Você está na página 1de 7

Downlink Performance Improvements in (A)synchronous GPRS Networks using Single/Multiple Antenna Interference Cancellation Techniques

Larissa Popova1 , Raimund Meyer2 , Wolfgang H. Gerstacker1,2 , and Wolfgang Koch1


Mobilkommunikation, Universit Lehrstuhl fur at Erlangen-Nurnberg Cauerstrasse 7/NT, D-91058 Erlangen, Germany Phone: +49 9131 85 25017, Fax: +49 9131 85 28682 Email: {popova, gersta, koch}@LNT.de 2 Com-Research GmbH, Wiesengrundstrasse 4, D-90765 F urth, Germany Email: com-reseach@t-online.de
Abstract: The aim of this paper is to assess GPRS data throughput of GSM networks for single and multiple antenna interference cancellation (SAIC/MAIC) receivers in the terminals. This is done by using a two-step top-down simulation methodology. Both synchronous and asynchronous networks with traditional European reuses of 1/3 and 3/9 are considered. It turns out that SAIC/MAIC receivers are very benecial also for European network planning. The results indicate a substantial gain in data throughput in the order of 20-35% for SAIC compared to terminals with a conventional receiver. A further dramatic increase in throughput up to 120% can be achieved by terminals using an MAIC receiver. In spite of the widespread view that a maximum gain will be achieved for a synchronized network, where the amount of overlap between desired signal and interference can be controlled, our results show that the performance of both network types is very similar.
1

1.

Introduction

Although 3G systems like UMTS with its progressive wideband technology enable a very high-rate transmission and its deployment continues to gain momentum currently, GSM will still continue to be a viable technique with some potential of further improvement. One challenge for GPRS/EDGE is still to provide high data rate services at high trafc load. On the one hand, the demand for high-rate services requires intensive use of the coding schemes CS-3 and CS-4 in GPRS, which are more sensitive to noise and interference. On the other hand, the higher trafc load requires lower frequency reuse distances in the radio access network, which increases cochannel interference. One attractive solution to this problem is interference cancellation (IC). For GSM/GPRS/EDGE such techniques have been developed and investigated since the mid of the 1990s, see [15, 18, 16, 6] and references therein. Depending on the number of receive antennas, single antenna interference cancellation (SAIC) and multiple antenna interference cancellation (MAIC) are distinguished. Both have a potential for signicantly increasing the spectral efciency of GPRS and EDGE. A good overview can be found in [9, pp. 224-229]. Most investigations consider very tight reuse factors like 1/1 with fractional cell loading, mainly driven by US-operators with small spectrum. In Europe, the interest in IC techniques was more moderate. Due to the fact that most European network operators have sufcient

spectrum in order to be able to apply large reuse factors, the gain from SAIC or MAIC is commonly assessed not to be signicant. The basis for this assessment are results from studies for speech services only, e.g. [2]. For GPRS, however, the scenario is different. The coding scheme CS-4 for example is much more sensitive to interference than any of the schemes used for speech such that it can be applied only very rarely. We present some results showing that its utilization and in turn the average throughput can be signicantly increased if SAIC/MAIC is used in GPRS networks even with a reuse of 3/9. This motivates its use also in networks with more conservative frequency reuse factors. A synchronous GPRS network with cloverleaf cell structure, reuse 1/1 and fractional cell loading using SAIC in the downlink has been studied in [14]. The results indicate interesting throughput gains especially at a high trafc load. In [17], dual antenna diversity reception for Enhanced (E)GPRS terminals for synchronous networks with reuse 1/3 and 3/9 is considered. Only 8ary phase-shift keying (8PSK) modulation is used and a minimum mean-squared error decision-feedback equalization (MMSE-DFE) receiver is employed. Signicant capacity enhancements are demonstrated. An interface from link level (LL) to radio network (RN) has been used for obtaining the numerical results in [17], cf. also Section 3. Of course, certain simplications are required for the denition of this interface. However, due to the great impact of the interface on the calculated performance, more investigations are required. This paper addresses the problem of coupling radio network and link level simulations from a different point of view. RN results are used as input to LL simulations. This allows to focus on aspects like adjacent channel interference, as well as synchronous and asynchronous networks. The results could be used for verifying or improving the interface for the LL to RN approach. We show results for 1/3 reuse and 3/9 reuse for SAIC and MAIC for both, synchronous and asynchronous GPRS networks. These results indicate that there is not a signicant difference for synchronous and asynchronous network operation as far as blind IC techniques [11, 12, 13] are applied. The paper is organized as follows. In Section 2 the main idea behind the MAIC algorithm is described. For a detailed description of the used SAIC algorithm, we refer to [11, 12, 13]. Section 3 explains our simulation

methodology which differs from the commonly used approach, where the link level is characterized by an analytical or numerical approximation to the block error ratio (BLER). Section 4 gives an overview of the used simulation assumptions and Section 5 presents the simulation results. Finally, some conclusions are drawn in Section 6.

2.

MAIC Algorithm

In the following, the used MAIC strategy, cf. also [11], is reviewed. We consider the case of a mobile terminal with NR receive antennas whose received signal is impaired by intersymbol interference (ISI), cochannel interference1 resulting from K interfering base stations, and additive white Gaussian noise. A transmission with linear modulation using realvalued coefcients2 is assumed for the desired signal as well as for all interferers. Thus, the discretetime complex baseband received signal at the ith receive antenna, i {1, 2, . . . , NR }, is given by
K

For K 2 NR 1, it can be shown that the lter transfer functions Pi (z ) (z transforms of impulse responses pi []) can be chosen in that way that all coefcients of each overall interferer transfer function after combinaNR Pi (z ) Gi, (z ), {1, 2, . . . , K }, have the tion, i=1 same phase as c . Therefore, also the total interference after ltering and combination has the same phase as c and vanishes after projection onto c, i.e., zero forcing with respect to all interferers can be performed if K 2 NR 1. Furthermore, it can be shown that the useful signal after projection does not vanish in almost all cases, i.e., under very mild conditions for the subchannel impulse responses. Hence, perfect interference suppression is possible using the proposed receiver structure. In a practical implementation of the proposed MAIC strategy, FIR lter impulse responses pi [], i {1, 2, . . . , NR }, are chosen in that way that the signal after projection approximates a desired signal
qd

w[k ] = gi, [] b [k ]+ni [k ]


=0

d[] a[k k0 ]

(5)

ri [k ]=

hi [] a[k ]+
=1

(1) (a[k ], b [k ], {1, 2, . . . , K }: realvalued data symbols of desired signal and of th interferer, respectively; n[k ]: white Gaussian noise). hi [] and gi, [], i {1, 2, . . . , NR }, {1, 2, . . . , K } are the discrete time overall impulse responses valid at the ith receive antenna corresponding to the desired and the th interfering signal, respectively. All mobile subchannels are assumed to be (approximately) constant during each data burst. For the MAIC algorithm, an arbitrary nonzero complex number c is selected and a corresponding number c = Im{c} j Re{c} (2)

(Re{}, Im{}: real and imaginary part of a complex number, respectively) is generated. c and c may be interpreted as mutually orthogonal twodimensional vectors. Each received antenna signal ri [k ] is ltered with a complexvalued lter with coefcients pi [], i {1, 2, . . . , NR }. After ltering the signals are combined and then projected onto c, i.e., the realvalued signal
NR

y [k ] = Pc
i=1

pi [] ri [k ]

(3)

is formed, where Pc {x} denotes the coefcient of projection of a complex number x onto c, Pc {x} = < x, c > Re{x c } = 2 |c| |c|2 (4)

(< , >: inner product of two vectors).


1 In the derivation we restrict ourselves to cochannel interference, but the algorithm is also benecial for the case of adjacent channel interferers (ACIs). 2 Gaussian minimum-shift keying (GMSK) used in GSM/GPRS is an example for such a modulation.

with realvalued coefcients d[] and a delay k0 which are both free parameters for optimization. The FIR lter coefcients pi [], i {1, 2, . . . , NR }, and d[] are jointly optimized for minimization of the error consisting of cochannel interference and noise, i.e., a minimum meansquared error (MMSE) criterion is applied instead of zero forcing (ZF) with respect to the interferers. This is because MMSE lters can be calculated more efciently. In the MMSE optimum solution, the output signal of projection y [k ] is expected to be approximately free of cochannel interference for K 2 NR 1, which is possible in principle according to the previous considerations, and has minimum noise enhancement. Intersymbol interference is not removed since it can be better taken into account by a subsequent trellisbased equalizer. In order to avoid the trivial solution in lter optimization, the additional constraint d[0] = 1 is adopted. It should be noted that this constraint is similar to that in optimization of the lters of decisionfeedback equalization (DFE). Because the feedforward lters of a multipleantenna ZFDFE produce white output noise after signal combination, the noise component in y [k ] is approximately white for the optimum lters. The lter order qd can be chosen for a tradeoff between performance and complexity of trellisbased equalization. Adaptive adjustment of the lter coefcients pi [], i {1, 2, . . . , NR }, and d[] may be performed according to a leastmeansquare (LMS) algorithm, which is a straightforward generalization of the LMS adaptation algorithm of the SAIC receiver of [11, 12, 13]. Alternatively, a recursive leastsquares (RLS) algorithm may be selected for lter adaptation, or the optimum lters may be determined via direct minimization of the timeaveraged squared error within a window of size K 1 2 K , 1/K k=0 (y [k ] w[k ]) . All algorithms perform blind adaptation with respect to the interferers because only the training sequence of the desired signal has to be known.

Finally, we note that for NR = 1 the described MAIC algorithm specializes to MIC (mono interference cancellation), the algorithm of [11, 12, 13] used for SAIC simulations in this paper.

3.

Simulation Methodology

In order to assess the effect of a receiver algorithm on network performance, two types of simulations are required: LL and RN simulations. A good overview of the characteristics of the different types of simulators can be found e.g. in [9, App. E]. For the performance analysis of GPRS we use a time-driven RN simulator based on the concept described in [19]. In order to evaluate the radio network performance it is necessary to have access to link quality measures like the bit error probability (BEP) or BLER. However, the mapping from physical parameters like the carrier-to-total-interference ratio (CIR) and Eb /N0 (Eb : average received bit energy, N0 : power spectral density of white Gaussian noise) to BLER is not an easy task and can only be obtained accurately from LL simulations: An interface between LL and RN simulations is required. In principle, two approaches are possible (see Fig. 1):
Statistical analysis e.g. frame-wise block error indicator RN simulator analytical model for BEP = f(CIR, DIR, ...) LL simulator LL2RN approach Statistical analysis RN2LL approach RN simulator received power of useful and interfering signals e.g. frame-wise block error indicator

LL simulator

Figure 1: Simulation methodologies. From LL simulations a mapping from physical quantities like instantaneous Eb /N0 or CIR to BLER is obtained which is approximated by a simple function. This function is then used in the RN simulator to assess the instantaneous quality of each individual link. In this approach, the nal evaluation is done in the RN simulator. We denote this approach by LL2RN (link level to radio network) simulation. This is the commonly used method described in the literature [7, 14, 2]. From time-driven RN simulations we get for each mobile a sample sequence with physical parameters like power of desired signal, power of strongest cochannel interferer, power of 2nd strongest cochannel interferer, power in the upper adjacent channel, etc. For each time slot in GSM/GPRS we get such a set of physical parameters. This sample sequence is used as input to the LL simulator in order to accurately determine the BLER. In this approach, the nal evaluation is based on results from the LL simulator. We denote this approach by RN2LL (radio network to link level) simulation. Both approaches have their pros and cons, which will be briey discussed in the following. LL2RN has the main advantage that it allows the RN simulator to model dynamic procedures based on link quality, e.g. handover, power control, and link adaptation (LA). Also, higher

layer protocols can be integrated in order to study latency times and throughput. This is the preferred approach as long as the mapping from physical parameters to BLER can be described in a mathematically tractable way [8]. For the study of SAIC algorithms it is known that the performance depends not only on CIR, but also significantly on the dominant-to-residual interference power ratio (DIR), requiring a two-dimensional mapping depending on CIR and DIR. This approach is used e.g. in [14]. Especially when two or more receive antennas are used (MAIC), the BLER depends on much more parameters and it is difcult to come up with a simple and realistic model, which could be used in RN simulations, cf. [17]. An additional complexity comes into play, when asynchronous networks are studied (in [17], only synchronous networks have been investigated). In asynchronous TDMA systems, the BLER depends also on the amount of time shift of the interferers against each other and against the desired signal. This last drawback motivates the RN2LL approach, where we can obtain as many physical parameters as we need from RN simulations for each individual time slot without additional complexity. In the RN simulator we model path loss and shadowing only, but no multipath fading. This allows studying effects of different power delay proles, antenna correlations as well as frequency correlations (for realistic frequency hopping) in the LL simulation. As a major advantage, this approach allows to compare different receiver algorithms on a common basis without the need of an analytical approximation of the BLER. A disadvantage of this approach is the impossibility to simulate quality-based algorithms like power control, handover and link adaptation in the RN simulation. For the purpose of our investigations we prefer the RN2LL approach due to the lack of availability of a satisfactory BLER mapping for MAIC. This requires a few words to the mentioned disadvantages. Power control is not applied here, i.e., we assume each base station transmits with constant power on the TCH (Trafc Channel). Instead link adaptation is applied in the LL simulator. The RN simulator works for 5 sec length. It is reasonable to assume that no handover occurs during such a short connection time due to measurement and averaging delays in the handover decision process. LA can be modelled in the LL simulator, cf. Section 4. It has no effect on physical parameters in the RN simulator, since the base station transmit power is constant. In Section 5, it is demonstrated that simplied models are not sufcient for accurate performance predictions, especially for the presence of adjacent channel interferers (ACIs), which justies our approach.

4. Simulation Setup
4.1. Radio Network Simulation The time-driven radio network simulator works with time steps of a TDM frame (i.e., 4.615 ms). It employs a regular cell plan consisting of 75 hexagonal sector cells

in a cloverleaf arrangement for a 1/3 reuse pattern and 225 sector cells for a 3/9 reuse pattern. To avoid the boundary effects of the nite simulation area, the wraparound technique is applied. The propagation conditions are specied by the Okumura-Hata model [10], and the UMTS 30.03 path loss model is used (120.9 + 37.6 log10 d @ 900 MHz).
Parameter Reuse Sys. bandw. for TCH TCH Carrier freq. Cell radius Radio freq. Freq. hopping Log normal fading STD Correlation distance Noise oor BTS transmit power Call duration Mobile speed Downlink power control Handover Cong. 1 from [2] Cong. 4 from [2] 3/9 1/3 5.4 MHz 7.2 MHz 27 36 0.5 km 0.3 km 900 MHz Ideal random hopping 6 dB 110 m -113 dBm (noise gure = 8 dB) 20 Watt (43 dBm) 5 sec (xed) 30 km/h OFF OFF

Table 1: Main simulation parameters. Ideal random frequency hopping is implemented on burst level. Mobiles are distributed randomly over the cell area. The mobility model is quite simple: Each mobile has a given constant speed during a simulation session. Only the movement direction is randomly selected for each mobile. Initially, all mobiles are connected to the base station to which they have the highest path gain. During the simulation (duration of 5 sec), a continuous transmission of data is assumed for each mobile. The most important radio network parameters are taken from congurations 1 and 4 of [2] and are summarized in Table 1. As a measure for radio network capacity and spectrum utilization the well-known effective frequency load (EFL) is used, which is dened as the ratio of carried trafc per sector (in Erlangs) to the total number of available TCHs (= no. of TCH frequencies times the number of time slots per carrier). Each simulation run contains approximately 1000 calls. 4.2. Radio Network to Link Level Interface After the radio network simulation is completed, the following power levels are delivered on a burst-by-burst basis: useful signal, three strongest cochannel interferers (CCIs), three strongest ACIs, separately for higher and lower carrier frequency (ACI high and ACIlow ), and power level of the sum of all residual interferers, separately for CCI, ACIhigh , and ACIlow (13 power levels in total). 4.3. Link Level Simulation The LL simulator models the radio link between the base station and the considered mobile station separately for all four GPRS coding schemes CS-1 to CS-4 according to the 3GPP standard. Multipath fading is modelled according to the 3GPP GERAN [1] Typical Urban (TU) channel model. Three receiver types are analyzed: a conventional GSM mobile receiver (equalizer) (CEQ) [3], an SAIC receiver [11, 12, 13], and an MAIC receiver [11]. For these receivers a frame error indicator is deter-

mined by the channel decoder on a block-by-block basis for CS-1 to CS-4. To simulate the interferer environment in asynchronous networks the time shift of each co-and adjacent channel interferer is modelled independently as proposed in [2]. Due to the top-down simulation approach (open loop simulation without feedback path from LL simulator to RN simulator) a-posteriori LA based on the average BLER provided by the LL simulator was applied. The average BLER for CS-1 to CS4 is computed four times per call, i.e., every 1.25 sec. For each of these periods the highest coding scheme is selected, for which this average call BLER is below the given target error rate of 10%. As this LA is done a-posteriori and outside the RN simulator the absolute data throughput might show some deviation compared to simulations where LA is part of a closed loop simulation. However, it can be assumed that the observed gains of SAIC/MAIC relative to the conventional receiver are quite reliable although their absolute numbers might still be subject to moderate variations. After the physical layer simulation is completed, a statistical analysis of the frame error indicator is performed in order to compute the overall data throughput per time slot per mobile.

5. Numerical Results
In this section, results from RN2LL simulations are presented assuming 100% CEQ or SAIC or MAIC mobile penetration, respectively. Both synchronous and asynchronous networks are considered. Simulation parameters for the MAIC algorithm are identical to those for CEQ or SAIC receivers except that additionally two receive antennas are assumed with a 70% mutual correlation and an antenna gain imbalance of 3 dB. These values are currently under discussion in standardization. It should be noted that the throughput results are not very sensitive to these numbers [5, 4]. For SAIC and MAIC the DSP code of an implementable prototype receiver is used. Also, typical hardware impairments are included in the simulations [5]. Fig. 2 shows the system performance versus the frequency load for CEQ, SAIC, and MAIC. As a criterion for system performance, we consider the average user information rate (excluding header bits and higher layer protocol overhead) for a given trafc load. The left graphics depict GPRS throughput and relative throughput gain for a 3/9 reuse and the right ones quantify the performance gain for a 1/3 reuse. For the synchronous case, training sequence 0 (TSC 0) and TSC 1-7 randomly selected have been used for the desired signal and the interferers, respectively. The difference in performance between synchronous and asynchronous operation is very small. Although the common view is that synchronous networks allow higher SAIC/MAIC gains than asynchronous ones, our performance results indicate the excellent capability of the considered SAIC/MAIC solutions to cope with asynchronous interference conditions without any signicant loss in performance for highly loaded data networks. For the investigated 1/3 and 3/9 reuse patterns the results do

GPRS (CS1 to CS4) data throughput (1 slot) for 3/9 reuse


18

GPRS (CS1 to CS4) data throughput (1 slot) for 1/3 reuse


16 CEQ async SAIC async MAIC async CEQ sync, TSC17 SAIC sync, TSC17 MAIC sync, TSC17

GPRS data throughput (kbits/s)

16

14

14

12

GPRS data throughput (kbit/s)


CEQ async SAIC async MAIC async CEQ sync, TSC17 SAIC sync, TSC17 MAIC sync, TSC17 2 3 4 5 6 7 8 9 10 11 12

12

10

10

10

15

20

25

30

80

SAIC/MAIC relative throughput gain (3/9 reuse)


SAIC async MAIC async
140

SAIC/MAIC relative throughput gain (1/3 reuse)


SAIC async MAIC async

70

120

60

100

Throughput gain (%)

50

Throughput gain (%)


2 3 4 5 6 7 8 9 10 11 12

80

40

60

30

40

20

20

10

Effective frequency load (EFL)

10

Effective frequency load (EFL)

15

20

25

30

35

Figure 2: Data throughput and throughput gain for CEQ, SAIC, and MAIC receivers for 3/9 reuse (left) and 1/3 reuse (right) obtained by RN2LL simulator. not indicate the requirement for network synchronization, as far as interference cancellation is concerned. For MAIC, an even higher throughput than for SAIC can be achieved as was expected3 . In Fig. 3 the statistics of the coding scheme usage are depicted for CS-1 to CS-4 for asynchronous networks with 3/9 reuse pattern for CEQ, SAIC, and MAIC (from left to right). EFL=6% has been valid. As expected, when using a conventional receiver the LA has to select a lower coding scheme compared to SAIC/MAIC to suit the current radio channel conditions. EFL = 15%, when grouping the 13 interferers (3 CCI, 3 ACIhigh , 3 ACIlow , and 3 nonfading residual interference [2]) in different ways. It should be noted, that the total interference power is constant in all cases, when assuming an adjacent channel protection (ACP) [2] of 18 dB. For the CEQ, a receiver input lter with a typical bandwidth of 180 kHz has been used. The rst row of Table 2 is valid for allowing only the strongest CCI to fade. All other interferers have been collected to one residual nonfading CCI. Each ACI has been modelled as an equivalent CCI with an 18 dB lower power (ACP = 18 dB). According to the table, the throughput differences when resolving more and more cochannel interferers are only moderate. However, when resolving the strongest adjacent channel interferer, a signicant degradation can be observed which becomes even worse when the strongest adjacent channel interferers at both sides are considered. Surprisingly, this holds for all three receivers, especially also for the CEQ. These results conrm that a simplied interface between radio network and link level (and vice versa in the form of a simple mapping table) is not sufcient in order to obtain a precise performance estimate.

Figure 3: Relative coding scheme (CS) usage for CEQ, SAIC, and MAIC receivers (from left to right). In the following, we investigate the inuence of different transitions from RN to LL (RN2LL interface). Table 2 shows the GPRS data throughput for all three receivers for the 1/3 reuse (synchronous mode) and an
3 The principal shape of the curves in Fig. 2 agree with the results of [9, Chapter 7].

6. Conclusions
In this paper, the effect of the SAIC/MAIC algorithm of [11, 12, 13] on GPRS capacity has been analyzed using a top-down (RN2LL) simulation approach. The main benet of such a simulation methodology in contrast to

Scenario CCI1, CCIres CCI1-5, CCIres CCI1, CCIres , ACI1high CCI1, CCIres , ACI1high , ACI1low CCI1-3, CCIres , ACI1-3high , ACIres,high ACI1-3low , ACIres,low

CEQ [kbit/s] 8.42 8.44 6.58 5.60 5.23

SAIC [kbit/s] 9.48 9.58 8.21 7.05 6.90

MAIC [kbit/s] 13.16 13.64 12.21 11.46 10.62

Table 2: GPRS throughput for different interference congurations for CEQ, SAIC, and MAIC (1/3 reuse, synchronous network, EFL = 15%, no use of TSC for interferers).

LL2RN simulations is that the interference situation can be taken into account precisely, especially adjacent channel interference. This enables to get simulation results close to reality and may be helpful to develop a suitable interface for the LL2RN approach which does not exist yet for MAIC. The following conclusions can be drawn from the numerical results: Very similar gains can be expected for both synchronized and unsynchronized networks. The throughput gains for unsynchronized networks are only approximately 2% smaller than for the synchronized case. This might be a very encouraging result for operators with existing unsynchronized networks to introduce SAIC/MAIC. SAIC provides substantial gains in data throughput for GPRS compared to a conventional receiver. Throughput gains roughly in the order of 20 to 35% can be expected. SAIC starts to become effective even at low effective trafc load. The used MIC algorithm for SAIC proves to be very robust to real world effects like amplitude and phase jumps at burst boundaries, multiple simultaneous interferers, etc. A further dramatic increase in throughput can be achieved by multiple antenna terminals with enhanced signal processing algorithms like MAIC. For the considered network parameters the gains in data throughput can be up to 120% compared to a single antenna terminal with a conventional receiver.

[4] Example Link Level Gains of MS Receive Diversity. Technical report, Nokia, GP-052125, GERAN#26, Schaumburg, USA, 2005. [5] Link level performance of MS RX diversity. Technical report, Philips, GP-051901, GERAN#26, Schaumburg, USA, 2005. [6] S.L. Ariyavisitakul, J.H. Winters, and N.R. Sollenberger. Joint equalization and interference suppression for high data rate wireless systems. IEEE Journal on Selected Areas in Communications, 18:1214 1220, July 2000. [7] S. Brueck, H.-J. Kettschau, and F. Obernosterer. Emission Reduction and Capacity Increase in GSM Networks by Single Antenna Interference Cancellation. International Journal of Electronics and Communications, 58:274283, April 2004. k, S. J [8] A. Furusk ar, M. H oo averbring, H. Olofsson, and J. Sk old. Capacity Evaluation of the EDGE Concept for Enhanced Data Rates in GSM and TDMA/136. In Proceedings of the IEEE Vehicular Technology Conference, pages 16481652, Houston, TX, May 1999. [9] T. Halonen, J. Romero, and J. Melero. GSM, GPRS and EDGE Performance: Evolution towards 3G UMTS. 2nd Edition, John Wiley & Sons, New York, USA, 2003. [10] M. Hata. Empirical formula for propagation loss in land mobile radio services. IEEE Transactions on Vehicular Technology, 29, 1980. [11] R. Meyer, W. Gerstacker, and R. Schober. Method for Cancelling Interference during TDMA Transmission and/or FDMA Transmission. International Patent Application WO02/054660A1, PCT/EP01/15019 (granted), December 2000. [12] R. Meyer, W.H. Gerstacker, R. Schober, and J.B. Huber. A Single Antenna Interference Cancellation Algorithm for GSM. In Proceedings of the IEEE Vehicular Technology Conference, Stockholm, Sweden, May/June 2005. [13] R. Meyer, W.H. Gerstacker, R. Schober, and J.B. Huber. A Single Antenna Interference Cancellation Algorithm for Increased GSM Capacity. IEEE Transactions on Wireless Communications, to appear, 2006.

Acknowledgement
We would like to thank Philips Semiconductors, Nuremberg, Germany, for valuable input and suggestions.

REFERENCES
[1] Radio transmission and reception. 45.005. 3GPP TS

[2] Feasibility Study on Single Antenna Interference Cancellation (SAIC) for GSM Networks (Release 6). 3GPP TS 45.903, 2004. [3] Interference Performance of Conventional Receivers. Technical report, Philips, GP-041045, GERAN#19, Cancun, Mexico, April 2004.

[14] M. Moisio and K. Ashan. The Effect of SingleAntenna Interference Cancellation on GPRS Performance. In Proceedings of the International Symp. on Wireless Communication Systems, pages 15, Mauritius, September 2004. [15] M. Moisio and S. Nikkarinen. Capacity Gains of Single Antenna Interference Cancellation in GSM. In Proceedings of the IEEE Int. Symp. on Personal, Indoor, Mobile Radio Comm. (PIMRC), pages 27002704, Barcelona, September 2004. [16] A. Mostafa, R. Kobylinski, and M. Austin. Interference Cancellation for Downlink GSM Network. In Proceedings of IEEE Vehicular Technology Conference (VTC-Fall 2003), pages 10891093, Orlando, FL, October 2003. [17] P.A. Ranta, H. Berg, E. Tuomaala, and Z. Uykan. Dual Antenna Diversity Reception for EGPRS Terminals. In Proceedings of IEEE Vehicular Technology Conference, pages 510 514, Milan, May 2004. [18] 3G Americas SAIC working group (Editor: M. Austin). SAIC and Synchronized Networks for Increased GSM Capacity. White paper, 3G Americas, Bellevue, WA, September 2003. [19] J. Zander and S. Kim. Radio Resource Management for Wireless Networks. Artech House, Boston, 2001.

Você também pode gostar