Evaluation of Load Control Strategies in an UTRA/FDD Network

Evaluation of Load Control Strategies in an UTRA/FDD Network W. Rave, T. Köhler, J. Voigt, G. Fettweis Mobile Communications Systems Chair, Dresden University of Technology, D-62 Dresden, Germany P.Schneider,M.Berg Mannesmann Mobilfunk GmbH, Am Seestern, D-4543 Düsseldorf, Germany Abstract For a regular array of hexagonal cells in a suburban outdoor environment, we examined resource allocation schemes in the radio network controller (RNC) for the up- and downlink in a UMTS network. In the uplink the noise rise is the parameter of choice. In the downlink the power for different data rates and the connection as a whole serves as the means to control the network. Allocating this resource in the admission and congestion control, the throughput of a packet service with variable data rate was optimised. Loading of UTMS cells The link equation for the CDMA uplink [], relating the carrier to interference ratio CI of user i to the processing gain PG i, to the received power of the desired signal P i, to the power of unwanted signals P j6i, and to the thermal noise power within the CDMA bandwidth P th, can be writtenintheform C I i PG i P i P th +(+f ) P j6i ν jp j PG i P i P total P i : () The average ratio between inter and intra cell interference is represented by the value f, while the user activity is incorporated in the coefficient ν j and P total is equal to the total received power at the network element. Using the noise rise together with the uplink load factor UL noise rise P total P th P total P total P user the well-known pole equation is obtained: P i + PGi (CI) req: UL ; (2) P th UL : (3) It displays the fact, that for a given spreading factor (processing gain) a critical number of users exists for which the required transmit power diverges. The minimum distance to be maintained from the critical user density or loading of the network is specified by a maximum allowed load factor UL, defining the percentage of user interference with respect to the total interference. For the downlink, the link equation has to be modified to take into account the common control channels and the intra cell orthogonality, represented by the coefficient ff: C P PG i P i P I i P th + ff ν j6i jp j + k Pl ν (4) k;lp k;l Nevertheless the system behaves in a similar way, and with suitable approximations another pole equation can be derived [2]. For two reasons the situation for the network control is rather different, however: At first the control channels are not power controlled, and secondly the receivers are located at many different positions. As a consequence, the noise rise in the downlink can exhibit very high values, even if the mobile is located close to the base station and receives its signal with good quality. Therefore the noise rise is useless for network control in the downlink and has to be replaced by an intelligent allocation of the available base station transmission power. The use of these different control parameters for up- and downlink by the UMTS radio network controller will be illustrated in sections 3 and 4 after the simulation scenario is defined in the section 2. 2 Simulator and Simulation Environment 2. UMTS Simulation Tool To analyse the behaviour of a UMTS radio network, we developed a simulation tool which captures the dynamic processes in mobile communications networks using a discrete event simulation scheme. The discrete event mechanism together with an object-oriented modelling approach leads to separate computational entities of network elements which exchange messages/data similarly to a real communications network. The sequence of messages or events is determined by stochastical processes which are implemented in the activation of the network elements (throughout the paper a standard Poisson inter arrival process is used) as well as in the

movement and service models, associated with each mobile station which implement mobility in the simulation area and a sequence of active and silence times. The central part of the simulation is the repetition of a standard sequence of events (like transmission and reception of signals, C/I evaluation according to eq.(4) etc.). From this standard sequence a simulation cycle results [3], which in the present case corresponds to a UMTS frame of ms. Service requests by the network elements (user equipment UE, Node B and RNC) followed by accepting or refusing commands from the RNC implement features like soft or hard handover, power control, and the dynamic change of transport formats. In the downlink, control channels (PC- CPCH, SCCPCH and primary CPICH) are permanently transmitted. The dedicated uplink control channel is modelled by a power increase corresponding to the data rate ratio between user and control data on the dedicated channels. Evaluation of frame errors is performed by relating the CI-value computed for every transmission time interval in each network element to a link level curve specified by a linear relation between the logarithm of the frame error rate and the CI value in db. 2.2 Simulation Environment As a test case, a regular hexagonal cell layout based on the parameters proposed for UMTS network simulations by the ETSI [4] was used together with periodic boundary conditions to avoid edge effects. Six cells were simulated with a base station distance of 6 m. The area of the resulting hexagons corresponds to a circle with an effective (cell) radius of 3.5 km. The path loss equation was modified slightly with respect to [4] to match the german COMCAR research project [5], which studies the possible coexistence of DVB-T with UMTS at UHF frequencies. For a frequency of 8 MHz in a suburban outdoor scenario with a path loss exponent of 3.76, the path loss L is described by the equation L[dB] 9:8 + 37:6 log(r[km]): (5) The user speed was set to km/h (.33 m/frame). Shadowing was modelled with a zero mean log-normal distribution of standard deviation 5.66 db (assumed as the uncorrelated portion of a fading process with total standard deviation of 8 db) and a correlation length of m. All simulations were performed with hard handover to save codes in the downlink and to avoid additional optimisation parameters introduced by macro diversity. 3 Uplink Control using the Noise Rise To study the uplink behaviour, we chose a circuitswitched service of 64 kb/s with a processing gain PG 6. For the CI requirement we took a link level result available from TSG-RAN WG 4 for fading propagation conditions (case 3), as specified in [6]. The target for the power control loop was set to (CI) req :7 db. With the inter to intra cell interference ratio of f ' :45 determined for omni antennas in our simulation environment, the pole capacity M pole can be estimated according to M pole '. This leads to a pole capacity PG :(CI)req: (+f ) of ' 3 users per cell, where the required uplink power should diverge, which is seen in the plots of the average uplink transmit power level and the average uplink load as a function of the average UE density (Fig. ). Different values of the maximum allowed uplink load were used as parameter in the admission control of the RNC. Without constraints the system is unstable under user fluc- avg. UL Tx [dbm] avg. UL load [%] 25 Parameter: max. uplink load in RNC admission control 99% 9% 8% 7% 6, % 2 4 6 8 2 9% 99% 7% 8% 6% % 2 4 6 8 2 Figure : Mean uplink transmit power (top) and uplink load (bottom) as a function of the mean mobile density for different values of the maximum allowed noise rise in the admission control. tuations already well below the pole capacity. Therefore we illustrate the divergence for 99 % maximum allowed uplink load (which is sufficient to keep the system stable) and unlimited power available at the UE. The other curves for lower values of the maximum uplink load were obtained using additional constraints of 27 dbm maximum power level at the UE and a radio link timeout counter which caused the session to be dropped, if the frame error rate could not be met during % of the average life time (thus timeout occurs for UEs close the cell edges which hit their maximum power level for some time). We then studied how to choose the operating point for the admission control from the set of curves in Fig. and how blocking, dropping, and throughput could be optimised. The behaviour of the network seen from the RNC admission control is illustrated in Fig. 2. A maximum allowed uplink load of 6 % was used and the resulting uplink load distribution is plotted for the user density of 6.67 users/cell. To reduce the blocking rate, we low-pass filtered the RSSI level (received signal strength indicator, equivalent to the total received power level) measured at the cell site which is used by the RNC to determine the

uplink noise rise. Pdf.2.5..5 Traffic volume: 6.67 UE/cell max.ul Load: 6% Instantaneous UL Load 4 6 8 Uplink Load [%].2 Filtered Uplink Load.8.5 Pdf..5 Thereby the standard deviation of the.8.6.4.2.6.4.2 4 6 8 Uplink Load [%] Cdf Cdf UL RSSI [dbm] 98 99 2 3 Node B 6 4 Figure 2: Uplink load distribution (left) with and without low pass filtering and the uplink RSSI level (right) of the six cells as a function of simulation time. The RSSI level in each cell is averaged over cycles. For comparison non filtered values are shown during the first period. measured noise rise is reduced, which prohibits that service requests are prematurely rejected. The blocking rate due blocked [%] dropped [%] 4 Parameter: max. uplink load in RNC admission control % 6% 7% 8% 9% 2 4 6 8 2 9% 5 8% 7% 5 6% % 2 4 6 8 2 Figure 3: Block (top) and drop rate (bottom) as a function of the mean mobile density for different values of the maximum allowed uplink load in the RNC admission control. to rejections for too high uplink load from the admission control and dropping because of insufficient power at the UE is shown in Fig. 3. An acceptable dropping rate of 2 % allows about 5-6 users per cell (for the present investigation we used permanently transmitting users, for smaller activity coefficients, the effective number of users increases accordingly). To block other users early enough, a maximum uplink load of 6 % seems appropriate. For a base station distance of 6 km the system is coverage limited, because the uplink transmit power level for the first user entering the system is ' 8 dbm (a link budget with a processing gain of 2 db, E b N -.7 db, noise power (thermal noise + noise figure of 3 db) at the Node B of -3.8 dbm and the path loss for 2.5 km of 34.8 db leads to 8.3 dbm). Thus a noise rise of 9 db drives the mobiles into their limit. 4 Downlink Control with the Node B Power In the downlink a packet switched service, modeling WWW access, was used. Two alternative CI target values of and db (after despreading), equivalent to a frame error rate of %, leading to interference and code limited situations were compared. The intra cell orthogonality factor was set to ff :4 representing outdoor conditions. Primary and secondary common control as well as the pilot channel were permanently transmitted with, 27 and dbm, respectively. In contrast to the circuit switched service which permanently kept its code resource, the packet service released its code after each packet call. The parameters of the packet service radio access bearer were set corresponding to a data rate of 92 kb/s (spreading factor SF 6) with a mean packet call size of 7 bytes and an average number of 4 packet calls per session. A transmission time interval of ms and a transport block set size of 288 bits were assumed. Setting the average reading time to second ( frames) led to an activity factor of approximately.3. To control the network we used the following mechanisms: In the admission control connection setup was delayed due to lack of codes or power. A single OVSF code tree was available in each cell. The available base station power for downlink data channels was varied between 33 and 45 dbm. Dropping occured, if the accumulated delay of a session reached times the transmit time of an average packet call. If this happened before a single packet call was transmitted, the session was counted as blocked. In the congestion control we allowed the data rate to vary, if base station power was available or exhausted (up- and downgrading of ). This was done by doubling or halving the data rate and the associated transmit power, if lower or upper power limits in the base station were exceeded. The upper limit was the above mentioned total base station power for dedicated channels, while the lower limit was set below it. The data rate wasallowedtovaryin2stepsfrom92over96to48kb/s (SF 6, 32, and 64). The same link level result after despreading was assumed to give no preference to either one of the data rates. The operation of the variable rate service was compared to a service at constant rate, for which the connection requiring the highest power in a cell was dropped by the congestion control, when the power limit was exceeded. 4. Operation with Constant Data Rate The downlink behaviour of the network is shown in Fig. 4fortheCI target value of with an estimated pole

PG capacity of M pole ' :CI ' 7:5. Throughput per (ff+f ) cell and the sum of block and drop rate (which we considered together, because for a packet service blocking appears not favourable to dropping) are shown with the base station power as parameter. As can be seen in the bottom part of drop and block rate [%] 4 8 6 4 39 dbm 42 dbm 45 dbm 4 8 6 39 dbm 42 dbm 45 dbm power code 4 8 6 Figure 4: Throughput (top) and the sum of blocking and dropping reasons (bottom) per cell as a function of the offered traffic for different values of base station power (target CI 4dB). Fig. 4, for a CI target value of and for typical values of the combined blocking and dropping rate below %, power is the only reason for blocking or dropping of sessions which means that the network is interference limited. The operation with fixed data rate is quite inefficient, however, because only 2 3 users can be sustained on the average, which is much less than the pole capacity. For the lower target CI value of db, the pole capacity 8 6 4 blocking and dropping rate: : % : 2% coverage db C/I target code interference 33 35 37 39 4 43 45 available base station power [dbm] Figure 5: Throughput comparison for quality levels of 2 % and % of combined block and drop rate as a function of available base station power for CI targets of and. is estimated to be M pole ' 5 users, equal to the number of available codes (one node of the code tree with SF 6is used for the three control channels). As a consequence, the dominant dropping reason is expected to change (for sufficient base station power) from interference to code, because a non negligible probability exists, that the number of required codes exceeds 5. Our investigation of this situation as a function of base station power (Fig. 5) showed, that now around 5 6 users (out of 5) with constant data rate are accepted by the system and successfully finish their service. With respect to the target value the throughput is appr. doubled. Only for rather low base station power, sessions are dropped or blocked due to a lack of power as for the target CI value of. This represents rather a coverage problem due to high path loss than an interference. The real under these conditions is code, if only one code tree is used. 4.2 Operation with Variable Data Rate The operation of the up- and downgrading mechanism is illustrated in Fig. 6. The total number of present and the number of actively transmitting ones using the three different data rates are compared for two traffic loads as a function of simulation time. In addition, the cor- # of Code usage probability all present SF64 SF32 SF6 4 6 8.25.2.5..5 833 kb/s/cell active 833 kb/s/cell SF6 SF32 SF64 Code number 4 4 6 8.25.2.5..5 67 kb/s/ cell all present SF64 SF6 SF32 67 kb/s/cell SF6 SF32 SF64 4 Code number Figure 6: Connections using 92, 96, and 48 kb/s as a function of simulation time for offered traffic loads of 833 and 67 kb/s/cell (top). The associated code usage for SF 6, 32, and 64 is shown below. responding code usage, displayed as distribution functions of the OVSF codes with spreading factors 6, 32, and 64, is shown. On the left, a situation below the critical traffic volume is seen, where almost all operate at the highest data rate and almost only codes for SF 6 are used. Only during one excursion to a higher traffic load a marked usage of SF 64 occurs which is damped out, after the traffic volume has dropped again. Close to the critical traffic load (right), where interference and dropping start to increase significantly, the down- and upgrading mechanism becomes much more active, reacting to traffic fluctuations. If the thresholds are exceeded, it selects those (with highest or lowest powers in their cell) that can still change their data rate. For the illustrated case the lowest data rate is already used most fre-

quently. Another interesting feature is the dependence of the average data rate on the distance between the UE and Node B (Fig. 7). The average data rate during the life time of a connection is plotted as a function of distance to the Node B for different traffic volumes (and CI target value of ). For low load only a slight reduction relative to 92 kb/s is seen as a function of distance, while a significant decrease of the data rate is observed for higher traffic volumes. Comparing the results for CI target values of and Data rate [kb/s] 8 6 4 8 6 33 kb/s 267 kb/s kb/s 833 kb/s kb/s 4 UE NodeB distance [m] Figure 7: Dependence of the mean data rate of the UE on the distance to the Node B for different values of the offered traffic load. and the cases with fixed and variable data rate (Fig. 8), we find, that for the target value the drop and block rate is significantly reduced. By allowing the data rate to vary, the cell can adapt itself to traffic variations (if sufficiently long delays are allowed). This leads to a strong throughput increase for the same quality of the network (desired maximum drop rate). The throughput for variable rate is slightly enhanced at higher traffic loads. If the cell capacity can be divided into smaller portions, it can be better exploited. For the db CI target the dropping behaviour is more or less unchanged by allowing the data rate to vary, because the congestion control is only rarely operating at the limit of the base station power. Therefore throughput at the same quality level does not much increase. While for fixed data rate the throughput ratio between and target values was appr. 2, for variable data rate only a ratio of '.5 is found. A corresponding behaviour is seen in the life times and distances between UE and Node B. For the higher target value the life time decreases for fixed rate at higher traffic loads, because longer have a higher probability to get dropped. With variable data rate, the average life time (proportional to the average session delay) increases. This is due to the fact that the average data rate decreases. For the db target both for fixed and variable data rate a decrease of life is observed, because long will finish with smaller probability at high traffic load in both cases. Cell shrinking (preferential dropping of UEs close to the cell edge) is observed for both target values for fixed data rate at high traffic volumes. With variable rate this coverage reduction was markedly reduced for the higher CI target value. 6 8 throughput 4 block + drop rate 4 6 8 4 6 : variable data rate db throughput 8 4 block + drop rate 4 8 6 24 block + drop rate [%] block + drop rate [%] UE Node B distances [m] UE Node B distances [m] 24 2 life times distances 8 6 8 4 4 6 8 4 24 db 2 distances life times 8 4 4 8 6 24 Figure 8: Throughput and dropping as a function of traffic load (left) for fixed and variable data rates. Average distance between UE and Node B and average life time (right). CI targets of (bottom) and (top) are compared. 5 Conclusions The UMTS uplink is well controlled with the noise rise and adapted filtering of the RSSI level. In the downlink interference and code limited situations together with possible data rate variations were studied. For interference the adjustment of the data rate according to the path loss allows to enhance throughput significantly, while for code limited scenarios almost nothing is gained. Acknowledgment: The authors thank J. Huschke (Ericsson) and P. Herhold (TU Dresden) for critically reading the manuscript as well as M. Schweigel (TU Dresden) for his work on the simulator at an earlier stage. References [] K. S. Gilhousen, I. M. Jacobs, R. Padovani, A. J. Viterbi, L. A. Weaver, On the capacity of a cellular CDMA system, Trans.Veh.Techn. 4, pp. 3-32, 99 [2] K. Hiltunen and R. de Bernardi, WCDMA Downlink Capacity Estimation, pp. 992-996, VTC Conf. Tokyo ; K. Sipiläet al., Estimation of Capacity and Required Transmission Power of WCDMA Downlink Based on a Downlink Pole Equation, pp.2-5, VTC Conf. Tokyo [3] J. Voigt and G. Fettweis, A cycle-based approach for parallel simulations of mobile communications networks, Symposium on Performance Evaluation of Computer and Telecommunications Systems (SPECTS ), Vancouver, pp. 39-43 [4] ETSI Technical Report Selection procedures for the choice of radio transmission technologies of UMTS, 998 [5] COMCAR belongs to the UMTSplus initiative of the German ministry for research & education; see http://www.comcar.de [6] 3G TS 24, UTRA (BS) FDD; Radio Transmission and Reception UE life times [cycles] UE life times [cycles] 8 6