1 -WEDNESDAY DYNAMIC ENHANCED RADIO RESOURCE ALLOCATION FOR WIRELESS COMMUNICATION NETWORKS

Transcription

1 DYNAMIC ENHANCED RADIO RESOURCE ALLOCATION FOR WIRELESS COMMUNICATION NETWORKS KllA. Egner* Northern Telecom Inc. 220 Lakeside Blvd. Richardson, Texas Vasant K. Prabhu Department of Electrical Engineering, The University of Texas at Arlington, Texas ABSTRACT Mobility of the wireless user creates uncertainty in demand and non-optimum use of radio resources. Current wireless networks are periodically re-configured manually to improve network performance. These network reconfiguration and performance optimizations consume thousands of engineering hours and are often done by trial and error. However, the benefit of these efforts will be minimized as increasing competition throughout the world causes the service provider to dynamically change pricing structures to attract new customers and maintain existing customers. These changes in wireless service tariff and user base can drastically impact the network usage. This uncertainty further reduces the period of time a non-dynamically tuned network will perform optimally. Therefore, the need to re allocate the network and radio resources on the fly to maximize network infrastructure usage and minimize network costs will become increasingly important. This paper will propose a new enhanced method to dynamically allocate radio resources and will evaluate the performance impact of this scheme using the configuration and operational measurements of an existing Global Standard Mobile (GSM) network over a period of several days. The new enhanced dynamic radio method reduces the total radio channel requirements by 0-20% over the currently used fixed allocation scheme. This new method improves the radio resource performance (the average C/l per Erlang) by 2.4 db over a fixed allocation radio system. I. INTRODUCTION Current radio networks are statically engineered based on busy hour traffic. This results in over-allocation of radio channels for a large percentage of the day. These extra radio channels consume more spectrum and create additional interference in the network. A dynamic allocation system adaptively defines radio resource requirements based on time intervals. Allocating radio resources on a time interval basis can reduce the total channels required and result in less interference in a network. Differing from conventional approaches, this paper defines a radio deployment modeling which will become the basis for a new radio frequency allocation scheme. This allocation scheme is based on a new performance metric which uses both the deployment information and operational measurements. The modeling assumes that an accurate estimate of temporal demand can be generated via continuous network monitoring. This is frequently done in actual networks. Section II discusses the demand estimation and trafec channel requirement *. This work is part of doctoral research program at the University of Texas at Arlington. computation over time. Section I highlights the importance in radio deployment modeling and its impact on received signal and interference strength. Network voice quality is modeled based on a potential interference matrix in Section IV. Section V discusses how the frequency allocation algorithm will work. Section VI compares the results of current fixed allocation to the new enhanced dynamically deployed radio resource approach.. DEMAND ESTIMATION AND RADIO CHANNEL REQUIREMENTS Before radio channels can be assigned, a computation of how many traffic channels are required to meet the time interval demand must be performed. The current state of the art ~olves the nondynamic allocation based on estimating the traffic during the cell's busy hour. In a dynamic radio resource allocation scheme, cells are allocated based on interval demands. The demand estimation is based upon actual teletraffc usage in a real network. Current cellular teletraffc patterns have been shown to be approximately periodic based on day of week and time of day. Fig. shows the statistical cyclostationaty of daily wireless traffic usage for one region of an operation DCS800 network. This traffic cyclostationary is typical across wireless network and will be used to predicted interval demands. Current measurements will be continuously incorporated into the estimates to evolve traffic predictions as the network demand evolves. Relative Call Attempt Volume for Area ---.-TUESDAY -WEDNESDAY _....-_-... Fig.. -Periodicity of Wireless Traffic Patterns All radio resource calculations assume that traffic dema follows an M/M/N/N queue model [l]. This model is valid for call patterns that have Poisson arrival statistics and exponentially distributed service times. The model also assumes that the call arrival process is not affected by number of calls in progress. The /98/%5.00 Q 998 IEEE

2 grade of service calculation based on ErIangl3 formula is given in Eq. (, Fig. 2 is at the edge of BTS 0. GOS = EIA.N\ = N \--I I i=o Ai/i! where A represents the offered traf ic in Erlangs and N is the number of allocated radio channels. The parameter GOS represents the blocking probability and is typically engineered at % or 2%. This formula can be numerically solved when the desired GOS and A are known and N is desired. The function which solves for N will be referenced from this point forward as Erlan@(A,GOS). In a static radio resources scheme, radio channels are allocated to cells based on individual busy hour data. The static radio resources computation is given in Eq. (2). Note that the peak offered traffic at cell i during the day is used to engineer the radio channel requirements, and K N, = ErlangB[Max,(A,, *), GOS] (2) i=l where Ns represents the number of radio traflic channels required for static allocation and Ai,t represents offered traffic at cell i at time t. The Max t(.) function is used to find the peak traffic for each cell. Equation (3) shows the radio requirements for dynamic radio resource method where radio resources are engineered on a per period basis, K N, = ErlangB[A,,, GOS] i= where Nd represents the number of radio traffk channels required for cell i at time t. Dynamic radio allocation will achieve sigtuficant savings over static radio assignments. This method improves voice quality through the decrease of interference from radio channels which will not be needed during a particular period. The channel requirements for both the dynamic and static allocation methods will be processed by the interference minimizing algorithm which will be discussed later. The paper will use 200kHz TDMA[2] radio channel, Each channel will have 8 time slots. The analysis will assume one time slot is reserved for signalling and 7 time slots are available for traffic.. RADIO NETWORK DEPLOYMENT MODELING First the paper will consider the geographic locations of the different power sources. The power source model does not assume a uniform spacing for cells [3 J but instead uses actual cell position data. All mature networks will have non-homogeneous deployed cells and frequencies which reflect the non-uniformity in the network demand [4]. These networks cannot be modeled using homogenous clusters. Fig. 2 illustrates the interfering mobile station in the surrounding cells and the desired signal being sent by the mobile station (MS) in the center cell. The receiving antenna is basestationq3ts) 0 and the MS shown in (3) Fig. 2. -Interference in a Non-Homogeneous Cell Plan Cellular Radio path loss models are typically based on real measurements which have been fitted to predict path loss under different parameters which typically include environment, transmission frequency, and antenna height [5]. This section defines the maximum cell radius for a particular cell under a given set of conditions. Cell Radius The link budget can be used to compute the approximate cell radius for different sites in different environments. Maximum cell radius can be estimated by computing the distance at which the link budget equals the path loss and the transmitter is at maximum power. Eq. (4) shows thls equality LinkBudgetiJ = Pi- MiJ- Pi+ G = ai,jlog(d) + pi,. where Pi equals the maximum power at transmitter i, Pj is the minimum receive power at j, Mij is the total fade margin budgeted for path from i to j. MiJ including path loss due to penetrating cars and buildings along with log-normal fading. G in (4) represents antenna gain due to both transmit and receive antennae. The path loss along ij is modeled using a linear model with a: slope, a:j,j and intercept, B,( Solving the maximum cell radius, R, becomes the following. (LinkB4?;f+, j - Pi, j ) R,,, U = 0 (5) Eq. (5) shows the computed receiver noise limited case. Most urban and suburban coverage areas are not receiver noise limited but are interference limited. Therefore the cell radius in these areas will be smaller than the maximum cell radius. For the interference limited case, the maximum cell radius will be defined based on co-channel interference distances. These minimum Carrier to Interference Ratios,CIRs, are driven by the corresponding technologies receiver sensitivities. North American TDMA standard has a minimum CIR 2 or 6 depending on whether antenna diversity is or is not used, respectively. The GSM standard has a minimum CIR of 9dE3 if frequency hopping is turned on or db in case it is off. Daktance MatrLx The distance matrix shown in Eq. (6) is computed and stored in a matrix to reduce duplicate computations, (4) /98/$ IEEE 70 VTC '98

3 where R, is the cell radius for BTS and di,j is the average distance between the interfering MS in cell and the cell s receiving antenna in j. All &stance units are in kilometers (km). The non-diagonal elements in the D matrix represents the average physical distance between the receiving antenna and the interfering mobile stations. The diagonal terms represent distance to communicating mobile subscribers on the edge of the cell. Antenna System Modeling Typical suburban and urban cell sites are deployed in a tri-sector configuration which focuses more radio resources on smaller areas for traffic limited cases. In coverage limited cases, the tri-sector configuration can also be used to extend the site footprint. Since a sectorized antenna pattem significantly reduces gain in the non-pointing directions, it is extremely important to take the antenna orientation into account when computing the channel interference from other cells. Fig. 3 shows the orientation of an interfering MS signal in relation to the receiving cell X. radio channel for each period. This interference model will use both the deployment information and operational measurements to define a new performance metric. The potential interference matrix will be used by dynamic radio selection algorithms to select radio channels that minimize interference and maximize voice quality. Fig. 4 gives an overview of the dynamically enhanced radio resource allocation scheme. Each processing box is referenced with the section number in which the process has been discussed. The potential interference will be computed based on the deployment model. The dynamic radio channels per interval is computed based on the periodic demand estimates. The objective function for th~s algorithm is to minimize the global Interference to Carrier, UC, metric. The first pass algorithm selects radio channels which minimize the potential interference in the local area. The second pass algorithm adjusts the first pass channel assignments to reduce the actual I/C for the network. The second pass algorithm reassigns channels which reduce the global I/C metric. r Fig. 3. -Antenna System Modeling This antenna angle is represented by 0,-0,. The angle between the transmitting MS in sector x and the receive cell tower antenna is represented by 0, -fir The antenna system gain for each MS is computed based on the relative angle from the receiving sector antenna mapped into Gsector~(B). The system gain calculation is represented as where Gij is the antenna system gain experienced by the interfering signal and Gi,i is the antenna system gain experienced by the desired signal from the MS in cell x. IY DYNAMIC ENHANCED FREQUENCY ALLOCATION ALGORITHM The new algorithm defines a method for constructing a network wide interference matrix and the algorithm used to select (7) Fig. 4. Dynamically Enhanced Radio Resource Algorithm Potential Interference Modeling Ths section departs from conventional frequency planning and enters the dynarmc frequency planning arena. The objective of dynamic frequency planning will be to improve the overall voice quality of a cellular network by reducing dominating signal impairment sources. This analysis will consider only the radio reverse link, which is typically the limiting link for most cellular networks. The signal sources will consist of the transmitting MS in the cell and the potential interfering MS s in the vicinity. The carrier signal strength can be computed as the mobile transmission power plus the antenna system gain minus the propagation loss. c, = PI,, + GI, 2 - a,,log(r,) - PI, dbm Since most suburban and urban wireless networks will be operating in an interference limited environment, and most (8) l98/$ IEEE I702 VTC 98

4 networks actively control the power level of the transmitting MS, it is reasonable to model the MS power at the receiver to be fairly constant. This eliminates the difficulty in estimating the MS transmit location. The minimum received signal strength at each cell receiver is assumed to be in the range of -93 to -97 dbm. The interference power level is calculated as the average interference power level added to antenna system gain minus the path loss from average distance between the interfering MS position and the receive antenna, Zi, j = P. '.I. + Gi, j - ai, jlog (ai, j) - pi, j dbm The above carrier signals and interference signals are used in the construction of the potential power level matrix represented as (9) frequency allocation scheme. Each column in the matrix describes the frequencies which are allocated for each cell. An entry in the matrix will be either zero if the frequency is not allocated or one if the frequency is allocated. The actual interference to signal ratio matrix, i/c, is computed by multiplying the radio channel allocation matrix by the potential interference matrix. i/c = F[N x K ] ~ / c ~ ~ x K ~ ] ~ ~ ~ ~, [ K i/c = or i/cl, i/cl, i/cl, i/c2, i/c2, i/c2, ' '.2 'i,k] p = p. c2 dbm where P is a K by K matrix containing all potential interferers from each cell to all other cells. The power matrix above is introduced to eliminate recomputation of potential interferers. The diagonal elements contain the power level received from mobiles under constant transmit power control. The non-diagonal matrix elements represent the interference power levels. For example matrix element 2,~ represents the interference power level from cell 2 to cell. Note that this matrix in general will be non-symmetric due to the difference in effective radiated power from each BTS. In order to add interference power levels, P will be converted from logarithmic units back to milliwatts and stored in matrixp, Frequency Planning Objective Function This matrix contains the actual interference levels existing in the network based on the given allocated frequencies. The matrix will have N rows, one for each radio frequency, and K columns, one for each cell in the network The dynamic frequency planner uses both deployment information and traffic measurements to maximize network voice quality. This section defines the objective criteria for optimizing the radio frequency plan. Automatic frequency planning systems [7] attempt to maximize the voice quality in a cellular network by maximizing the C/I under limited system bandwidths. The objective function maximizes C/I by minimizing the average I/C ratio per Erlang in the network, and Pij are elements of P and the values are in dbm. Each potential interference pij will be normalized by the corresponding competing carrier signal cj to form nit The matrix representation i/cpolential contains the normalized potential power levels for all potential interfering signals. A radio channel allocation matrix is used in computation of actual signal interference. Equation (5) shows the radio channel allocation matrix F, where index N is the maximum radio channel number and index K is maximum cell number. The F matrix represents the where Ai,k is the offered traflic generated on frequency k in cell I. Minimum Carrier to Interference Level The objective will have the following constraint associated with each element in the i/c matrix: i/qk = l/(c/imin) for VCelli (6) This constraint is imposed to guarantee a minimum grade of service in voice, typically quantified by a bit error < Adjacent Channel Interference Constraints A channel exclusion constraint is used by the automatic frequency planning algorithm to minimize adjacent channel interference. This test prevents the assignment of adjacent channels within BTS. For example, if channel 5 is assigned to a BTS, then channels 4 and 6 will be excluded from site assignment. Equation (7) shows an exclusion matrix for a two site network with seven radio channels. Each column is a cell index and each row is a frequency index. Columns -3 refer to BTS and columns 4-6 refer to BTS 2. For example, the entered at El,l indicates radio channel is assigned to cell in BTS, and the 3 x's entered in E2,, &,2, &,3, indicate radio /98/% IEEE i 703 VTC '98

5 Ohmori, Kawano, frequency 2 is excluded from use in all cells of BTS. E= -I lxxxxx xxxlxx xlxxxx xxxxlx xxlxxx xxxxxl xxx V. RESULTS AND CONCLUSION Fig. 5 shows the traffic statistics for the entire 366-cell network over the period from 0 a.m. to 4 p.m. for a typical Saturday in an actual operating Network. The entire network has three different peaks during the day, one at p.m., one at 3 p.m., and final peak around 6 p.m. The data was pulled from an Asian market, where business people typically work half a day on Saturday Fig. 6. Required Radio Channels vs P 750 a Fig. 5. Network Erlangs vs. Fig. 6 shows the radio channel requirements for a GSM System. The fixed channel requirements remain flat at 494 channels for the entire day. The dynamic radio requirements peak during the network busy hour, PM. The dynamic radio scheme savings will vary based on the hour, ranging from 0-20%. The dynamic allocation scheme reduced the average radio channels per cell per period by 5.6%. The dynamic radio resource scheme not only decreased the radio equipment requirements but also improved voice quality by reducing radio interference. Fig. 7 shows the relative improvement in average C/ I from a dynamic network radio resource allocation scheme versus a fixed allocation scheme. The average C/I per Erlang per period increased by approximately 2.4 dl3. Future work will focus on the exploring of other algorithms to minimize average I/C using such methods as steep descent gradient and back propagation neural networks. The network deployment model will be extended to model the impact of antenna down-tilt and the dynamic algorithm will be extended to include tuning of down-tilt angle as a means to further reduce interference levels. 6.0 WI -4-B namc cn 4 0 io Fig. 7. Network Voice Quality Performance vs. Allocation Scheme REFERENCES Bertsekas, D., Gallager,R., Data Network, Prentice Hall Inc., 987, p.40 Mouly, M., Pautet M., The GSM System for Mobile Communication, Cell & Sys, 992, p Yacoub, M.D., Foundations of Mobile Radio Engineering, CRC Press, 993, pp Moreno, A. S., An Analytical Procedure for Cellular Frequency Planning, IEEE Conference 993, pp Okumura, Y ~ E ~ T, and Fukuda, K., Field strength and its variability in VHF and UHF land mobile services, Rev Elec. Comm Lab., 6, , September- October 968 Raith, K., Capacity of Digital Cellular TDMA Systems, ZEEE Transactions on Vehicular Technology, V0.40, No. 2, May 99, pp Gill, T.M., An Automated System for Frequency Planning, EEE Conference 993, Publication /98/$ Em 704 VTC 98