there is a trade-off between the number of colors and number of BOA iterations. More colors result in less problems at the boundary of the radio coverage between neighbors in each oscillation, as the BSs with same color are far from each other, and is seen to decrease the number of iterations necessary to converge, although this increases the waiting time of bubble algorithm and information exchange cost over the entire network. The number of oscillations for different numbers of colors is shown in Figure.2(b). C. Robustness Performance The distributed BOA is performed in the number of color set, and it is important to decide the order in which the coloring set performs the computation of the distributed BOA. Three different orders in terms of traffic load were investigated below: • Descending order: the collective load of every cell in a color set was determined and the set with the highest load performed the distributed BOA first. • Ascending order: the collective load of every cell in a color set was determined and the set with the lowest load performed the distributed BOA first. • Random order: the bubble oscillation algorithm was performed in a random order but due to test conditions care was taken to ensure that the random order didn’t match the ascending or descending order.

0.05

0 0

10

20

30

40 50 60 Traffic Snapshots

70

80

90

100

Figure 3. Results for the different orders

represents the call blacking rate. Although the decreased order has a little better performance, the results show that no matter which order the computation of the bubble oscillation algorithm is performed the result is nearly the same. This shows off the robustness and versatility of the distributed BOA and also helps in applying it in distributed fashion as any order of computation could be used without making a telling difference to the performance of the bubble oscillation. V.

CONCLUSIONS

A fully distributed real time algorithm that exploits different color labels of contiguous base stations and integrates with a previously described cooperative algorithm for optimal wireless coverage in cellular networks is presented. The experiments demonstrate the efficiency and robustness of the distributed algorithm to provide cooperative cell coverage. Mobile cellular networks are considered in this paper, but the distributed approach could easily be extended to the wireless sensor networks and other optimization algorithms. REFERENCES [1]

[2]

[3]

[4]

[5] [6] [7]

The performance of the different orders is shown in Figure 3. The x-axis represents the traffic snapshot, and the y-axis

1568

L. Du, J. Bigham, and L. Cuthbert, “Towards to intelligent geographic load balancing for mobile cellular networks” IEEE Transactions on Systems, Man, and Cybernetics – Part C: Applications and Reviews, Vol. 33, No. 4, November 2003, pp 480-491. L. Du, J. Bigham, and L. Cuthbert, “A bubble oscillation algorithm for distributed geographic load balancing in mobile networks,” in Proceedings of the Twenty-third Annual Joint Conference of the IEEE Computer and Communications, IEEE INFOCOM’2004, vol. 1, Hong Kong, March 2004, pp. 330–338. J. Wu, J. Bigham, P. Jiang, "Tilting and Beam-shaping for Traffic Load Balancing in WCDMA Network", European Conference on Wireless Technology ECWT2006, Manchester UK,September 2006. P. Nahi, C. Parini, S. Papadopoulos, L. Du, J. Bigham and L. Cuthbert, “A Semi- Smart Antenna Concept using Real-Time Synthesis for use in a Distributed Load Balancing Scheme for Cellular Networks”, IEE International Conference on Antennas and Propagation, ICAP’2003, April 2003. Network group, Electronic Engineering Department, Queen Mary University of London http://www.elec.qmul.ac.uk/networks/index.html S.Gosh, M. H. Karaata ‘A self stabilising algorithm for coloring planar graphs’ , Distributed Computing (1993) 7:55-59. “Mobile WiMAX – Part I: A Technical Overview and Performance Evaluation”, WiMAX Forum, 2006.