Call Admission Control Techniques for UMTS Antonio Capone and Simone Redana DEI, Politecnico di Milano Piazza L. da Vinci 32, 20133 Milan, Italy
[email protected]
Abstract-The main advantage of UMTS radio interface is its flexibility in resource management allowed by the W-CDMA (Wideband Code Division Multiple Access) access technique. However, to exploit this flexibility it is necessary to design an efficient call admission control (CAC) scheme able to adapt the number of active calls in each cell according to interference levels and power availability. In this paper we compare the performance of two CAC styles for voice services, one based on the number of active calls and one on run-time measures of the power emitted by the base station or of the total received interference. For this last class of CAC algorithms we propose a simple scheme that allows to perform measures when silence suppression is adopted. I. INTRODUCTION With second generation cellular systems, like GSM, the decision if a new call can be accepted or not can be considered an easy task since it only depends on the available number of channels in the cell. On the other side, the planning of the number of channels and the frequencies to be assigned to each cell is not an easy job [ I ] as proved by the number of people working in the planning divisions of GSM service providers. With the new third generation cellular system, UMTS, the two problems have changed their roles. The network planning has no more to deal with the frequency allocation problem, so in a sense it is an easier task. On the contrary, the call admission control can become a very complex problem due to the soft capacity of W-CDMA systems. As a matter of fact, the system is mainly interference limited and the number of connections can not specify the actual capacity of each cell. The acceptance of a new connection depends on the SIR (signal-to-interference ratio) values achievable by each existing connection once the new one is activated. These values are functions of the emitted powers which, due to power control (PC) mechanisms, depend on the mobile user positions. Since the power available at each base station (BS) is limited, the number of users that can be served is large if the former are close to the BS and small if they are far away. The PC mechanism adopted by UMTS controls the power emitted on each channel in order to keep the SIR at the receiver at a target value. In normal conditions, an equilibrium point is reached after some algorithm iterations and all channels achieve the SIR target. The acceptance of a new call can create two possible situations: the new call is safely activated since a new equilibrium can be reached, or the new call is erroneously admitted since a new equilibrium can not be reached due to the interference levels and the power constraints. In this last case, the PC raises the power levels until some power limit is reached and some call is dropped due to the low value of
0-7803-7005-8/01/$10.00 0 2001 IEEE
925
SIR. Ideally, call admission control should be able to accept a call only if a new equilibrium of the power control can be reached and to reject it otherwise. This ideal behaviour can be obtained with a complete knowledge of the propagation conditions or allowing the new call to enter the system for a trial period. More practical schemes implemented with a distributed control must cope with a partial knowledge of the system status and may erroneously accept or reject a call. As performance metrics it is useful considering the average accepted load and the call dropping probability separately. A good CAC scheme must be able to guarantee a dropping probability below a quality threshold (we assume 0.5%) even at high offered loads and to keep the accepted load as high as possible. In this paper we compare the performance of the two CAC styles for voice services, one based on the number of active calls and one on run-time measures of the power emitted by the base station or of the total received interference. For this last class of CAC algorithms we propose a simple scheme that allows to perform measures when silence suppression is adopted. The results are obtained by simulation considering uniform and non-uniform traffic distributions. We define criteria for setting the parameters of the CAC algorithms to meet the dropping probability constraint and then compare the different methods on the basis of the accepted load. The paper is organized as follows. In Section I1 the different call admission control styles are described and their characteristics are discussed considering constant rate traffic sources and ON/OFF sources. The simulation model is presented in Section 111, while numerical results are shown and commented in Section IV. Section V concludes the paper. 11. CALLADMISSION CONTROLSCHEMES Interactive CAC algorithms are based on the idea of admitting the new call with a low power level and then evaluate if a new power control equilibrium can be reached [2], [3]. These algorithms can be considered ideal admission policies since they admit a new call if and only if the system can actually provide the required SIR level to all calls. Unfortunately, their implementation in real systems presents some drawbacks. First they present serious convergence speed problems and, above all, they can actually work only with always active connections and can not exploit discontinuous transmission, which is one of the most important issues of UMTS. A different approach to the CAC problem is based the idea that the CDMA capacity is strictly related to power limits which prevents the PC to reach a new equilibrium when the load is too high. In fact, the power levels are increased by the PC mechanism when interference increases to keep the SIR at the target value and, as a result, the level of power emitted with respect to the limit can be adopted as load indicator in the admission decision (power-based CAC) [4], [8], [9]. The decision rule can be quite simple like considering a threshold, Pthr, on
the emitted power and admitting new calls only if the powers considered are below the threshold. In the downlink direction the only power level that can be considered is that emitted by the base station even if also limits to the power used for each individual channel can be adopted. In the uplink direction the emitted power levels on each channel need to be considered. Each mobile station must inform the base station on the actual level of power so that the base station can take the admission decision on new call requests. The downlink and uplink directions can be considered independently since the W-CDMA interface of UMTS adopts a FDD (Frequency Division Duplexing) scheme in which separate bandwidths are used for the two directions. To avoid the periodic reports of the power levels from the mobile stations to the base station in the uplink direction, the total interference level at the base station can be adopted as load measure in the admission procedure (interference-based CAC) [4]. In fact., this interference level is that considered by all the PC loops operating on each channel to set the power emitted by each mobile station. Naturally, these CAC schemes can not guarantee a safe admission and call droppings can occur. They can not maximize either the admitted traffic since a call can be sometime rejected even if it could be safely admitted. Moreover, in the simple versions considered in the literature they do not solve the problem of discontinuous transmissions. Nevertheless, they are very simple schemes and their use is appealing in real systems like UMTS where complexity is already a big deal. Another even more simple possible approach to CAC is to derive an average cell capacity in number of connections so that a simple CAC based on the available number of circuits can be adopted (connections based CAC) [ 5 ] . This is equivalent to considering constant the contribute of the neighboring cells to interference and prevents an efficient management of resources with non-uniform traffic distributions. However, if the admission controller can know also the number of active connection in neighbors cells, this additional information can be used in the decision criteria to make the resource management more flexible (Looking Around CAC) [lo]. In particular it is possible to define the load of a cell as the weighted sum of the number of active call in that cell and in its neighbors. With these schemes based on the number of active connections, is quite simple to take into account the multiplexing gain due to the silence suppression of voice calls. As known the use of discontinuous transmissions results in an increased system capacity [6] that can be exploited by the Looking Around CAC just increasing the admission threshold. On the contrary, with the schemes based on the emitted power or the total interference levels the admission method must be adapted to take into account the time variations of the measured values. The general problem of the evaluation of the amount of resources needed by a set of variable rate connections is not simple and with CDMA systems it is even more involved since the network resources used by a connection do not only depends only on the bandwidth but also on the interference generated [ 7 ] . However, with these CAC schemes the assumption is that the instantaneous measured value is an indicator of the system instantaneous load and is quite natural to extend the measure over a period of time performing a statistical test on the varying level observed. This approach is similar to that of the measurement-based admission control (MBAC) schemes proposed for wired networks [ 111. An easy way to perform the test is to observe the level (power emitted or total interference) over a sliding time window and
to check whether a threshold value is exceed in any interval within the window. The window width must be tuned in order to capture the time dynamic of the process, and the admission threshold must be selected so to meet the dropping probability constraint. 111. SIMULATION MODEL
To evaluate the performance of the CAC schemes considered in the previous section we have used a C++ simulator. The simulator reproduces a system composed of 49 exagonal cells that lay on a torus surface to avoid border effects. The base stations (BS) are located at the center of each cell and irradiate with omni-directional antennas with unit gain.
A. Propagation model The propagation model follows the guidelines in [12]. In particular, the relationship between the received power Pr and the transmitted power Pt is given by Pr = P t a 2 1 0 & where L is the path loss, 10% accounts for the loss due to slow shadowing, being E a normal variate with zero mean and u2 variance, and a2 represents the gain, with an exponential distribution of unit mean, due to fast fading. The cell radius is 300 m, and the path loss L is expressed as
2
1OlogL = 128.1
+ 37.610gr(dB)
where T (in meters) represents the distance between the mobile and the base. Furthermore, we assume the shadowing standard deviation equal to 5 dB.
B. Trufic model We have adopted a traffic model with voice calls where each user generates a sigle call and calls arrive to the system according to a Poisson process of intensity A. The call length is exponentially distributed with mean 180 s and a user leaves the system as soon as the call ends. Two different traffic patterns have been considered: homogeneous and hotspot. With the homogeneous traffic pattern, when a new user is generated, its position is chosen randomly over the torus surface and is assigned to the BS with the minimum attenuation. With the hotspot traffic model a fraction of offered load arrives in the hotspot cell while the rest of load is uniformly distribuited over the other cells. The percentage of the offered load to the hotspot cell is given by 9=
(ncell
- 1). P + 100 ncell
where ncell is equal to 49 and p represent the loss of balance among the cells. No user mobility is considered. When silence suppression is adopted the traffic source status is described by a two state Markov chain [13]. The average time spent in the active state is equal 1s, while that spent in the silent state is equal to 1.35 s.
C. Receiver model At the receiving side the SIR after despreading is evaluated, for each transmission, as PT
SIR = S F . Itntra
926
+ I i n t e r + PN
(1)
r-'
22
1
20
I
I
4.850% 3.972% 0.974% 0.074%
10
I , 10
,
1
$
20
30
40
50
20 40 Offered fraffic [ErVcelll
Offered traffic [ErVcell]
Fig. 1. Power-based CAC, downlink, homogeneous traffic distribution: offered traffic vs. accepted traffic and maximum dropping probability for different values of the ratio P t h r l P m a z .
Fig. 3. Interference-based CAC, uplink, homogeneous traffic distribution: offered traffic vs. accepted traffic and maximum dropping probability for different values of the threshold level.
38 -22 20
36.512%
i i
1
P O.M4% 0.419% O.POG%
-
F
$
22
4
18
2 030%
14
0 465% 0 074%
D
5 725%
/
10 L
I 10
0
10
20 30 40 Offered traffic [ErVcell]
8
50
iL
-
1-62]
1
1
I
20 40 offered fraffic [ErUcell]
Fig. 2. Power-based CAC, downlink, effect of the maximum power value: offered traffic vs. accepted traffic and maximum dropping probability for different values of the maximum power.
Fig. 4. Looking Around CAC, downlink, homogeneous traffic distribution: offered traffic vs. accepted traffic and maximum dropping probability for different values of the threshold.
where P, is the received signal strength, PN is the thermal noise power assumed equal to -99 dBm in downlink and - 103 dBm in uplink, IinteTis the sum of the signal powers received from the other cells, IintTais the sum of the signal powers due to other transmissions within the same cell, and S F is the spreading factor assumed equal to 128. To take into account the overhead due to physical layer signaling, we assume S F = 512 during voice silent periods.
value, the maximum value is adopted. If the sum of powers required by the downlink channels exceed the base station maximum power, all the powers are proportionally reduced to limit the total power at the maximum value. After each power control iteration the actual S I R values experienced by each user are evaluated. If the S I R is lower than the call is dropped and the user a minimum value, SIR,i,, leaves the system. The SIRtar values adopted in the simulation are 6.1 dB for the uplink and 7.9 dB for the downlink, while the SIR,i, is equal to 3 dB for both directions.
D. Power control model The power control mechanism model adopted is based on the procedures defined in UMTS specifications for the dedicated channel (DCHs). The transmitted power is adjusted at each algorithm iteration to maintain the SIR at the target value, SIRt,,. In our model a power control iteration is execute every 100 ms for constant traffic sources and 10 ms for ON/OFF sources and the new power level is evaluated as:
Each channel can not exceed a transmitted power of 30 dBm in downlink and 21 dBm in uplink, whereas the overall power transmitted by a base station, Pmax,is limited to 43 dBm [ 141. If the power control requires a power higher than the maximum
927
IV. NUMERICAL RESULTS Considering first the homogeneous traffic pattern, in Figure 1 we show the accepted traffic versus the offered traffic curves and the corresponding maximum dropping probability obtained considering the power-based CAC in the downlink direction with different value of the PthT/Pmaz ratio. We observe that the CAC control mechanism is actually able to limit the maximum accepted traffic even at high offered load and that this maximum value and the dropping probability increase as the Pthr/Pmaxratio increases. To guarantee a dropping probability lower than 0.5%we need PthrlPmaz = 0.5. The idea of considering the ratio PthTlPmax instead of the absolut threshold value Pthr is based on the assumtion that it
34
30 E.
3
w
26
34
/I
I
c
2 26
-
w U
U
5 22 -
E22
0
0
t 18 -
18 4
4
10
1
ma
P n
14
1
30
14
~
''
100
200 Offered traffic[ErVcell]
300
10
400
200 Offered traffic[Erl/cell]
100
300
I 400
(4 30
2I26
t
1
30
t
/
il -
U
5 22
2%
5% 10%
0
20%
9
Bn g
-
1t
I4t 101
18
1
-
5%
.
10%
b-d
20%
l4
100 Oflered traffic [Erl/cell]
200
I 100 Offered traffic[Ell/cell]
200
(6)
(b)
Fig. 5. Power-based CAC, downlink, hotspot traffic: offered traffic vs. accepted traffic. (a) Hotspot cell, (b) other cells.
Fig. 6. Looking Around CAC, downlink, hotspot traffic: offered traffic vs. accepted traffic. (a) Hotspot cell, (b) other cells
would be easier in real system to have a simple tuning criteria based on a relative threshold. To verify if this approach is viable we have considered in Figure 2 a fixed ratio PthT/Pmaz = 0.5 and different values of Pma,.Unfortunately, we see that as the maximum power increases also the accepted traffic and the dropping probability increases. This behavior is due to the effect of the closed-loop power control which can find a new equilibrium if a solution to the power control equations exists and if that solution is compatible with the available maximum power [15]. If the maximum power never prevents the power control to reach an equilibrium point when a solution exists, we say the system is interference limited, while we say it is power limited otherwise. Therefore, if the maximum power is high the system is interference limited and, as the traffic increases, the emitted power remains well below the maximum value until the point in which the power control can no longer find a solution and the power rapidly increases. For this reason, when the system is interference limited to have an effective CAC we need a Pthr close to emitted power measured just before this point. Since this value depends on the propagation environment, we cannot define a simple general criteria to tune PthT. Similar results have been obtained for the uplink case with a threshold on the emitted power of each channel, but they are not reported for the sake of brevity. On the contrary, we report in Figure 3 the results for the interference-based CAC which
does not need any signaling between mobile and base station. Also this simple scheme is able to control the accepted load and to guarantee a low dropping probability with a proper setting of the admission threshold. The Looking Around CAC scheme is based on a different approach since only the number of connections active in the cell and in its neighbors is considered. According to what proposed in [lo], we have estimated the cell load as:
where ni is the number of active calls in cell i , ih/i is the set of the six cells adjacent to i , and w is a parameter set equal to 0.47 in our simulations. A new call request in cell i is accepted only if the load is below a threshold in cell i and in its neighbors. The results obtained with different admission threshold are shown in Figure 4. Also for this CAC algorithm we observe that it is possible to tune the acceptance threshold in order to guarantee a low dropping probability. Moreover, with the threshold values that guarantee a dropping probability below 0.5% the accepted traffic is almost the same accepted by the power-based CAC mechanism (see Figure 1). This is quite interesting even if it is evident that this good behavior has been observed in a homogeneous propagation environ-
928
26 I
I
1.125%
0.8060/.
0.433%
0.233%
0.230% 0.042%
i
1
I 0
10
20
30
40
12 10
50
Offered traffic [ErVcell]
Fig. 7. Power-based CAC with ON/OFF sources, homogeneous traffic distribution, window lenght 10 s: offered traffic vs. accepted traffic and maximum dropping probability for different values of the ratio Pthr/Pmal.
ment with ideal hexagonal cells. In real systems the weight w adopted for each neighboring cell in the cell load formula should be different and properly tuned taking into account the propagation conditions and the real shapes of the cells. In order to verify the flexibility of the admission methods with respect to traffic conditions, the acceptance threshold values able to guarantee a dropping probability below 0.5% have been taken to evaluate their performance with the hotspot traffic pattern. The results obtained for the power-based CAC and the Looking Around CAC are shown in Figures 5 and 6, respectively. With both schemes the accepted traffic in the hotspot cell is higher that than of the other cells. But with the Looking Around scheme, it remains higher even at high loads, while with the power-based scheme a fairer behavior is observed. A behavior similar to that of the power-based CAC has been observed also for the interference-based CAC in the uplink direction. Finally, we have evaluated the performance of the CAC schemes with ON/OFF voice sources. In Figure 7 we show the offered traffic versus accepted traffic curves obtained with the power-based CAC adopting the measurement scheme based on the time window and the power threshold. We see that with a window lenght of 10 s, which is able to capture the source dynamics, it is still possible to set the acceptance threshold in order to guarantee a low dropping probability. The same capability is offered by the Looking Around scheme, for which it is just necessary to increase the acceptance threshold in order to expliot the multiplexing gain (Figure 8).
V. CONCLUSIONS In this paper we have evaluated the performance of two classes of call admission control methods for W-CDMA UMTS networks. The results obtained with the power-based and interference-based algorithms show that with a proper acceptance threshold setting it is possible to keep low the dropping probability even at high loads, but the optimal threshold value depends on propagation environment. The same behaviour has been observed with voice sources with silence suppression adopting our scheme based on a measurement time window. The Looking Around scheme, which is based on the number of active connections in the considered cell and its neighbors, is also able to guarantee a low dropping probability and an accepted load almost equal to that of the power-based method.
929
1 -
I
I
io
.
I
, 20 30 40 Offered traffic [ErVCell]
50
Fig. 8. Looking Around CAC with ON/OFF sources, homogeneous traffic distribution: offered traffic vs. accepted traffic and maximum dropping probability for different values of the threshold.
Moreover, this scheme appears more flexible than the powerbased one with non-uniform traffic distributions, even if its implementation in real systems is quite complex mainly due to the difficult parameters tuning.
REFERENCES [I] A. Capone, M. Trubian, Channel assignment problem in cellular systems: a new model and Tabu Search algorithm, IEEE Trans. on Vehicular Technology, Vol. 48 (1999) 1252-1260. [2] M. Andersin, Z. Rosberg, J. Zander, Soft and Safe Admission Control in Cellular Networks, IEEE/ACM Trans. on Networking, vol. 5, no. 2, April 1997, pp. 255-265. [3] D. Kim, Efficient Interactive Admission Control in Power-Controlled Mobile Systems, IEEE Trans. on Vehicular Technology, vol. 49, no. 3, May 2000, pp. 1017-1028. [4] C.Y. Huang, R. Yates, Call Admission Control in Power Controlled CDMA Systems,Proc. of IEEE Vehicular Technology Conference, 1996, vol. 3, pp. 1665 -1669. [5] Y. Ishikawa, N. Umeda, Capacity Design and Performance of Call Admission Control in Cellular CDMA Systems, IEEE Journal on Selected Areas in Communications, vol. 15, no. 8, Oct. 1997, pp. 1627-1635. [6] A.M. Viterbi, A.J. Viterbi, Erlang Capacity of a Power Controlled CDMA System, IEEE Journal on Selected Areas in Communications, vol. 11, no. 6, Aug. 1993, pp. 892-900. [7] J.S. Evans, D. Everitt, Effective Bandwidth-Based Admission Control for Multiservice CDMA Cellular Networks, IEEE Trans. on Vehicular Technology, vol. 48, no. 1, Jan. 1999, pp.36-46. [8] J. Knutsson, P. Butovitsch, M. Persson. R. Yates, Downlink Admission Control Strategies for CDMA Systems in a Manhattan Environment, Proc. of IEEE Vehicular Technology Conference, 1998, pp. 1453-1457. [9] J. Knutsson, P. Butovitsch, M. Persson, R. Yates, Evaluation of Admission Control Algorithms for CDMA Systems in a Manhattan Environment, Proc. of CIC-97. [IO] Y. Ma, J.J. Han, K.S. Trivedi, Call Admission Control for Reducing Dropped Calls in Code Division Multiple Access (CDMA) Cellular Systems, Proc. of IEEE INFOCOM 2000, pp. 1481-1490. [ 1 I] M.Grossglauser, D. Tse, A Framework for Robust Measurement-Based Admission Control, IEEE Trans. on Networking, vol. 7, no. 3, June 1999, pp. 293-309. [12j UMTS 30.03, Annex B, Test environments and deployment models, TR 101.112, April 1998. [I31 Brady, P.T., A Model for Generating On-Off Speach Patterns in TwoWay Conversation, The Bell System Technical Journal, September 1969, pp.2445-2471. [ 141 3rd Generation Partnership Project, RF System Scenarios, TR 3G 25.942, Dic. 1999. [I51 R. D. Yates, A framework for uplink power control in cellular radio systems, IEEE Journal on Selected Areas in Communications, vol. 13, 1995, pp. 1341-1347.
