Emission Reduction and Capacity Increase in GSM Networks by Single Antenna Interference Cancellation

Emission Reduction and Capacity Increase in GSM Networks by Single Antenna Interference Cancellation

¨ 58 (2004): 274–283 Int. J. Electron. Commun. (AEU) http://www.elsevier.de/aeue Emission Reduction and Capacity Increase in GSM Networks by Single A...

540KB Sizes 0 Downloads 12 Views

¨ 58 (2004): 274–283 Int. J. Electron. Commun. (AEU) http://www.elsevier.de/aeue

Emission Reduction and Capacity Increase in GSM Networks by Single Antenna Interference Cancellation∗ Stefan Brueck, Hans-Juergen Kettschau, and Frank Obernosterer Abstract: There is an increasing demand to utilize the frequency spectrum of mobile communication systems most efficiently. This means in particular to GSM networks that the frequency reuse shall be planned as low as possible. In this case the system may become limited by interference rather than coverage. One promising technology for GSM mobiles in interference-limited systems is single antenna interference cancellation (SAIC). This receiver technology allows both for increasing the network capacity and for reducing the base station transmit power. The aim of this paper is to assess the emission reduction as well as the system capacity capabilities when SAIC technology is applied in downlink receivers. Keywords: Base station power emission, System capacity, Single antenna interference cancellation, MAIO management, GSM network

1. Introduction There is an increasing demand to utilize the frequency spectrum of a mobile communication system most efficiently. For voice applications, the number of served users, and for data services the data throughput is of primary interest. Common to both criteria is that the spectral efficiency of the system, measured in bit/s/Hz/km2 , is maximized. Typically, for FDMA based systems like GSM a frequency reuse of a certain cluster size is applied, depending on the system itself and the propagation environment [1]. Optimizing spectral efficiency means to GSM networks that this frequency reuse is minimized both for traffic (TCH) and broadcast channels (BCCH). A long list of technologies is available for second-generation (2G) mobile communication systems to increase their performance. Among them are Tx and Rx diversity [2], frequency hopping [3], fractional loading [4] and many others. Also technologies based on spatial diversity are applicable [5]. Many of those intelligent antenna solutions require multiple receive antennas and, hence, are mostly used in uplink direction (UL). In the downlink (DL), the limited equipment size at receiver site prevented multiple

Received September 1, 2003. Revised January 13, 2004. Lucent Technologies, Thurn-und-Taxis Strasse 10, 90411 Nuremberg, Germany (E-mail: sbrueck,hkettschau,[email protected]). Correspondence to: S. Brück. ∗ This work was supported by the German Ministry of Education and Research (BMBF) under grant 01BU161.

antenna solutions to be exploited in the past. Consequently, the downlink is the limiting direction in many mobile communication systems, because no advanced receiver technologies could be applied in the mobile so far. Today the limitation in most of the existing GSM networks is due to coverage problems, because the same frequency band is reused with a large spatial distance. Hence co-channel interference is negligible in many of today’s GSM systems. However, for future optimized networks the system capacity is expected to be interference-limited, if the frequency reuse is reduced to very low cluster sizes. This stimulated an increasing interest in interference cancellation or suppression techniques for FDMA based systems in the last couple of years [6]. Most of these algorithms are based on multiple receive antennas and may not be directly applicable in downlink direction. State-of-the-art technology in GSM like frequency hopping and Mobile Allocation Index Offset (MAIO) management allows for reducing the frequency cluster size to three or even to one in selected environments. But the achievable spectral efficiency is still rather low, because of downlink receiver limitations. This downlink bottleneck in existing GSM systems led to intensive research on receiver technologies for interference suppression based on a single receive antenna only, called Single Antenna Interference Cancellation (SAIC). Meanwhile, first contributions on SAIC are made to the standardizations bodies [7], [8]. Herein significant performance gains at link level are presented. These results stimulated a large industrial interest. A feasibility study for introducing SAIC in GSM/EDGE networks has already started in January 2003 within GERAN [9]. First capacity investigations are also already available, based on estimations [10] and field tests for a selected network [11], [12]. The objective of this paper is to investigate potential SAIC performance gains more principally by means of simulations. Our focus is on the reduction of the radiated base station power as well as on system capacity enhancements in terms of the number of simultaneously active voice users. Here, we merely focus on GSM downlink transmission. It will be shown that SAIC technology allows for reducing the average transmit power of a base station significantly for a given number of served users without violating the quality of service defined by the block error rate (BLER). A detailed investigation for different scenarios will be given. Alternatively to reducing the transmit power, the capacity of a GSM system can be increased considerably compared to conventional downlink equalizers, if the full amplifier capability of the 1434-8411/04/58/04-274 $ 30.00/0

S. Brueck et al.: Emission Reduction and Capacity Increase in GSM Networks by Single Antenna Interference Cancellation 275

base station is used. Expected capacity gains will also be shown. The outline of this paper is as follows: In Section 2 a cellular model used for all simulations and its main parameters are introduced. Terminologies used in later sections are defined and explained. Also frequency hopping and MAIO management are repeated, because of their importance for reducing the frequency reuse to very low cluster sizes. In interference-limited FDMA based networks the co-channel interference stems from several independent neighbor cells. Therefore in Section 3 an interference model is introduced that reflects the fact that typically several sources of interference are present. Physical layer simulations of SAIC for different interference models are presented. The coupling of physical layer and system level simulations is an important key for predicting accurate results. In Section 4 a so-called instantaneous value interface between physical layer and system level simulation is presented for SAIC. System level simulation results for the reference scenario are given in Section 5. Both the reduction of average base station transmit power and possible capacity gains are investigated, if SAIC technology is applied in mobile phones. Section 6 concludes the investigation with a summary and an outlook to currently ongoing work within GERAN.

2. Cellular aspects in GSM networks As already mentioned in the introduction, there is an increasing demand to enlarge the spectral efficiency of 2G systems. Random frequency hopping and MAIO management allow for reducing the frequency reuse of the TCH layer in GSM networks significantly. A practically achievable TCH frequency reuse is 1/3 or even 1/1 in selected environments, if enough spectrum is available1 . Such an aggressive frequency reuse requires the mitigation and reduction of both co-channel and adjacent channel interference. Frequency hopping is briefly repeated in the next subsection. Further we describe MAIO management and show how it can be applied in order to lower both cochannel and adjacent channel interference. 2.1 Frequency hopping and MAIO management Frequency hopping is one common means in GSM to improve the system performance. For each GSM burst of duration of 577 µs a new carrier frequency is chosen. The sequence of frequencies to be used is derived from 1 We assume in this paper that each site consists of three sectors. The notation of reuse x/y means that x sites and y sectors within x sites use different carrier frequencies. For example, a reuse of 1/3 means that all sites share the same spectrum, but within each site the frequencies are divided in three distinct parts used in three sectors. Sometimes in literature the terminologies cell and sector have a different meaning. In this paper both expressions denote the same sectorized area of a site.

an algorithm specified in [13]. The input parameters to this algorithm are the number of hopping frequencies N, the Frame Number (FN), the Hopping Sequence Number (0 ≤ HSN ≤ 63) and the Mobile Allocation Index Offset (0 ≤ MAIO ≤ N–1), which are all under control of the operator. A definition of these parameters is given in [13]. The output of the algorithm is the Mobile Allocation Index (MAI), which is a function of these parameters. The MAI is an integer number within the range {0, · · · , N − 1}. The MAI states which frequency has to be used for each burst. Each MAI is allocated uniquely to a certain frequency. Frequency Hopping is further separated into two types: Cyclic Hopping (HSN = 0) and random hopping (0 < HSN ≤ 63). For random hopping the generation of the current MAI is almost independent on the previously generated indices. That means that the number of different frequencies used within 8 hops (length of a TCH/FS frame) is generally not as large as for cyclic hopping. Even the same frequency may occur several times in a consecutive order. On the other side the interference is caused by different sources for each hop in case of random hopping. Thus extreme good or bad interference situations during a longer period of time are avoided in contrast to cyclic hopping. Hence compared to cyclic hopping, random frequency hopping provides less frequency diversity [2], but improved interference diversity in general [14]. In network planning, a MAIO is required for each transmitter (TRX), and a HSN needs to be allocated to each sector. The FN is a base station internal counter, incremented in each frame. The selection of these parameters impacts the network performance significantly. The most important properties of the hopping algorithm in [13] are: • Property 1: Two TRXs with the same HSN and the same FN but different MAIOs, never use the same frequency at the same time. • Property 2: The difference of the MAIOs translates into a shift (modulo the MAIO difference) of the generated MAIs and, therefore, of the frequencies. • Property 3: Two TRXs, using the same frequencies but a different HSN > 0 (random hopping), will interfere randomly. The collision probability is about 1/N, where N is the number of hopping frequencies. These properties can be used for the generation of orthogonal hopping sequences in order to avoid cochannel interference and for the reduction of adjacent channel interference. Such algorithms are typically called MAIO management. It can be distinguished between inter-site and intra-site MAIO management. The principles are explained in more detail, when basic terminologies have been introduced in the next subsection. 2.2 Cellular system model and its parameters In this subsection definitions and assumptions required later are introduced. Further a reference cellular model is presented, which is used for all following investigations.

276 S. Brueck et al.: Emission Reduction and Capacity Increase in GSM Networks by Single Antenna Interference Cancellation We assume for the reference model that a total frequency spectrum of 5 MHz is available for the GSM network. This spectrum is divided into twenty-five 200 kHz channels. We further assume that the BCCH layer is planned with a reuse of 4/12. This means that twelve sectors in four sites require a unique 200 kHz channel for transmitting the BCCH. Typically, one 200 kHz channel is spent as guard band between TCH and BCCH layer or to adjacent frequency bands. Then twelve frequencies are still available for the TCH layer. For this number of frequencies being available for the TCH layer it is known from practical experience and simulations that in certain propagation scenarios the TCH layer can be planned with reuse 1/1 and three TRXs per sector, when random frequency hopping is applied and a channel activity factor of 50% can be assumed2 . All twelve TCH frequencies are common to all sectors, each TRX hops over all frequencies. Because only three TRXs are installed in each sector, only three out of twelve different frequencies are on air for each transmission period of 577 µs. This ratio is denoted as Hardware Fractional Load (HFL) of 3/12. Generally, the HFL is defined as ∆

HFL =

No. of TRX/Sector . No. of Frequencies/Sector

(1) Table 1. Simulation Parameters.

The probability how often a carrier frequency is effectively on air, depends on the carried traffic. It is a measure for the expected interference and the required frequency cluster size in the network. A widely used definition is the Erlang Fractional Load (EFL) ∆

EFL =

carried Traffic/Sector . 8 · No. of Frequencies/Sector

(2)

The EFL denotes the average usage of a frequency per time slot. This is also a measure for the capacity of the system. The larger the EFL for a given number of frequencies the more voice users are served simultaneously. An improvement in EFL is directly convertible in a gain of the system capacity in Erlang/sector. We now want to describe MAIO management in more detail. Inter-site and intra-site MAIO management are based on the HFL the cells are planned for. Inter-site MAIO management applies cyclic hopping (HSN = 0) and, thus, leads to a large frequency diversity. Reuse of the same MAIO with a sufficiently large spatial distance is possible. For very low HFL values property 1 of the hopping algorithm can be applied to avoid co-channel interference even between sites. Note that this requires time synchronization between sites. For a HFL larger than 1/3 it is not possible to avoid collisions between frequencies within one site. Therefore, inter-site MAIO management generally assumes a HFL of less than 1/3 for a three-sector site. Intra-site MAIO management 2

applies random frequency hopping (HSN > 0). Here only co-channel interference within a site is suppressed. Because transmission within a site is always synchronous, no synchronous network operation is required. Because of random collisions, intra-site MAIO management is also applicable, if the HFL is larger than 1/3. Due to property 2 both MAIO management strategies can be used to reduce adjacent channel interference. For a target HFL of 3/12 with frequency reuse 1/1 we apply intra-site MAIO management in this paper. Such a HFL allows to avoid co-channel interference within all sites and to reduce adjacent channel interference significantly. The most important parameters of the cellular model are summarized in Table 1. The system model holds for the TCH layer in the 900 MHz band. The BCCH layer is neglected in the following investigations. Adjacent channel interference is modelled only within the TCH layer by a reduction of the interference level by 18 dB, if the interferer occupies the directly adjacent frequency channel. This is according to the technical requirement given in [15]. For simplicity, channel occupancy of the stand-alone dedicated control channel (SDCCH) has been neglected,

The channel activity factor is slightly larger than the DTX factor.

Parameter

Value

Carrier Frequency

900 MHz

Power Amplifier

43 dBm

Voice Codecs

AMR 12.2, AMR 5.9 AMR 4.75

Frequency Plan

Reuse 1/1, HFL 3/12 Reuse 3/9, HFL 3/3

Channel Activity

50%

Offered Traffic

variable

Traffic Model

Mean call duration = 90 s Poisson distribution of arrival process

Cell Layout

1 serving cell 56 interference cells 3 sectors/site Cloverleaf cell planning

Antennas

60 deg. beam width

Shadow Fading

Constant per call 6 dB variance

Downlink Noise Floor

−115 dBm

Adjacent Channel Interference

18 dB reduction

Power Control

RxQual, RxLev based 1 dB granularity

Frequency Hopping

Uncorrelated fast fading

Physical Channel

TU50 FH, GMSK

Site-to-Site Distance

1500 m

Propagation Model

21 + 35 · log10 (d/m)

S. Brueck et al.: Emission Reduction and Capacity Increase in GSM Networks by Single Antenna Interference Cancellation 277

and all twenty-four time slots of the three TRXs are assumed to be available for traffic. Under this assumption, an offered traffic load of 16.6 Erlang/sector leads to a blocking probability of 2% according to the Erlang-B formula [16]. It has further been assumed that the frequency diversity due to frequency hopping is ideal. This requires that the carrier frequencies experience stochastically independent fast fading, i.e. the separation between the frequencies is larger than the coherence bandwidth of the channel [2]. Because of random hopping over twelve frequencies this assumption is rather well justified. The path loss exponent is 3.5 and the distance between two sites has been chosen to be 1500 m. This assumption corresponds to a typical urban environment. The physical layer results are included in the system level simulation tool by look-up tables. Because they already take the impact of fast fading into account, it is not considered at system level anymore. The parameters given above are used in all subsequent investigations. It is further assumed that all base stations are synchronized to each other. In general, synchronization improves the performance of the physical layer algorithms in GSM. But it is known that bad cross correlation properties of certain training sequences may cause difficulties for signal identification. Proposals to overcome this problem in synchronous network operation are given in [17].

3. Performance gains of SAIC at physical layer For a TCH frequency reuse of 1/3 or even of 1/1, cochannel interference between cells cannot be neglected anymore. It may even be the case that the coverage limitation has been changed into an interference limitation. According to [15], co-channel interference in GSM is typically simulated by one single GMSK modulated interference source only. This reflects the situation in a real world scenario insufficiently and leads to physical layer performances, which cannot be met in a cellular scenario. In order to adapt the interference model better to the statistical nature of the co-channel interference, we approximate the total co-channel interference by four stochastically independent interference sources. In this model the overall interference is given by the following expression Itotal = Idominant + Iresidual ,

Iresidual =

3 

I j , E(I1 ) = E(I2 ) = E(I3 ) .

ratio E(Idominant )/E(Iresidual ) are variable. All interference sources are GMSK modulated. In the following link level results will be presented for the interference model taken from [15] and the interference model consisting of four independent sources, both for a conventional equalizer and SAIC being applied in downlink receivers. The main difference between both algorithms is that a conventional equalizer treats interference simply as additional background noise, whereas any SAIC algorithm tries to exploit or reduce this interference. We investigate a SAIC approach based on blind interference cancellation techniques, similarly exploiting inherent properties of GMSK modulation as described in [18]. The implementation of both algorithms fulfils practical restrictions, encountered in real world mobiles. A brief general description of SAIC methods can be found in [19]. Because the downlink is considered, no Rx diversity is assumed in all simulations. First the results are given for the standard model with only one source of interference. In Figure 1 a typical urban channel with a velocity of 50 km/h and ideal frequency hopping has been considered for AMR 12.2 and AMR 4.75 codecs. The rates of the channel encoder are R = 1/2 for AMR 12.2 and R = 1/5 for AMR 4.75, respectively. The BLER is given for the class Ia bits of the speech codec. It is seen that SAIC leads to an improvement of more than 10 dB compared to a conventional equalizer at a reference point of 1% BLER, which is a typical value for good voice quality. Further it is seen that AMR 12.2 with SAIC outperforms AMR 4.75 with a conventional equalizer. This indicates that the SAIC algorithm compensates the improved error correcting capabilities of the R = 1/5 encoder being used for the AMR 4.75 codec. Figure 2 considers a typical urban channel with only 3 km/h without frequency hopping. In this case neither SAIC nor the conventional equalizer benefits from frequency diversity. It is seen in the figure that the gains are even larger in this case. At a reference point of 2% BLER

(3)

(4)

j=1

E(·) denotes the expected value. The residual power is the sum in Watts of three independent interference sources with the same mean power but independent fast fading. The mean power of the dominant interferer and the

Fig. 1. BLER for TU50 with frequency hopping, standard interference model.

278 S. Brueck et al.: Emission Reduction and Capacity Increase in GSM Networks by Single Antenna Interference Cancellation

Fig. 2. BLER for TU3 without frequency hopping, standard interference model.

the gain due to SAIC is about 12 dB for an AMR 12.2 codec. Due to the properties of the SAIC algorithm its performance is very good, if an interference model consisting of one interferer only is used. It has already been mentioned that in a real world scenario with low frequency reuse the co-channel interference stems from many cells. Thus it can be expected that the gains due to SAIC are reduced, if the interference model consisting of four interference sources is used. Figure 3 shows the performance of SAIC for this model with E(Idominant )/E(Iresidual ) given in dB as parameter for a full rate (FR) codec. Here the most important property of SAIC becomes evident. If only one dominant interferer is present, the gain due to SAIC is largest. For the TU50 channel model, the improvement is about

Fig. 3. BLER for TU50 without frequency hopping, improved interference model.

12 dB. If the received powers of the interferers become more equal, the gains are reduced. But it can be noticed that even for very low E(Idominant )/E(Iresidual ) ratios a gain of about 2.5 dB is always achievable. This is due to the assumption of independent fast fading of the four sources of interference, which is well met in reality. In this case it is very likely that one of the interferers is stronger than the others for each short period of time. This fact is of benefit to the SAIC algorithm. It is most important to note that the SAIC gains are dependent on the instantaneous strength of the individual interferers. They are largest, if only one dominant interferer is present. Such an interference situation cannot be expected in a fully loaded system with low frequency reuse. In such a scenario the interference powers are more equally distributed and the gains due to SAIC are reduced. Because of the stochastically independence of the interference sources a gain is still obtained even if all interferers disturb with the same average power.

4. Interface between physical layer and system level simulations If physical layer performance indicators like BLER or signal to noise ratio are of interest at system level, the coupling between physical layer and system level simulations is of considerable importance. This is in particular true, if the signal to interference ratio varies considerably within one TCH frame, as it is the case, when frequency hopping is applied. Therefore we introduce an interface being used in the following system level simulations. Furthermore it is explained, how the dependence of the SAIC performance on E(Idominant )/E(Iresidual ) is modelled. Interfaces between these two types of simulations are often called actual or Instantaneous Value Interface (IVI), when the instantaneous signal to interference ratio at burst level is used instead of a value, averaged over a longer period of time (Averaged Value Interface, AVI). Examples of such interfaces are given in [20], [21]. The goal of the IVI is to calculate an effective signal to interference ratio that approximates the variations of the signal to interference ratios within each frame or codeword reliably. This effective signal to interference ratio is typically further used for finding the instantaneous BLER by using (BLER, C/I) look-up tables. The approach being used in this investigation has been derived in [22]. Here only a brief summary is given. The concept is based on union bound techniques for convolutional codes [23]. Assuming the signal to interference levels within a codeword are independent from burst to burst, it has been shown in [22] that an accurate effective signal to noise ratio can be calculated by       C C −1 =Z E Z . (5) I eff Ij Z(·) is a channel dependent function, called Bhattacharyya function [23]. The expectation E(·) has to be taken over

S. Brueck et al.: Emission Reduction and Capacity Increase in GSM Networks by Single Antenna Interference Cancellation 279

all signal to interference variations. Taking into account that a TCH frame consists of eight bursts, we use the following approximation for calculating the effective signal to interference ratio       8  C C 1 . ≈ Z −1  · Z (6) I eff 8 j=1 Ij As already mentioned the function Z(·) is channel dependent and does not have a simple analytical form for frequency-selective fading channels. Instead we use an approximation of the form 

1 A1 · x Z(x) = · 1 − , 2 1 + A2 · x

(7)

where the constants A1 and A2 are found by simulations. With A1 = A2 = 1 this function is equivalent to the error probability of BPSK transmission over a Rayleigh fading channel [2]. This approach allows for estimating the resulting BLER easily and with good accuracy by look-up tables. The described calculation of the effective signal to interference ratio is based on the C/I ratio at the decoder input, i.e. after the equalizer or the interference canceller. In the previous section it has been demonstrated that the performance gain of SAIC depends on the instantaneous power ratio of the dominant and the residual interferers. This dynamical behavior has to be modelled at system level properly in order to predict the benefits of SAIC in a cellular scenario. Figure 4 shows the average link level gain (taken from Figure 3) of SAIC compared to a conventional equalizer in dependence on the ratio between dominant and residuals interferers for a reference BLER of 1%. This curve is approximated by a slightly conservative step-function as

shown in Figure 4, which is used in the system level simulator. In each time step of the system level simulator the instantaneous ratio between the dominant and the residual interferers is calculated, and the corresponding link level gain from Figure 4 is used as a correction term of the (BLER, C/I) look-up tables for a conventional equalizer. The overall methodology for coupling physical and system level simulations is as follows: • The time resolution of the system level simulation tool is one GSM burst, i.e. 577 µs. For each block of eight bursts of a TCH frame the instantaneous signal to interference ratio is calculated. • The instantaneous ratio between the dominant and the sum of all residual interferers is calculated for each burst, and the correction term in case of SAIC receiver is applied, taken from Figure 4. • These values are mapped by the instantaneous value interface to an effective signal to interference ratio using equations (6) and (7). • The resulting BLER for this TCH frame is found in the look-up table, which has been produced in advance by link level simulations. • Finally, a coin with this BLER as fairness is thrown to decide, whether the frame is received correctly.

5. Performance gains of SAIC in a cellular scenario In the following the average base station transmit power and the capacity improvements gained by SAIC shall be investigated. In all simulations the reference model, given in Table 1, has been used. Synchronous network operation has been further assumed.

5.1 Average base station transmit power

Fig. 4. SAIC performance gains.

The TCH layer is considered in a reuse 1/1 configuration with three TRXs per cell. An offered traffic of 16.6 Erlang/sector is assumed leading to a blocking probability of 2%. As a measure for voice quality the average BLER per call is used. Figure 5 depicts the mean base station transmit power for a certain target average BLER/call both for a conventional equalizer as well as for a SAIC receiver, being applied in the mobile for a typical urban channel model with frequency hopping and a velocity of 50 km/h. The maximal base station power has been fixed to 43 dBm. It is seen that application of SAIC in the mobile reduces the required transmit power significantly. For a target of 1.5% BLER/call the mean power can be reduced by about 4.7 dB. For a very good voice quality of less than 1% BLER the mean power is larger than 40 dBm, when a conventional equalizer is used. This indicates that an amplifier larger than 43 dBm is required

280 S. Brueck et al.: Emission Reduction and Capacity Increase in GSM Networks by Single Antenna Interference Cancellation

Fig. 5. Power emission reduction with SAIC for TCH reuse 1/1 and TU50 FH.

Fig. 6. Power emission reduction with SAIC for TCH reuse 3/9 and TU50 FH.

due to the variance in transmit power, if the users shall be served with a mean BLER/call of 1%. A maximal amplifier power of 43 dBm only would result in overheating and subsequent thermal degradation leading to lowered amplifier reliability. This figure also shows that 16.6 Erlang/sector is the critical load for the reference network, if all mobiles apply conventional equalization. Further increase of the number of users would result in larger interference and, hence, larger BLER. The load cannot be enlarged, if no loss in voice quality can be tolerated. In contrast, a load of 16.6 Erlang/sector is not critical to the network in terms of voice quality, if SAIC is applied in all mobile receivers. Now we assume a frequency reuse of 3/9. Reuse 3/9 means that in three adjacent sites with nine sectors different frequencies are applied. The overall bandwidth is assumed to be 8 MHz in this example now. Thirteen frequencies again are reserved for BCCH and the guard band, i.e. twenty-seven frequencies are left for the TCH layer and frequency hopping. Because the frequency reuse is 3/9, three hopping frequencies are available in each sector. The number of TRXs per sector shall again be three, i.e. the HFL is 3/3. The offered traffic is again 16.6 Erlang/sector. Figure 6 depicts the average transmit power for a TCH reuse of 3/9 for the same channel model as in Figure 5. In this case the gain by SAIC is reduced, because cochannel interference is less severe in a reuse 3/9 system. For a mean voice quality of 1% the gain is about 2 dB in average transmit power.

a traffic load of 16.6 Erlang/sector is the maximal capacity in the chosen environment, which can be tolerated with sufficient voice quality and conventional equalization. With SAIC a lower saturation level in the mean BLER is recognized. The system load can still be increased without decreasing the voice quality of each individual call. Now it shall be shown, what capacity gains can be expected by SAIC. In a previous section the EFL has been introduced as a measure for the system capacity. If the EFL increases in an interference-limited system, the quality of the voice service in terms of average BLER shrinks, because the interference becomes more severe. Alternatively it can be said that the number of users, which still experience a mean BLER/call of 1%, is reduced. Customer needs typically demand a BLER/call of 1% or better for at least 95% of the users in each sector. The blocking probability shall typically not exceed 2%. Figure 7 shows the number of users with a mean BLER/call of 1% depending on the EFL for downlink direction and AMR codecs 12.2 and 5.9. The blocking probability is 2% or less in all cases. It is seen that about 95% can be served with a BLER/call of 1%, if the EFL is about 12% and if all mobiles apply conventional equalization with AMR 12.2. If we again assume that twelve frequencies are available for the TCH layer, this EFL corresponds to a carried traffic of about 11.5 Erlang/sector according to equation (2). Increasing the traffic to 16.6 Erlang/sector does not allow 95% of the users to be served with such a good voice quality. This shows again that the corresponding HFL of 3/12 is the capacity limit in the reference model for conventional equalization. It is further observed in the figure that SAIC allows for increasing the EFL to about 21% for the same grade of satisfied users with AMR 12.2. Compared to conventional

5.2 System capacity enhancements with SAIC With conventional equalization the mean BLER reaches its saturation level at 1% in Figure 5. This reveals that

S. Brueck et al.: Emission Reduction and Capacity Increase in GSM Networks by Single Antenna Interference Cancellation 281

6. Summary and outlook

Fig. 7. Achievable Erlang fractional load for different AMR codecs.

equalization this is a gain of 75%. According to the definition of EFL, this gain is directly convertible into the same increase in Erlang/sector with SAIC. Figure 3 also shows that the capacity can be further enhanced for both receiver technologies, when an AMR 5.9 speech codec is applied. The SAIC gain is not as large as compared to AMR 12.2 due to the increased interference robustness of an AMR 5.9 codec. For AMR 5.9 an EFL of 55% is achievable with SAIC according to this figure. Assuming again twelve hopping frequencies, this EFL corresponds to a carried traffic of 52.8 Erlang/sector according to equation (2). For a required blocking probability of 2%, eight TRXs per sector were necessary according to the Erlang-B formula [16]. The HFL were 8/12 in this case. The gain in EFL due to SAIC allows a further interpretation, namely that the required bandwidth can be reduced. About the same EFL of 21% for AMR 12.2 is achieved, if the traffic is kept fixed at 11.5 Erlang/sector, but if now only seven hopping frequencies are used instead of twelve frequencies as before3. This means that the number of frequencies could by reduced from twelve to seven with SAIC in order to achieve the same capacity with the same grade of satisfied users. This corresponds to a bandwidth reduction of 42% for the TCH layer4. The above simulation results have shown that SAIC both allows for increasing the number of users per sector or for reducing the required frequency spectrum. In both cases the spectral efficiency of an interference-limited GSM system is enlarged significantly.

The EFL is: 11.5/(7 · 8) = 0.205 Here it has been assumed that the BLER performance is unchanged in the reduced spectrum, i.e. correlation effects of fast fading are still negligible and do not lead to a larger requirement in the signal to interference ratio. 3 4

Single antenna interference cancellation is a very promising technology to increase the spectral efficiency of GSM networks. In this paper the expected performance gains in terms of power reduction and capacity enhancements have been investigated. First the performance and the cancellation properties of SAIC have been considered at link level. These results were used for the definition of a more realistic interference model for GSM networks with low frequency reuse compared to the one stated in [15]. Further an instantaneous value interface for link level and system level simulations has been derived in order to reflect the dynamical properties both of frequency hopping and SAIC. These concepts were used in a system level simulation tool with very high time resolution in order to demonstrate the capabilities of SAIC in terms of base station power reduction and system capacity improvement. As reference scenario such a traffic load and HFL have been chosen in a reuse 1/1 system, which are known to be the maximal system capacity for the reference model, when conventional equalization is applied together with speech codec AMR 12.2. It has been demonstrated that SAIC is able to reduce the average base station transmit power by about 4.7 dB in this scenario for a given mean BLER/call. In less interference-limited scenarios this gain is reduced, but still present. With frequency reuse 3/9 a reduction in average transmit power of about 2 dB has still been observed. These gains can either be used to increase the cell radii, i.e. for reducing the number of base stations in the network deployment, or to reduce the sizes of the power amplifiers. It was further seen that the mean BLER/call could not be reduced below 1% in the reference scenario with conventional equalizers, which is the critical target for good voice quality in mobile communications. With SAIC technology a much lower error floor was observed. This showed that SAIC increases both the system capacity and voice quality considerably. As measure for the capacity and frequency usage of a network the EFL has been used. Simulations have shown that the EFL can be increased by SAIC up to 75% without violating the grade of satisfied users. This corresponds to a capacity increase of 75% in Erlang/sector. This expected gain is in line with the results given in [7]. Alternatively, SAIC allows for reducing the required frequency spectrum considerably, if the offered traffic is kept the same both for SAIC and conventional equalization. In the simulations it has been observed that about 42% less TCH spectrum is needed to serve the same amount of users, while maintaining the same voice quality.

6.1 SAIC activities in GERAN For completeness of the paper we summarize briefly ongoing GERAN activities. In the mean time SAIC has been

282 S. Brueck et al.: Emission Reduction and Capacity Increase in GSM Networks by Single Antenna Interference Cancellation established as a separate work item within GERAN. The goal is the definition of commonly agreed models and the prediction of expected performance gains both for link and system level. Special attention is given to the definition of suitable link level interference models both for synchronous and asynchronous network operation and different system loads [24]. In particular, the number of co-channel and adjacent channel interferers is specified in order to match realistic scenarios closely. First link level results for the synchronous and asynchronous GERAN models have been presented in [25], [26], showing significant gains of SAIC compared to a conventional equalizer. Also system level simulation parameters have been defined for the 900 MHz and 1900 MHz bands. Different bandwidths, cluster sizes and modulation types (GMSK, 8-PSK) for source and interference signals have been taken into account [27]. First results are given for this model in [28]. Performance measurements for a real cellular network are available in [29]. Both contributions show that the expected capacity gain of SAIC is in the order of 40%–60%. Even gains of up to 100% seem to be achievable.

References [1] Walke, B.: Mobilfunknetze und ihre Protokolle – Band I. B.G. Teubner, 1998. [2] Proakis, J.: Digital Communications. McGraw–Hill, 1995. [3] Carneheim, C.; Jonsson, S.O.; Ljungberg, M.; Madfors, M.; Naslund, J.: FH-GSM Frequency Hopping GSM. IEEE VTC (pp. 1155–1159, 1994). [4] Wigard, J.; Mogensen, P.; Johansen, J.; Vejlgaard, P.: Capacity of a GSM Network with Fractional Loading and Random Frequency Hopping. IEEE PIRMC (pp. 723–727, 1996). [5] Winters, J.H.: Optimum Combining in Digital Mobile Radio with Cochannel Interference. IEEE Journal on Select. Areas in Commun. (pp. 1222–1230, 1984). [6] Lucent Technologies: Diversity Interference Cancellation for GSM/EDGE. 3GPP TSG GERAN#4, Tdoc GP-010772 (Biarritz, April 2001). [7] Ericsson: Downlink GMSK Interference Suppression – Performance Evaluation. 3GPP TSG GERAN#9, Tdoc GP020822 (Seattle, April 2002). [8] Cingular Wireless; Philips Semiconductors; SBC TRI: Single Antenna Interference Cancellation in MS for GSM Networks. 3GPP TSG GERAN#9, Tdoc GP-021013 (Seattle, April 2002). [9] Cingular Wireless; Philips Semiconductors; Nortel; Nokia; Motorola; Ericsson; AT&T Wireless; Intel: Single Antenna Receiver Interference Cancellation (SAIC). 3GPP TSG GERAN#12, Tdoc GP-023400 (Sophia-Antipolis, November 2002). [10] Rysavy, P.: Voice Capacity Enhancements for GSM Evolution to UMTS. 3G Americas (July 2002). [11] Cingular Wireless; Philips Semiconductors: Laboratory & Field Testing of SAIC for GSM Networks. 3GPP TSG GERAN#11, Tdoc GP-022701 (Los Angeles, August 2002). [12] Austin, M. (Ed.): SAIC and Synchronized Networks for Increased GSM Capacity. 3G Americas (September 2003). [13] 3GPP: Multiplexing and Multiple Access on the Radio Path. GSM TS 05.02, ETSI (August 1999).

[14] Olofsson, H.; Naslund, J.; Skold, J.: Interference Diversity Gain in Frequency Hopping GSM. IEEE VTC (July 1995). [15] 3GPP: Radio Transmission and Reception. GSM TS 05.05, ETSI (July 1996). [16] Schwartz, M.: Telecommunication Networks – Protocols, Modelling and Analysis. Addison-Wesley, 1987. [17] Cingular Wireless: Methods to improve Downlink Radio Performance in synchronous Networks. 3GPP TSG GERAN#15, Tdoc GP-031433 (Ft. Lauderdale, June 2003). [18] Trigui, H.; Slock, D.T.M.: Cochannel Interference Cancellation within the current GSM Standard. IEEE ICUPC (October 1998). [19] Kempf, P.; Kalveram, H.: SAIC Method maximizes Voice Capacity of GSM Networks. Wireless Europe (May 2003). [20] Haemaelaeinen, S.; Slanina, P.; Hartman, M.; Lappetelaeinen, A.; Holma, H.; Salonaho, O.: A Novel Interface between Link and System Level Simulations. ACTS Mobile Communications Summit (pp. 599–604, 1997). [21] Olofsson, H.; Almgren, M.; Johansson, C.; Hook, M.; Kronestedt, F.: Improved Interface between Link Level and System Level Simulations applied to GSM. IEEE ICUPC (pp. 79–83, 1997). [22] Brueck, S.: Modeling Interference Diversity in GSM Networks. IEEE VTC (Sept. 2000). [23] Viterbi, A.J.; Omura, J.K.: Principles of Digital Communication and Coding. McGraw-Hill, 1985. [24] Rapporteur: Exemplary Link Level Assumptions, Configuration 2/3. 3GPP TSG GERAN SAIC Adhoc#2, GAHS030025 (Seattle, March 2003). [25] Philips Semiconductors: SAIC Link Performance for Synchronous GERAN Models. 3GPP TSG GERAN#15, Tdoc GP-031514 (Ft. Lauderdale, June 2003). [26] Philips Semiconductors: SAIC Link Performance for Asynchronous GERAN Models. 3GPP TSG GERAN#15, Tdoc GP-031515 (Ft. Lauderdale, June 2003). [27] Rapporteur: SAIC System Simulation Parameters for Characterization of Link Level Scenarios. 3GPP TSG GERAN SAIC Adhoc#1, GAHS-030015 (Atlanta, January 2003). [28] Cingular Wireless: SAIC System Level Evaluations based on GERAN System Simulator. 3GPP TSG GERAN#15, Tdoc GP-031270 (Ft. Lauderdale, June 2003). [29] Cingular Wireless: SAIC System Level Evaluations based on Network Data from a Real Cellular System. 3GPP TSG GERAN#15, Tdoc GP-031271 (Ft. Lauderdale, June 2003).

Stefan Brueck was born in Darmstadt, Germany in 1969. He studied mathematics and electrical engineering at University of Technology Darmstadt, Germany, and Trinity College, Dublin, Ireland. In 1994 he received the Dipl.-Math. degree. From 1994 until 1999 he was with the Institute of Network and Signal Theory at University of Technology Darmstadt. In 1999 he received the Dr.-Ing. degree in electrical engineering from this university. Since 1999 he is with Lucent Technologies, where he is involved in the optimization of GSM and UMTS networks. His main research interest is currently radio resource management for mobile communication networks.

S. Brueck et al.: Emission Reduction and Capacity Increase in GSM Networks by Single Antenna Interference Cancellation 283

Hans-Juergen Kettschau was born in Saarbruecken, Germany in 1954. He studied electrical engineering at University of Saarbruecken, Germany and received his Dipl.-Ing. 1982. In 1982 he joined Philips/PKI Nuremberg, Germany. He worked in the development of telecommunication equipment and simulation software in this area. Since 1995 he is with Lucent Technologies working in the Wireless Research Area. Here he developed several simulation tools for alarm correlation, physical layer both for GSM and UMTS and packet data applications. He holds several patents in the area of data mining.

Frank Obernosterer was born in Nuremberg, Germany, in 1965. He received his Diploma (M.Sc.) degree in electrical engineering in June 1990 and the Dr.-Ing. (Ph.D.E.E.) degree with a thesis on modeling nonlinear distortions in high density digital magnetic recording in 1995 from the University of Erlangen-Nuremberg, Erlangen, Germany. From 1991 to 1998, he was a Research Assistant at the Technical Electronics Institute, University of Erlangen-Nuremberg. In 1995, he was working on nonlinear equalization concepts for digital magnetic recording. Since 1998, he has been with Lucent Technologies, Germany. His research interests include equalization, interference cancellation, and adaptive and nonlinear filtering.