CogART: A Fair Cognitive Channel Allocation Method for ... - USC

5 downloads 8249 Views 247KB Size Report
schemes at moderate call traffic levels making it very relevant for unforeseen situations. ... wireless services, especially mobile telephony, has increased dramatically. However, the ... Center (MSC) from time to time. The MSC is the central.
A Fair Cognitive Channel Allocation Method for Cellular Networks ∗ Dept.

Sunav Choudhary∗ , Shaunak Mishra∗ , Nachiket Desai∗ , N. Swathi Priya∗ , Dhaval Chudasama∗ and R.V. Rajakumar∗ of Electronics and Electrical Communication Engineering, Indian Institute of Technology, Kharagpur, India Email: [email protected], [email protected], [email protected] [email protected], [email protected], [email protected]

Abstract—Rapid growth in the consumer base for mobile services against the backdrop of limited spectrum demands intelligent networks for optimum utilization of the allocated spectrum with minimal changes in the existing infrastructure. This paper introduces a novel three stage predictive channel allocation scheme called Intelligent Channel Allocation (ICA) based on long term call statistics, instantaneous call statistics and event driven decisions supported by cognitive radio techniques for an opportunistic but fair usage of spectrum. The scheme realizes a trade-off between enhanced spectrum utilization in moderate call traffic conditions and fast response to emergency situations modeled by a step input in call attempts. Experimental simulations show that the proposed scheme is superior to the existing schemes in emergency situations while providing a comparable performance with the existing hybrid channel allocation schemes at moderate call traffic levels making it very relevant for unforeseen situations. Index Terms—Radio spectrum management, Prediction methods, Cognitive radio, Intelligent networks.

I. I NTRODUCTION The radio spectrum is divided into several bands with each service having its own dedicated band specified by a regulatory authority. Over the last decade the demand for wireless services, especially mobile telephony, has increased dramatically. However, the allocated band for 2G and 3G mobile communications standards is seriously limited. In current cellular networks, the number of channels for a given cell is fixed and hence the number of users that a cell can serve is limited. The increasing users have to be accommodated within the available band. This is typically done by reducing the cell size at the cost of equipment. Below a certain cell size the handoff traffic increases, which again limits the capacity in view of the need for bandwidth reservation for handoff. Therefore, introducing dynamism in the allocation of spectrum and effectively using the same becomes inevitable. It is generally observed that spatial and temporal variations associated with call traffic in cells of a cluster lead to some cells being congested while others are idle. These variations can be exploited in order to efficiently utilize the spectrum by making a significant part of the available spectrum dynamic and using it whenever and wherever there is heavy traffic. In view of the co-channel interference and other problems, allocation issues are better addressed by using a balanced and well-designed dynamic scheme rather than by pooling the entire capacity of a cluster. Such a balanced flexibility can

978-1-4244-4583-7/09/$25.00 © 2009 IEEE

also be made to provide a capability to deal with capacity requirements in emergency situations, viz. a natural calamity, a terrorist attack, etc. which may result in a sudden rise in demand for wireless services causing serious congestion and failure. Such situations can be examined by studying the behavior for a step input in the number of call attempts in the affected cell. Therefore, in order to provide capability to cater to emergency situations, the step response of the frequency allocation mechanism is improved. This is achieved by including the necessary apparatus required to identify and cater to such a situation with response time and additional throughput forming the metrics for performance evaluation in emergency situations. The remainder of the paper is organized as follows. Section II briefly describes cognition in radio networks. Section III briefly lists the schemes for channel allocation found in literature. Section IV describes the proposed algorithm which uses cognition to adapt to sudden as well as gradual changes in cellular traffic. Section V compares simulation results for various channel allocation algorithms against the proposed algorithm. Conclusions are presented in Section VI. II. C OGNITION IN R ADIO N ETWORKS In order to ensure efficient spectrum utilization, users of several services need to be given freedom to access a free spectrum within certain limits. This is called fullest Dynamic Spectrum Access (DSA) [8]. A new class of radios which can make DSA possible are cognitive radios. Cognitive radios sense the spectrum over a wide band, detect the legacy users (primary users) and accommodate secondary users in the free bands of the spectrum without causing interference to the primary users. Cognitive radios can be used to make mobile radio networks capable of adapting themselves to spatial and temporal variations in user traffic and serve eligible secondary users, in order to increase the overall throughput. Intelligence is added at the Mobile Station (MS) and Base Station (BS). The BSs exchange the status of the traffic and details of channel occupancy with the Mobile Switching Center (MSC) from time to time. The MSC is the central controller capable of taking decisions based on the current and past traffic status as to which channel should be allocated to which cell under the specified constraints. Having allocated

138

CogART’09

the channels by MSC, the BSs tune themselves to those frequencies to serve their corresponding users. Decentralizing the control of MSCs to the BSs increases the inter BS communication traffic for all BSs as cognition requires knowledge of past traffic statistics of all neighbouring cells within the minimum frequency reuse distance. Additionally, it may result in conflict of demand for the use of available channels. Its only advantage of very low failure probability due to equipment malfunction is neutralized by the rarity of complete equipment failure. III. E XISTING C HANNEL A LLOCATION S CHEMES Several channel assignment strategies have been described in the literature [5]. Channel assignment strategies are mainly classified as Fixed Channel Assignment (FCA), Dynamic Channel Assignment (DCA) and Hybrid Channel Assignment (HCA). In FCA strategies, a nominal number of channels are permanently allocated to each cell. Several fixed channel borrowing schemes like Simple Borrowing scheme(SB) [3], [7], [10], Borrow from the richest (SBR) [2], Basic Algorithm with Reassignment (BAR) [3], Borrow First Available (BFA) [2], exist in literature. In DCA however, all channels are kept in a central pool and are assigned dynamically to the cells as new calls arrive in the system [7]. After the call is over the channel is returned to the pool. DCA schemes can be centralized or distributed. Simulations show that DCA schemes perform better under light traffic while the performance of FCA schemes is superior under heavy traffic. This is because a channel transfer leads to its use being blocked in three of the nearby cells where it could have been used earlier complying with the minimum frequency reuse distance. At lower traffic levels, the increase in the call acceptance rate in the demanding cell outweighs the increase in call drop rate in the blocked cells giving an advantage to the DCA scheme. However, above a certain threshold, the two competing factors reverse in weightage which results in better performance of FCA scheme. HCA schemes try to combine the advantages of both FCA and DCA schemes. In HCA [4], [6], the set of channels assigned to each cell is divided into two sets, namely fixed and borrowable channels. Commonly used HCA strategies are Simple Hybrid Channel Borrowing (SHCB) [4], [9], [10]. Borrowing with Channel Ordering (BCO) [9], [10] and Borrowing with Direction Channel Locking (BDCL) [9]. Despite realizing effective utilization of the allocated spectrum, the existing approaches for channel allocation lack focus on the algorithm’s performance in an emergency situation which forms a crucial component of the proposed scheme. IV. P ROPOSED I NTELLIGENT C HANNEL A LLOCATION S CHEME The proposed Intelligent Channel Allocation (ICA) Scheme is a hybrid channel allocation scheme. The system adapts itself to spatial and temporal variations in the traffic by varying the fixed to dynamic channel ratio, working on a slot-by-slot basis. All the dynamic channels are under the control of the

Fig. 1.

Block Diagram of the Predictor

central dynamic pool. In this way, the system running in the nth time slot will predict the traffic of the (n + 1)th time slot. Reassignment schemes efficiently utilize the dynamic channels. The proposed algorithm also has the capability to adapt to emergency situations characterized by a sudden rise in traffic in certain areas. The same is discussed in the following subsections. A. Predictor Block The future traffic in each cell is predicted using a two-stage predictor whose block diagram is shown in Fig. 1. It has a Long Term Predictor (LTP) and a Short Term Predictor (STP), copies of which run independently for all cells. Call statistics is collected as number of calls per time slot. The LTP estimates the traffic level at a given time slot, using the traffic data at the same time slot of the previous days. Since this data can be assumed to be stationary and with good autocorrelation, the Auto Regressive (AR) model of prediction has been used for the LTP. The AR model coefficients of an N th order predictor are calculated using the equation a = R−1 r

(1)

where a is the N ×1 vector of AR coefficients, R is the N ×N autocorrelation matrix and r is the cross-correlation vector of size N × 1. The value of the input x as predicted by an N th order AR predictor is calculated by using the equation x ˆ(t) =

N 

ai x(t − i)

(2)

i=1

where x ˆ is the predicted value of x. Thus, the error in prediction is e(t) = x(t) − x ˆ(t)

(3)

A first order AR has been chosen as the LTP. Hence, the input x fed to it is the traffic level at the same time slot of the previous day. The LTP output error values corresponding to adjacent time slots of the same day can also be assumed to be stationary and with good autocorrelation. Hence, these values are fed to the STP, which is a third order AR predictor. The STP output error is very low compared to the LTP output error. An important point to note here is that days in the past that contained emergencies are excluded from the prediction process. The estimation of the call traffic is thus made more accurate using a two-stage prediction process by using the characteristics of the input data.

139

B. Determination of Fixed and Dynamic Channels

D. Co-channel Interference Mitigation

Each cell in the system is initially allocated a group of channels which, henceforth have been referred to as primary channels belonging to that cell. The number of primary channels allotted to each cell is altered at the start of each day depending on the average traffic predicted for that cell and its neighbors at the start of the day, i.e. the LTP output x ˆ1 since the STP output is known only just before the start of the time slot. The number of primary channels npri allocated to a cell c, is given by ⎞ ⎛ Nt  x ˆ1 (c, t) ⎟ ⎜ ⎟ ⎜ t=1 ⎟ × Nchan (4) npri (c) = ⎜ ⎟ ⎜ Nc  Nt ⎠ ⎝ x ˆ1 (c, t)

With the introduction of cellular concept in mobile communications, frequency reuse factor and co-channel interference have become important issues. They need to be addressed while considering channel borrowing. In the proposed algorithm, co-channel interference mitigation is done by invoking the directional channel locking scheme as described in [10]. This considerably reduces the co-channel interference by restricting the inter-cellular movements of a channel along a single direction in order to ensure no two adjacent cells end up with the same channel. E. Adaptation to Emergency

where Nc and Nt are the total number of cells and the number of time slots in a day, respectively and Nchan is the total number of channels allotted to the service provider. At the beginning of each time slot, the output of the STP x ˆ2 is used to prepare each cell for the expected traffic. This is done by evaluating an additional channel requirement function for every cell

The predictor described in section A works very well as long as the autocorrelation is good for the LTP output. If this autocorrelation was not good, the predictor would fail. This usually happens in emergency situations wherein the call traffic over a specific area increases suddenly. The ICA algorithm has a feature that prevents failure in such a situation. Upon encountering call traffic levels that are significantly higher than those predicted by the LTP, the use of the twostage predictor is discontinued. Instead, a simpler predictor using only the previous few traffic samples for the prediction is used. An emergency is declared if the condition

x2 (c, t)/Nts  − f ree(c, t) nreq (c, t) = ˆ

Sum > T hreshold

c=1 t=1

(5)

where f ree(c, t) is the total number of free channels in the cell c at time t, Nts is the maximum number of calls that can share a channel by TDM and · denotes the ceiling function. Nts is used to translate the expected error from STP to number of additional channels required. A negative value of nreq implies capacity to donate primary channels or return borrowed channels. Transfers follow a simple strategy of transfer from ‘richest to poorest’ to ensure fairness in distribution, while satisfying constraints imposed by reuse factor and co-channel interference (section D). C. Call Allocation Strategy Primary channels are given highest priority of allocation for an incoming call. If no primary channel TDM slots are available, then secondary channels are searched for free slots. If these too are not available, then the neighboring cells are searched for free transferable channels. This three-stage call allocation process gives each incoming call attempt a maximum chance to get accepted. It is also worth noting that the prioritization of channels is done only at a logical level. At a physical level, the logical channels are randomly mapped onto their physical counterparts. This is done to achieve better cochannel as well as adjacent-channel interference management. Apart from this, a reassignment strategy [7] which involves handing off a call in progress is also used. Transfer of calls from secondary channels to free slots in primary channels is given preference. This practice is essential for efficient utilization of the dynamic channels and also to further reduce the co-channel interference.

(6)

is satisfied. In Eqn. (6), Sum =

tobs 

x(c, t − i)

(7)

i=1

and T hreshold = thf ×

tobs 

x ˆ1 (c, t − i)

(8)

i=1

where the threshold factor thf and the observation time tobs are predetermined keeping in mind the minimization of response time and prevention of false emergency alarms, x(c, t) being the input call traffic. The emergency is revoked as soon as the condition given in Eqn. (6) ceases to hold true. However, there is a chance of the system switching back and forth repeatedly if the value of Sum fluctuates slightly around the value of T hreshold. To prevent this from occurring, the value of thf used for revoking an emergency is deliberately kept slightly lower than the value of thf used to declare one, i.e. a small amount of hysteresis is introduced in thf . This is necessitated by the fact that the threshold is the boundary of a complete change in the call allocation architecture and hence, frequent transitions across this boundary in both directions would lead to system instability. The number of borrowed channels must be as large as possible to increase the throughput in the affected cell(s). This is automatically achieved by using only a simple STP-like predictor which takes only the past few traffic sample values. This raises the value of the additional channel requirement function as defined in Eqn. (5), which automatically ensures maximum

140

Fig. 2.

Diagrammatic Representation of Algorithm

additional throughput. Emergency services are accorded high priority and thus, have dedicated channels allotted to them which are different than those available for regular commercial services. Hence, calls running on secondary channels may be forcibly terminated after around 120 seconds, which is a reasonable amount of time. Such terminations would be preceded by a warning given well in advance, and would serve to increase the overall throughput. Emergency services provided by cellular networks have high priority and thus, are alloted separate channels than those available for regular services. Since these channels are not included in the set of available channels in the foregoing algorithm, they are exempted from the described forcible termination scheme. The algorithm is represented in a flow chart form in Fig. 2. V. S IMULATION R ESULTS Several simulation studies were carried out to evaluate the performance of the proposed algorithm in moderate to high call traffic levels. The distribution of call attempts with time in a particular cell for a day was taken as a double-gaussian, as described in [1]. This profile was fed into a Poisson process as the mean. The call duration was exponentially distributed with a typical mean of 120 seconds. The time slot length T used for simulation was 10 seconds and the time window, tobs over which emergency situations were tested was taken to be 200 seconds.

Fig. 3.

Emergency Response of various schemes

Fig. 3 compares the blocking probability for the proposed algorithm with other schemes, namely HCA, BCO and simple FCA as described in section III. The proposed algorithm offers a slight improvement over the others at low to medium traffic levels. However, the margin of improvement increases with increasing traffic, which is evident from the significantly lower slope of the ICA plot compared to that of the closest competing other schemes as the traffic level increases. Having established the superiority of the proposed algorithm for extremely high traffic levels, i.e. in case of unforeseen situations, we now proceed to address the other relevant performance characteristics under such a condition which

141

Fig. 4.

False Emergencies and Rise Time Variations vs Threshold Factor

Fig. 6.

Plot of standard deviation in rejection percentage vs call traffic

VI. C ONCLUSION

Fig. 5.

Blocking Probability of various methods of channel allocation

depend on a variety of factors, most important of them being the threshold factor. A plot of the number of false emergencies generated versus the threshold factor is shown in Fig. 4. An optimum threshold factor of ‘2.5’ to ‘3’ corresponding to 150 to 200 percent increase was considered in the performance evaluation of ICA under emergency, which is a trade off between small rise time and higher likelihood of detection of false emergencies. In order to obtain the response time a step rise in the call attempts was given as input to a cell and the response of ICA and other algorithms was compared. Fig. 5 compares the transient performance of different schemes under such a situation. The ICA algorithm evidently outperforms the others in terms of both the additional throughput and the rise time. The rise time is a measure of how quickly the system detects an emergency when one has occurred. The ICA scheme also ensures a fairer distribution of channels among the different cells with unequal call traffic, as compared to the other three algorithms. Fairness has been quantified by the standard deviation in the call blocking probabilities of different cells, as shown in Fig. 6, where a fairer algorithm gives a lower standard deviation. The individual cellular call traffic values used for simulation across the cells of a cluster follow a non-monotonic as well as non-constant variance across increasing average cluster call traffic. The highly irregular standard deviation behavior of FCA scheme under non-uniform traffic suggests the inability of the FCA algorithm to absorb the non-uniformity in a cluster in terms of resource allocation as compared to the other three algorithms.

The proposed ICA algorithm incorporates intelligence in the form of traffic prediction into the existing mobile radio network. This is used both for achieving maximum resource utilization under routine traffic as well as providing a metric to classify an emergency situation for quick and effective response. This is coupled with an efficient allocation scheme that gives each incoming call a maximum chance of getting accepted, while reducing the amount of computation required by working on a slot-wise basis. The proposed algorithm achieves better throughput than other allocation schemes by exploiting the non-uniformity in traffic levels of adjacent cells without compromising on the fairness in call allocation. Apart from this, the features that have been included specifically to tackle emergency situations greatly reduce the response time of the system, enhacing the reliablity of the cellular network during unforeseen situations. R EFERENCES [1] S. Almeida, J. Queijo and L.M. Correia, “Spatial and temporal traffic distribution models for gsm,” in 49th IEEE Vehicular Tech. Conf., IEEE. IEEE, 1999, pp. 131–135. [2] L. Anderson, “A simulation study of some dynamic channel assignment algorithms in high capacity mobile telecommunications system,” IEEE Trans. on Vehicular Tech., vol. VT-22, p. 210, 1973. [3] J.S. Engel and M. Peritsky, “Statistically optimum dynamic server assignment in systems with interfering servers,” IEEE Trans. on Vehicular Tech., vol. VT-22, pp. 203–209, 1973. [4] T.J. Kahwa and N. Georganas, “A hybrid channel assignment scheme in large scale cellular-structured mobile communication systems,” IEEE Trans. on Communications, vol. COM 26, pp. 432–438, 1978. [5] Katzela and M. Naghshineh, “Channel assignment schemes for cellular mobile telecommunication systems, a comprehensive survey,” IEEE Personal Communications, pp. 10–31, June 1996. [6] H. Sekiguchi, H. Ishikawa, M. Koyama and H. Sawada, “Techniques for increasing frequency spectrum utilization,” in IECE Tech. Rep., 1984, pp. CS84–100. [7] R. Singh, S.M. Elnoubi and C. Gupta, “A new frequency channel assignment algorithm in high capacity mobile communications systems,” IEEE Trans. on Vehicular Tech., vol. VT-31, 1982. [8] K. Watanabe, K. Ishibashi and R. Kohno, “Performance of cognitive radio technologies in the presence of primary radio systems,” in IEEE international symposium on PIMRC. IEEE, September 2007. [9] M. Zhang and T.S. Yum, “The non-uniform compact pattern allocation algorithm for cellular mobile systems,” IEEE Trans. on Vehicular Tech., vol. VT-40, pp. 387–391, 1991. [10] M. Zhang, “Comparisons of channel assignment strategies in cellular mobile telephone systems,” IEEE Trans. on Vehicular Tech., vol. VT38, pp. 211–215, 1989.

142