Increased Capacity through Hierarchical Cellular ... - Springer Link
Mobile Networks and Applications 6, 471–480, 2001 2001 Kluwer Academic Publishers. Manufactured in The Netherlands.
Increased Capacity through Hierarchical Cellular Structures with Inter-Layer Reuse in an Enhanced GSM Radio Network ∗ JÜRGEN DEISSNER and GERHARD P. FETTWEIS Dresden University of Technology, Mannesmann Mobilfunk Chair for Mobile Communications Systems, D-01062 Dresden, Germany
Abstract. In today’s cellular networks it becomes harder to provide the resources for the increasing and fluctuating traffic demand exactly in the place and at the time where and when they are needed. Moreover, frequency planning for a hierarchical cellular network, especially to cover indoor areas and hot-spots is a complicated and expensive task. Therefore, we study the ability of hierarchical cellular structures with inter-layer reuse to increase the capacity of a GSM (Global System for Mobile Communications) radio network by applying Total Frequency Hopping (T-FH) and Adaptive Frequency Allocation (AFA) as a strategy to reuse the macro- and microcell resources without frequency planning in indoor picocells. The presented interference analysis indicates a considerable interference reduction gain by T-FH in conjunction with AFA, which can be used for carrying an additional indoor traffic of more than 300 Erlang/km2 , i.e., increasing the spectral capacity by over 50%, namely 33 Erlang/km2 /MHz. From these results we draw a number of general conclusions for the design of hierarchical cellular structures in future mobile radio networks. For example, we may conclude that they require reuse strategies that not only adapt to the current local interference situation, but additionally distribute the remaining interference to as many resources as possible. For a hierarchical GSM network this requirement is fulfilled by the T-FH/AFA technique very well. Keywords: hierarchical cellular structures, Global System for Mobile Communications (GSM), Total Frequency Hopping, adaptive frequency allocation
1. Introduction The provision of capacity for the increasing traffic demand in mobile radio networks comes along with the reduction of the cell size and, hence, stronger traffic fluctuations between the cells. Moreover, improved indoor coverage is required. It becomes harder to provide the resources exactly in the place and at the time where and when they are needed. Hierarchical cellular structures can serve indoor users and hot spots by pico- and microcell layers, respectively, while providing coverage in the area by the macrocell layer. Moreover, hierarchical cellular structures can compensate traffic fluctuations, e.g., by shifting overflow traffic from lower to higher layers. In order to avoid interference between the layers, their frequencies have to be coordinated. However, due to the heterogeneous cell sizes and propagation conditions, frequency planning for a hierarchical cellular network is a complicated and expensive task. That problem is even worse if picocells have to be deployed suddenly or only for short periods, e.g., at fairs, conferences, ad hoc events, for which it is not worth or even not possible to go through the frequency planning cycle. Furthermore, hierarchical cellular structures become a regular feature of future mobile radio networks. Although different multiple access techniques may apply, some experiences from GSM can also be useful for the design of other hierarchical cellular networks, where several layers share the same resources. ∗ This work was supported in part by the Alcatel Corporate Research Cen-
tre, Stuttgart, Germany.
In this paper, we study therefore the ability of hierarchical cellular structures with inter-layer reuse to increase the capacity of a GSM radio network by applying Total Frequency Hopping and Adaptive Frequency Allocation as a strategy to reuse the macro- and microcell resources without frequency planning in indoor picocells. After a discussion of previous work on hierarchical GSM networks in section 2, we describe our simulation environment in section 3. In section 4 we present the results of our comparative simulation study that aims at network configurations in a dense urban environment where a high additional traffic capacity in the picocell layer shall be achieved solely by reusing the microand/or macrocell frequencies. Finally in section 5, we discuss the results and draw a number of general conclusions for the design of hierarchical cellular networks. 2. Discussion of previous work on hierarchical GSM networks Interference reduction techniques like power control, frequency hopping (FH), discontinuous transmission (DTX), partial loading, and adaptive antennas improve the performance of hierarchical GSM networks. And, especially, partial loading in combination with FH can compensate traffic fluctuations to some extent [1]. In the macro- and microcell layer usually different frequencies are used to avoid the coordination effort between the layers. However, in order to not decrease the outdoor capacity of those layers it is not desirable to dedicate a large part of the scarce frequency band of a network operator to the indoor picocell layer. Facing this requirement, a strategy
472
to combat interference between the picocells and the other layers is to use wideband synthesized FH for the traffic channels [2]. Basically, Total Frequency Hopping (T-FH) in picocells is such a solution. T-FH is a slow FH technique which uses more frequencies for FH than typically chosen per base station, e.g., up to the whole frequency band of a network operator [3]. The more frequencies are used the more the interference is spread over all frequencies. The resulting interference diversity effect opens up capacity reserves in the mobile radio network. However, this is not sufficient for an acceptable performance in the picocells [4]. In fact, the performance of T-FH can be much improved by excluding the strongestly interfered frequencies and thus statically reducing the frequency list for T-FH in the picocells or by Adaptive Frequency Allocation (AFA) as well. A reduced frequency list may, for example, exclude the broadcast control channel (BCCH) and the traffic channel (TCH) frequencies of the local macro- and microcells. On the other hand, AFA comprises periodical measurements of the frequencies of the candidate frequency list. For each observed frequency, the measurements are averaged on the basis of a low-pass filter and put into a sorted list. A set of the least interfered frequencies is used for T-FH [5]. Thus, AFA eliminates fast variations of the field strength and of the channel occupation, but reacts on fundamental changes in the environment (frequency plan, new base stations, increased average load). In contrast to dynamic channel allocation schemes, as applied, e.g., in Digital Enhanced Cordless Telecommunication (DECT), AFA requires less measurement effort and has an intended longer response time. A combined T-FH/AFA technique has already become a part of an extension of the GSM standard towards a Cordless Telephony System (CTS) [6]. Previous simulation studies on GSM CTS [7,8] show the feasibility of this technique for uncoordinated CTS systems within the coverage area of a dense urban GSM network. However, a mobile network operator may not be interested in allowing a customer of a competing fixed network operator to reuse the mobile network’s frequencies for cordless access to that fixed network. Instead, the mobile network operator can use the revealed capacity reserves for the optimization of his own network. Beside the capacity-relevant advantages of the T-FH/AFA technique he can benefit from an uncomplicated fast deployment of small and inexpensive picocell base stations in the place where they are needed without frequency planning.
3. Simulation environment 3.1. Radio network model As a worst case for the capacity that is reusable for picocells, we investigate several network configurations in a very dense urban environment with busy hour traffic load. Furthermore, in comparison with real networks, we assume relatively high macro- and microcell transmit powers without
J. DEISSNER AND G.P. FETTWEIS
modeling power control and DTX for interference reduction, additionally challenging the picocell frequency reuse scheme. All base stations (BS) of the macro- and microcells are assumed to be outdoor, also all the macrocell mobile stations (MS) and 50% of the microcell MSs. The remaining microcell MSs as well as all picocell BSs and MSs are assumed to be indoor. Although a real radio network layout does not have a regular hexagonal cell layout like in our model, we may assume that a real reuse factor averaged over a couple of urban cells is higher than in a 1/3 or 3/9 macrocell reuse scheme with an additional microcell layer with a reuse factor of 5 and thus allows only for a lower traffic density. Since a reasonable FH gain occurs beginning at 4 FH frequencies, we allocate 4 frequencies to each sector cell in the 3/9 reuse scheme. In the case of 1/3 reuse, we allocate these 36 frequencies to only 3 sector cells per cluster and assume synthesized FH as well as partial loading. The resulting total number of 36 macrocell plus 10 microcell frequencies is a possible set a network operator may have available for a dense urban area. While the macro- and microcells apply FH following the standard GSM algorithm, the picocells use a different FH algorithm instead, which is also a part of the GSM CTS specification [9] and was proposed in [10]. Table 1 summarizes these assumptions. Each layer is modeled with an independent Poisson call arrival process. However, macrocells are usually intended to carry overflow traffic from the microcells. We assume that the microcells have a busy hour load of 80% and that the blocked calls (16%) are served by the macrocells. Thus, with respect to the macro- and microcell areas, the macrocells have to serve 3.9 Erlang overflow traffic from the microcells (table 2). For the picocells we do not consider such an overflow, because they should serve indoor areas that may have bad macro- and microcell coverage. For our interference analysis, the logical channels BCCH and SDCCH (broadcast and stand-alone dedicated control channels, respectively) are not separately modeled. For representing their interference, we load them like a traffic channel (TCH). The continuous transmission on all time slots of the BCCH frequency (with dummy bursts if not occupied by a TCH) is not modeled, however, without power control and at the high average load values of 72 and 80% we expect no considerable difference. Nevertheless, for the capacity calculation we considered that 2 and, in case of 6 transceivers (TRX) per cell, 3 time slots in a macro- or microcell are reserved for BCCH and SDCCH signalling. Each picocell in our model can serve one user at a time, so that the assumed traffic value of 0.1 Erlang per user leads to 10% average load in such a single-TCH picocell. For our investigation, mainly the average TCH load is of interest, which directly causes the interference to be analyzed. How this load is generated in the picocell layer can also be interpreted in other ways. For example, a group of 8 closely situated picocells, which use the same frequencies, cause similar interference in our model like one picocell that can allocate up to 8 time slots of a single transceiver according to
INCREASED CAPACITY
473
Table 1 Radio network configuration. Macrocell layer • hexagonal layout • inter-BS-distance: DM = 500 m • 3 sectors per BS site • 1/3-reuse with FH over 12 frequencies or 3/9-reuse with FH over 4 frequencies per sector • sector antenna diagram: 3 dB-beamwidth of 60◦ , maximum gain of 21 dBi, and front-to-back ratio of 40 dB • BS transmit power: 49 dBm EIRP (i.e., at least 37 dBm within the 120◦ sector) • MS transmit power: 33 dBm EIRP
3/9 reuse cluster:
Microcell layer • Manhattan grid layout • BSs at the intersections of the grid; inter-BS-distance: Dm = 200 m • 5-reuse scheme with FH over 2 frequencies per cell • omnidirectional BS antennas • micro-BS and -MS transmit power: 27 dBm EIRP
5 reuse:
Picocell layer • picocells, i.e., circles with 40 m diameter, at arbitrary positions, uniformly distributed over the investigation area • pico-BS and -MS at arbitrary positions, uniformly distributed within a picocell • pico-BS and -MS transmit power: 17 dBm EIRP • Total Frequency Hopping following the Lempel–Greenberger/Nonrepeating algorithm [10] Table 2 Layer-specific traffic parameters. Macrocells 1/3-reuse #frequencies per layer #TCH per cell Planning criterion Offered traffic per cell Average call duration Blocking probability Average load Carried traffic per area Spectral capacity
Macrocells 3/9-reuse
Microcells
36 36 10 45 (6 TRXs) 30 (4 TRXs) 14 (2 TRXs) 2% blocking probability 80% average load 50% partial loadinga 22.5 Erlang per cellb 21.9 Erlang per cellb 13.4 Erlang per cell 90 s 90 s 90 s 0 2% 16%c 50% 72% 80% 312 Erlang/km2 b 297 Erlang/km2 b 280 Erlang/km2 1/3 reuse: 64 Erlang/km2 /MHz, 3/9 reuse: 63 Erlang/km2 /MHz (for 36+10 freq.)
Picocells 3, . . . , 18 out of 46 or all 46 1 (like an MS hardware) 3000 users/km2 0.1 Erlang per user 120 s 0 10% 300 Erlang/km2 33Erlang/km2 /MHz (for 46 freq.)
a In relation to the 6 transceivers, i.e., 25% partial loading of the 12 frequencies per sector. b Including 3.9 Erlang per macrocell area, i.e., 54 Erlang/km2 , overflow traffic from the microcells. c The blocked traffic (2.1 Erlang/km2 per microcell area) is served by the macrocells as overflow traffic.
the GSM CTS specification [6]. Due to the trunking gain, of course, a group of 8 TCHs in one base station could be heavier loaded than the 10% of a single TCH. In comparison to the high number of individual base stations in the private CTS use case a radio network operator therefore could spare base stations while serving the same amount of traffic, but only as far as the traffic demand is high enough within the small coverage area of an indoor picocell. Handoff (due to the best server criterion) within the macro- and microcell layers is modeled, but not between different layers. Under those considerations, the simulated load is equivalent to the layer-specific traffic parameters as given in table 2.
The carried traffic in relation to the area and the spectrum can also be considered as the spectral capacity. This has been calculated in table 2 for the two-layer GSM radio network consisting of microcells and either 1/3 or 3/9 reuse macrocells on the one side and for the picocells on the other side. In both cases, the related number of frequencies is 46 (9.2 MHz). 3.2. Propagation models For the macrocell path loss calculation over the distance d in meters we use the Walfish–Ikegami model [11] with a base station antenna height hBS = 25 m, and a building height
474
J. DEISSNER AND G.P. FETTWEIS
hroof = 15 m: LMM = 7.7 + 38 lg d [m].
(1)
For d < 21 m, LMM is replaced by the free-space path loss. For the relation between the macrocell-BSs and the picocell-MSs an indoor penetration loss Lindoor = 12 dB is added: LMp = LMM + 12 dB.
(2)
In the microcell propagation we distinguish the line-of-sight (LOS) case with a two-slope model [11] 27 + 26 lg d [m] for d 500 m LmmLOS = (3) 10.8 + 40 lg d [m] for d > 500 m from the non-line-of-sight (NLOS) case. In the latter and for the relations between the pico-BS and the macro- or microMS as well as between the micro-BS and pico-MS we apply the Walfish–Ikegami model with hBS = 4 m LmmNLOS = −7.5 + 49 lg d [m].
(4)
The linear term which actually arises in this model from the multi-screen diffraction loss was neglected. Thereby, we overestimate the interference by 2 dB at a distance of 100 m to 8 dB at 500 m and beyond. For d < 24 m, LmmNLOS is replaced by the free-space path loss. If the microcell-MS is indoor or if any picocell is involved, Lindoor = 12 dB is added, which represents a relatively low average for different penetration depths and materials in order not to overestimate the isolation of indoor picocells within buildings: LpM = Lmp = Lpm = LmmNLOS + 12 dB.
(5)
For the picocells we use within the same building (d 40 m) a linear attenuation model [12] Lpp = 31.5 + 20 lg d [m] + 0.9d [m]
Figure 1. Propagation models.
(6)
and to other buildings the path loss equation (4) with 2 Lindoor added as the building penetration loss. Log-normal shadowing is considered with σ = 6 dB for indoor and microcell propagation and σ = 8 dB otherwise. Figure 1 gives an overview of all these propagation models. A complete definition of the applied propagation models can be found in [13]. 3.3. Adaptive frequency allocation In our model we determine for each frequency of the candidate list the mean value of the uplink and downlink received power levels of all time slots within the TDMA frame. These power level measurements include the path loss value only, no shadowing. According to our simulated time of 500 s, we measure only once at the simulation setup. For each picocell user, the N least interfered frequencies are used for Total Frequency Hopping (T-FH).
Figure 2. Investigation area for 1/3 macrocell reuse.
3.4. Interference analysis We assume that all BSs and MSs are slot-synchronous and we evaluate co-channel interference only. For 1/3 reuse, we investigate 3 tiers of interfering macrocells (figure 2), for 3/9 reuse one interfering tier. In both cases, the microcell area is a square with a side length of 1667 m and the picocell area a circle with a diameter of 1667 m. However, for the macrocell layer, the carrier-toco-channel-interference ratio (C/I) samples are only taken within the central cluster and for the micro- and picocell layers within a circle with a radius of 500 m. From the simulated C/I samples we derive the C/I distributions and apply the outage probability as performance criterion, which we define as: Poutage = P [C/I < 9 dB].
(7)
Each simulation setup we run with 5 different random generator seeds. Thereby we achieve 95% confidence intervals of 0.5% at 1% outage probability, ±2% at outage probabilities of 8%, and 3% at outage probabilities of 15%. On this basis we present a comparative study for a highdensity hierarchical GSM network with worst case assumptions with respect to the capacity of the picocells.
INCREASED CAPACITY
475
Figure 3. Outage probability in the macro- and microcell layers for 3/9 reuse1 .
Figure 4. Outage probability in the macro- and microcell layers for 1/3 reuse1 .
4. Simulation results 4.1. The allocation of the frequencies to be reused For both macrocell reuse structures, 3/9 with hard blocking and 1/3 with partial loading, we analyze the mutual interference between the picocells and the other layers in several scenarios either with static frequency lists (complete or reduced) or when using adaptive frequency allocation (AFA). In these scenarios the picocells reuse (a) all 46 frequencies of both the macro- and the microcell layer, (b) 18 frequencies according to a reduced static frequency list, which contains all 10 microcell frequencies, but only those macrocell frequencies which are not used in the potential serving cell and its neighbors (in the case of 3/9 reuse only), (c) the 10 microcell frequencies only, (d) 8 frequencies according to a reduced static frequency list, which contains no microcell frequencies and, as
in case b, only those macrocell frequencies which are not used in the potential serving cell and its neighbors (in the case of 3/9 reuse only), and (e)–(j) N = 18 down to N = 3 frequencies determined out of all 46 frequencies by measurements according to the AFA algorithm.
4.1.1. Interference within the macro- and microcell layers Figures 3 and 4 show the respective outage probabilities in the macro- and microcell uplink (UL; left pillar of each pair) and downlink (DL; right pillar), respectively. The reasons for the considerably higher macrocell outage probabilities for 1/3 reuse, which are mainly caused by the macrocell layer itself, are the denser cell layout and the use of 6 (instead of 4) transceivers per macrocell-BS. The benefit from partial loading would only be able to surface by additionally 1 Note, that in case (c) only the micro- and picocells interfere each other,
however, in case (d) only the macro- and picocells.
476
J. DEISSNER AND G.P. FETTWEIS
Figure 5. Outage probability in the picocell layer for 3/9 reuse2 .
considering power control and DTX, which were not modeled. The large difference in the microcell up- and downlink is due to the positions of the micro-BSs at the crossings of the Manhattan grid and the propagation model that assumes a considerably smaller path loss within the street corridors (grid lines) than around the corners and across the builtup area between the streets. So, in our simulated setup, a microcell-BS “sees” possible interfering microcell-MSs in four directions, whereas vice versa only in two directions. 4.1.2. Results for total frequency hopping The reuse of all 46 frequencies for Total Frequency Hopping leads only to a small increase of the outage probability in the macro- and microcell layers (at most 0.8% on the microcell uplink), figures 3 and 4. Also on the picocell uplink a good outage probability can be achieved (4.9% for 3/9 reuse and 5.3% for 1/3 reuse in the macrocells). However, in the assumed network configurations the outage probability on the picocell downlink is not acceptable (16.3% for 3/9 reuse and 15.1% for 1/3 reuse in the macrocells), figures 5 and 6. This means, that the interference averaging gain by the use of T-FH alone is not sufficient for the picocell downlink. Of course, especially a lower macro-BS transmit power and the consideration of power control and DTX would improve the results (see section 4.2). However, we aim at a solution for the picocells that also works under the condition of high interference levels from the macro- and microcell layers. 2 Note, that in case (c) only the micro- and picocells interfere each other,
however, in case (d) only the macro- and picocells.
Figure 6. Outage probability in the picocell layer for 1/3 reuse2 .
4.1.3. Results for T-FH with reduced static frequency lists The use of static frequency lists for T-FH, which contain a reduced number of frequencies with respect to the network configuration, improve the performance of the picocell downlink already to some extent. The results for the reuse of only 8 macrocell frequencies indicate that the influence of the picocells on the macrocells is negligible, see case (d) in figure 3. This is a real capacity reserve, which should be exploited for the picocells. However, the outage probability on the picocell downlink remains with 11.5% still unacceptably high, figure 5.
INCREASED CAPACITY
477
By including all the frequencies of the microcells the outage probability on the picocell downlink can slightly be improved, but only to the debit of the microcell as well as the picocell uplink performance, see case (b) in figures 3 and 5. The results for case (c), where only the microcell frequencies are reused, further emphasize the need for also excluding the strongestly interfered microcell frequencies, although the micro-BSs and -MSs only transmit with 27 dBm EIRP. Of course, the exclusion of locally used microcell frequencies would further improve the picocell performance. However, in real irregular cell layouts the improvement in the interference on the macro- or microcell layer by reduced static frequency lists would only be both achieved and intensified by measurements and, based on this, an adaptation to the local interference situation as the AFA algorithm eventually provides.
figurations at, approximately, 6 frequencies for 3/9 reuse. This value is shifted towards 4 frequencies for 1/3 reuse. Obviously, in the denser macrocell layout, less frequencies meet the low interference level, which is required for an effective reuse in the picocells. It is remarkable that, although the macro- and microcellcaused interference is in the cases (e) and (f) much higher for the denser 1/3 reuse, approximately the same picocell performance like in the 3/9 reuse can be reached in the cases (h)– (j). From these results we can see, that it is more efficient to exclude the strongest interferers than to achieve a higher interference averaging gain by using as much frequencies as possible for T-FH.
4.1.4. Results for T-FH with adaptive frequency allocation If AFA determines 18 frequencies or less for T-FH in the 3/9 macrocell reuse cases, figure 3, and 10 frequencies or less in the 1/3 reuse cases, figure 4, respectively, the impact on the macrocells from the frequency reuse in the picocells is negligible. In the 1/3 reuse cases, the picocells obviously make more use of the less interfered microcell frequencies than of the macrocell frequencies, which leads in comparison to 3/9 reuse to an increased picocell-caused outage probability in the microcells, compare especially case (f) in figures 3 and 4. Nevertheless, the microcell outages can still be decreased, e.g., on the uplink from 1.7% in case (f) to 0.3% in cases (i) and (j), figure 4. At low numbers of frequencies the picocellcaused outage probability in the microcells even reaches similar values, namely 0.2–0.4%, for both the 3/9 and the denser 1/3 macrocell reuse. In the picocells the use of AFA considerably decreases the outage probability, figures 5 and 6. According to the transmission characteristics in the macrocells and the interference conditions in the microcells, up- and downlink reach their minimal outage probability at different numbers of frequencies. On the uplink (left pillar of each pair) we find the minimum at 18 frequencies for 3/9 reuse and at 10 frequencies for 1/3 reuse, respectively. In both cases, the downlink outage probability (right pillar of each pair) reaches its minimum at 6 frequencies. This minimum is due to an optimal combination of excluding strongly interfered frequencies on the one hand and of interference averaging on the other hand. For less frequencies, the interference averaging gain decreases, because the probability of hits from other picocell users hopping over 5 or less frequencies increases. The resulting increase of the frame erasure rate will additionally be influenced by the limited frequency diversity, which slow moving or quasistationary picocell users encounter for very few hopping frequencies. If we consider up- and downlink together, we can find the optimal number of frequencies selected by the AFA algorithm for T-FH in the considered hierarchical network con-
In order to better assess the previous results, we want to discuss the influence of some other characteristics than the number of frequencies in the following, namely the picocell user density, the use of power control and DTX, and the building penetration loss. If we double the picocell user density to 6000/km2 in the 1/3 reuse case with 4 AFA frequencies (compare with case (i) in figures 4 and 6), the macrocell performance is not affected. The picocell-caused increase of the microcell outage probability is 0.7% on the downlink and even less on the uplink. The increase of the picocell outage probability by approximately 1% on both, up- and downlink, is mainly due to the higher interference within the picocell layer. Hence, these results indicate that also higher picocell user densities than 3000/km2 are possible and even 6000/km2 is not yet critical. Moreover, we expect that the picocell performance will still improve under consideration of power control and DTX. Further simulations with a 6 dB lower transmit power level of the sectorized macrocell-BSs, i.e., 43 dBm EIRP, were expected to have a similar effect in the reduction of interference like the application of power control and DTX. In the respective results for 3/9 macrocell reuse the outage probability on the picocell downlink drops from 16.3% (figure 5) to ca. 11% (not shown) in case (a) and from 10.8% (figure 5) to ca. 9% (not shown) in case (b), respectively. The picocell-caused interference in the macro- or microcells is not affected. Finally, among the propagation modeling assumptions, the building penetration loss has a strong influence on the results. For example, a 3 dB stronger isolation of picocells within buildings, i.e., Lindoor = 15 dB, would further decrease the outage probability on the picocell downlink cases (e), (f), and (h) for 3/9 reuse by ca. 3%, 2%, and 1%, respectively (see figure 5).
4.2. The influence of other characteristics than the frequency allocation
4.3. Summary Under worst-case conditions with respect to the interference from other layers to picocells in a dense urban environment,
478
the presented results indicate that AFA provides a considerable interference reduction gain in both investigated network configurations (at most a reduction of the outage probability on the sensitive picocell downlink by 12% and 11%, respectively). Moreover, approximately the same picocell outage probability values like in the macro- and microcell layers can be achieved, though those are frequency-planned and the picocells not. Since a real network with irregular cell layout is less tight than the 3/9 reuse case or at most within the range of 3/9 and 1/3 reuse, each overlaid with a 5-reuse microcell layer, we may assume that the performance results are a lower bound for the capacity of the indoor picocells. Hence, solely by reusing existing macro- and microcell frequencies, an additional indoor and hot-spot traffic of more than 300 Erlang/km2 can be carried by the picocells. This corresponds to an increase in spectral capacity by over 50%, namely 33 Erlang/km2/MHz. The influence of the picocells on the macrocells is in both reuse scenarios for less than 10 frequencies negligible. The outage probability in the microcells is increased by the picocells at most by 1.7%, in the cases, which are optimal for the picocells, only by 0.5%. In joint consideration of up- and downlink, the optimal number of frequencies selected by the AFA algorithm for TFH is for the considered hierarchical network configurations at approximately 6 frequencies for 3/9 reuse and is shifted towards 4 frequencies for 1/3 reuse. The results also show, that it is more efficient to exclude the strongest interferers than to achieve a higher interference averaging gain by using as much frequencies for T-FH as possible. Moreover, higher picocell user densities than 3000/km2 are possible and even 6000/km2 is not yet critical. And if we assume 6 dB smaller transmit powers at the macrocell-BSs, which can be compared with the case of downlink power control and DTX, the macrocell-caused outage probability at the picocell downlink can be further decreased by 5% for 46 reused frequencies and by 2% for adaptively 18 reused frequencies, respectively, without increasing the picocellcaused interference in the macro- or microcells.
5. Discussion of the results Since the configuration and the environment of cellular networks are usually very inhomogeneous, hierarchical cellular structures require reuse strategies that do not only adapt to the current local interference situation, but additionally distribute the remaining interference to as many resources as possible. Likewise the measurement effort should be kept to a minimum. For a hierarchical GSM network these requirements are fulfilled by the T-FH/AFA technique very well. In addition to that, this technique spares the planning of the picocell frequencies that are reused from the macro- and microcells. Frequency planning for these picocells would not be possible or at least incorporate a very high effort. For FDMA systems with a reuse factor greater than 1 adaptive and intelligent channel allocation techniques like
J. DEISSNER AND G.P. FETTWEIS
AFA can transform capacity reserves into a considerable additional traffic capacity solely by the reuse of spectrum already allocated to this system. The reasons for this are the different characteristics of the several layers of the hierarchical cellular network, mainly regarding to the power settings and the propagation, which overall lead to an inhomogeneous occupation of the radio resources. Moreover, in real irregular cell layouts, lacks of homogeneous coverage can appear, because the base stations can often not be placed at the optimal locations. Attenuations due to strong shadowing should not be eliminated by increasing the power, e.g., in order to provide indoor coverage. On the contrary, they should be exploited for reusing the attenuated resources. For example, dedicated indoor cells can benefit from the building penetration loss. At least partially they can reuse the resources from the outdoor cells and thus provide extra capacity without additional radio resources. Though the fixed frequency duplex offset of the GSM system prevents different frequency sets in uplink and downlink, adaptive channel allocation techniques in future cellular systems should be able to handle uplink and downlink separately. The two reasons for that are the different interference situations on uplink and downlink, as we can also see in most of the results presented in section 4, as well as different capacity and, where appropriate, quality requirements on uplink and downlink. For the reduction of the measurement effort and the speed-up of the AFA list update in new environments (especially for ad hoc installations), it is worth to incorporate information on the network structure in the AFA algorithm (such as the layer the channel belongs to and its average load as well as if it is a channel of the potential serving cell). Thereby, less promising channels, i.e., with potentially high interference, need not to be measured or may only be gradually included for the optimization of the AFA list after the initialization or an update. As frequency hopping is a special CDMA technique, we can find in our results for GSM with T-FH/AFA some similarities with results from other CDMA system analyses. For instance, as typical for CDMA systems, the downlink limits the performance of the radio network, as it is the case for the picocells in our study. Moreover, regarding the ability to avoid and distribute interference, T-FH/AFA is apparently very similar to a broadband CDMA system overlaying a GSM system with inter-layer reuse, which requires agile notch filtering according to studies in [14,15]. The increase in spectral capacity which we achieved in our study seems to be small in comparison to the results from those studies. In [14], for example, a CDMA overlay network cannot only carry 50% in addition to the GSM network traffic but nearly 300%. This requires 9 notch filters with depths of 35 dB on receive and 20 dB on transmit. Moreover, the GSM and the Broadband CDMA base stations are intended to be co-located and the overlay network does not “need” some sort of shadowing in the environment for an isolation of the cells that reuse the GSM resources
INCREASED CAPACITY
as in our study. However, we investigated a much denser GSM cell layout, where a large part of possible capacity reserves has already being exploited in the GSM macro- and microcells. In [14] a 4/12 reuse GSM cell layout with an inter-BS distance of 1.7 km and 67% traffic load is studied, which has a spectral capacity of merely 3 Erlang/km2/MHz. The CDMA overlay network increases this spectral capacity by some 7 Erlang/km2/MHz. Note that we achieved an increase by 33 Erlang/km2/MHz over an existing spectral capacity of 64 Erlang/km2/MHz. Hence, the use of the absolute spectral capacity as a measure for the density of the cell layout is strongly recommended for a comparison of the results of such kind of studies. Of course, the cost of additional base stations and new base station functionality is only worth with adequately high traffic demand. However, the quickly growing user penetration together with an increased use of data services on the basis of new techniques like the General Packet Radio Service (GPRS) and High-Speed Circuit-Switched Data (HSCSD) in GSM, which are all to be served by the limited bandwidth licensed to a network operator, indicate a rising necessity for such an optimization by means of hierarchical cellular networks. Naturally, hierarchical cellular structures become a decisive part of third generation mobile radio system concepts from their beginning. However, apart from the layer-specific optimization or the only interdependence through inter-layer handoffs, a further potential capacity improvement arises from the mutual reuse of the radio resources in the different layers on the basis of techniques that adapt themselves to the local interference situation. In order to further analyze such capacity improvements, we use simulation techniques that adequately cover the adaptivity and the dynamics, which are decisive for the expected gains [16–18]. For the deployment of hierarchical cellular structures it has also to be ensured that the reuse of the radio resources also works in rare extreme cases. Hence, in addition to the statistical analysis of the interference performance, system scenario calculations are always required, e.g., with respect to the transmitter output RF spectrum and spurious emissions as well as to the receiver blocking characteristics [19].
6. Conclusions In an interference study for a hierarchical GSM radio network we showed the capability of Total Frequency Hopping (T-FH) in conjunction with Adaptive Frequency Allocation (AFA) to transform capacity reserves into an increase of the spectral capacity from 64 Erlang/km2/MHz to 97 Erlang/km2/MHz. Thereby, a good interference performance can be achieved in all layers. Moreover, frequency planning for the indoor picocells is avoided. Such capacity reserves may still exist in very dense radio network configurations like those investigated as a worst case. Generally, they can be exploited by reuse strategies,
479
which do not only adapt to the current local interference situation, but additionally distribute the remaining interference to as many resources as possible. T-FH/AFA is an example for such a technique, which is well suited for FDMA systems and implies less measurement effort than dynamic channel allocation schemes. Moreover, it has already become a part of an extension of the GSM standard towards a Cordless Telephony System. Similarities with Broadband CDMA overlay scenarios were discussed. As in that case, the spectral capacity is a measure that enables a comparison of investigation results for very different radio network configurations. Acknowledgements The simulation software was developed within a project at Dresden University of Technology that was sponsored by the Alcatel Corporate Research Center, Stuttgart. Here, we wish to thank Ulrich Barth and Charles Skelton for providing us with the software version that was extended afterwards by Alcatel and their support. We would also like to thank Dr. Michael Hartmann, Mannesmann Mobilfunk GmbH, Dresden, for the fruitful discussion on the setup for this simulation study. References [1] M. Frullone, G. Riva, P. Grazioso and M. Barbiroli, Advanced planning criteria for cellular systems, IEEE Personal Communications (December 1996) 10–15. [2] M. Madfors, K. Wallstedt, S. Magnusson, H. Oloffson, P.-O. Backman and S. Engström, High capacity with limited spectrum in cellular systems, IEEE Communications Magazine (August 1997) 38–44. [3] M.I. Silventoinen, M. Kuusela and P.A. Ranta, Analysis of a new channel access method for Home Base Station, in: Proc. ICUPC’96, Cambridge, MA (October 1996) pp. 930–935. [4] J. Deißner, A. Noll Barreto, U. Barth and G. Fettweis, Interference analysis of a Total Frequency Hopping GSM Cordless Telephony System, in: Proc. PIMRC’98, Boston, MA (September 1998) pp. 1525– 1529. [5] First investigations of AFA performance in the CTS, TDoc SMG2 WPB 87/98, ETSI (1998). [6] Digital cellular telecommunications system (Phase 2+); GSM Cordless Telephony System (CTS), Phase 1; Lower Layers of the CTS Radio Interface; Stage 2 (GSM 03.52 version 7.0.0 Release 1998), ETSI (1999). [7] Interference performance of GSM-CTS: Simulation results, TDoc SMG2 WPB 95/98, ETSI (1998). [8] Simulation results of the interference performance of the AFA/TFH radio interface concepts for the GSM–CTS for scenarios including a microlayer in the GSM–PLMN, TDoc SMG2 WPB 174/98, ETSI (1998). [9] Digital cellular telecommunications system (Phase 2+); Multiplexing and multiple access on the radio path (GSM 05.02 version 8.0.0 Release 1998), ETSI (1999). [10] A. Noll Barreto, J. Deissner and G.P. Fettweis, A frequency hopping algorithm for cordless telephone systems, in: Proc. ICUPC’98, Florence, Italy (October 1998) pp. 1273–1277. [11] Digital cellular telecommunications system; Radio network planning aspects (GSM 03.30 version 5.0.0), ETSI (1996). [12] COST Action 231: Digital mobile radio towards future generation systems, Final report, European Communities (1999).
480
J. DEISSNER AND G.P. FETTWEIS
[13] Definition of the simulation environment for the evaluation of the interference performance of GSM–CTS radio interface concepts, TDoc SMG2 WPB 88/98, ETSI (1998). [14] D.M. Grieco and D.L. Schilling, The capacity of Broadband CDMA overlaying a GSM cellular system, in: Proc. IEEE VTC’94 (1994) pp. 31–35. [15] T. Widdowson, A CDMA overlay of the GSM network, in: Proc. IEEE PIMRC’97, Helsinki, Finland (September 1997) pp. 160–163. [16] J. Fischer, J. Deissner, G. Fettweis, D. Hunold, R. Lehnert, M. Schweigel, J. Voigt and J. Wagner, Object-oriented modeling of a generic mobile radio system for dynamic system simulation, in: Proc. of the Symposium on Performance Evaluation of Computer and Telecommunication Systems SPECTS’99, Chicago, IL (July 1999) pp. 240–247. [17] J. Deissner, G. Fettweis, J. Fischer, D. Hunold, R. Lehnert, M. Schweigel, A.Steil, J. Voigt and J. Wagner, A development platform for the design and optimization of mobile radio networks, in: Kurzvorträge der 10. GI/ITG- Fachtagung Messung, Modellierung und Bewertung von Rechen- und Kommunikationssystemen MMB’99, eds. D. Baum, N. Müller, R. Rödler, Universität Trier, Germany, Mathematik/Informatik, Forschungsbericht Nr. 99-17 (1999) pp. 129133. [18] WiNeS – Wireless Network System Simulator, TU Dresden, http://entmuc.et.tu-dresden.de/MNS/forschung. [19] System scenario calculations for GSM–CTS, TDoc SMG2 WPB 12/ 99, ETSI (1999).
Jürgen Deissner received his Dipl.-Ing. and Ph.D. degree in electrical engineering/information technology from Dresden University of Technology, Germany, in 1994 and 2001, respectively. From 1995 to 2000 he was a Research Associate at the Mannesmann Mobilfunk Chair for Mobile Communications Systems at Dresden University of Technology working on radio resource management techniques for increasing the capacity of mobile radio networks as well as modeling mobile
communications systems for interference-based capacity and quality analysis by means of radio-network-level simulations. In 2000 he worked with AT&T Labs Research, New Jersey, in the field of GPRS/EDGE modeling and analysis, and in 2001 he joined Radioplan GmbH, a startup spun out of Dresden University of Technology focusing on radio network analysis and optimization software. Jürgen Deissner is a member of the IEEE and the German VDE. E-mail: [email protected]
Gerhard Fettweis [IEEE S’82–M’90–SM’98] received his MSc/Dipl.-Ing. and Ph.D. degree in electrical engineering from the Aachen University of Technology (RWTH), Germany, in 1986 and 1990, respectively. From 1990 to 1991 he was a Visiting Scientist at the IBM Almaden Research Center in San Jose, CA, working on signal processing for disk drives. From 1991 to 1994 he was a Scientist with TCSI, Berkeley, CA, responsible for signal processor developments for mobile phones. Since September 1994 he holds the Mannesmann Mobilfunk Chair for Mobile Communications Systems at the Dresden University of Technology, Germany. He is an elected member of the SSC Society’s Administrative Commitee, and of IEEE ComSoc Board of governors, since 1999 and 1998, respectively. He has been an associate editor for IEEE Transactions on Circuits and Systems II, and now is an associate editor for IEEE Journal on Selected Areas in Communications wireless series. E-mail: [email protected]
Increased Capacity through Hierarchical Cellular Structures with Inter-Layer Reuse in an Enhanced GSM Radio Network ∗ JÜRGEN DEISSNER and GERHARD P. FETTWEIS Dresden University of Technology, Mannesmann Mobilfunk Chair for Mobile Communications Systems, D-01062 Dresden, Germany
Abstract. In today’s cellular networks it becomes harder to provide the resources for the increasing and fluctuating traffic demand exactly in the place and at the time where and when they are needed. Moreover, frequency planning for a hierarchical cellular network, especially to cover indoor areas and hot-spots is a complicated and expensive task. Therefore, we study the ability of hierarchical cellular structures with inter-layer reuse to increase the capacity of a GSM (Global System for Mobile Communications) radio network by applying Total Frequency Hopping (T-FH) and Adaptive Frequency Allocation (AFA) as a strategy to reuse the macro- and microcell resources without frequency planning in indoor picocells. The presented interference analysis indicates a considerable interference reduction gain by T-FH in conjunction with AFA, which can be used for carrying an additional indoor traffic of more than 300 Erlang/km2 , i.e., increasing the spectral capacity by over 50%, namely 33 Erlang/km2 /MHz. From these results we draw a number of general conclusions for the design of hierarchical cellular structures in future mobile radio networks. For example, we may conclude that they require reuse strategies that not only adapt to the current local interference situation, but additionally distribute the remaining interference to as many resources as possible. For a hierarchical GSM network this requirement is fulfilled by the T-FH/AFA technique very well. Keywords: hierarchical cellular structures, Global System for Mobile Communications (GSM), Total Frequency Hopping, adaptive frequency allocation
1. Introduction The provision of capacity for the increasing traffic demand in mobile radio networks comes along with the reduction of the cell size and, hence, stronger traffic fluctuations between the cells. Moreover, improved indoor coverage is required. It becomes harder to provide the resources exactly in the place and at the time where and when they are needed. Hierarchical cellular structures can serve indoor users and hot spots by pico- and microcell layers, respectively, while providing coverage in the area by the macrocell layer. Moreover, hierarchical cellular structures can compensate traffic fluctuations, e.g., by shifting overflow traffic from lower to higher layers. In order to avoid interference between the layers, their frequencies have to be coordinated. However, due to the heterogeneous cell sizes and propagation conditions, frequency planning for a hierarchical cellular network is a complicated and expensive task. That problem is even worse if picocells have to be deployed suddenly or only for short periods, e.g., at fairs, conferences, ad hoc events, for which it is not worth or even not possible to go through the frequency planning cycle. Furthermore, hierarchical cellular structures become a regular feature of future mobile radio networks. Although different multiple access techniques may apply, some experiences from GSM can also be useful for the design of other hierarchical cellular networks, where several layers share the same resources. ∗ This work was supported in part by the Alcatel Corporate Research Cen-
tre, Stuttgart, Germany.
In this paper, we study therefore the ability of hierarchical cellular structures with inter-layer reuse to increase the capacity of a GSM radio network by applying Total Frequency Hopping and Adaptive Frequency Allocation as a strategy to reuse the macro- and microcell resources without frequency planning in indoor picocells. After a discussion of previous work on hierarchical GSM networks in section 2, we describe our simulation environment in section 3. In section 4 we present the results of our comparative simulation study that aims at network configurations in a dense urban environment where a high additional traffic capacity in the picocell layer shall be achieved solely by reusing the microand/or macrocell frequencies. Finally in section 5, we discuss the results and draw a number of general conclusions for the design of hierarchical cellular networks. 2. Discussion of previous work on hierarchical GSM networks Interference reduction techniques like power control, frequency hopping (FH), discontinuous transmission (DTX), partial loading, and adaptive antennas improve the performance of hierarchical GSM networks. And, especially, partial loading in combination with FH can compensate traffic fluctuations to some extent [1]. In the macro- and microcell layer usually different frequencies are used to avoid the coordination effort between the layers. However, in order to not decrease the outdoor capacity of those layers it is not desirable to dedicate a large part of the scarce frequency band of a network operator to the indoor picocell layer. Facing this requirement, a strategy
472
to combat interference between the picocells and the other layers is to use wideband synthesized FH for the traffic channels [2]. Basically, Total Frequency Hopping (T-FH) in picocells is such a solution. T-FH is a slow FH technique which uses more frequencies for FH than typically chosen per base station, e.g., up to the whole frequency band of a network operator [3]. The more frequencies are used the more the interference is spread over all frequencies. The resulting interference diversity effect opens up capacity reserves in the mobile radio network. However, this is not sufficient for an acceptable performance in the picocells [4]. In fact, the performance of T-FH can be much improved by excluding the strongestly interfered frequencies and thus statically reducing the frequency list for T-FH in the picocells or by Adaptive Frequency Allocation (AFA) as well. A reduced frequency list may, for example, exclude the broadcast control channel (BCCH) and the traffic channel (TCH) frequencies of the local macro- and microcells. On the other hand, AFA comprises periodical measurements of the frequencies of the candidate frequency list. For each observed frequency, the measurements are averaged on the basis of a low-pass filter and put into a sorted list. A set of the least interfered frequencies is used for T-FH [5]. Thus, AFA eliminates fast variations of the field strength and of the channel occupation, but reacts on fundamental changes in the environment (frequency plan, new base stations, increased average load). In contrast to dynamic channel allocation schemes, as applied, e.g., in Digital Enhanced Cordless Telecommunication (DECT), AFA requires less measurement effort and has an intended longer response time. A combined T-FH/AFA technique has already become a part of an extension of the GSM standard towards a Cordless Telephony System (CTS) [6]. Previous simulation studies on GSM CTS [7,8] show the feasibility of this technique for uncoordinated CTS systems within the coverage area of a dense urban GSM network. However, a mobile network operator may not be interested in allowing a customer of a competing fixed network operator to reuse the mobile network’s frequencies for cordless access to that fixed network. Instead, the mobile network operator can use the revealed capacity reserves for the optimization of his own network. Beside the capacity-relevant advantages of the T-FH/AFA technique he can benefit from an uncomplicated fast deployment of small and inexpensive picocell base stations in the place where they are needed without frequency planning.
3. Simulation environment 3.1. Radio network model As a worst case for the capacity that is reusable for picocells, we investigate several network configurations in a very dense urban environment with busy hour traffic load. Furthermore, in comparison with real networks, we assume relatively high macro- and microcell transmit powers without
J. DEISSNER AND G.P. FETTWEIS
modeling power control and DTX for interference reduction, additionally challenging the picocell frequency reuse scheme. All base stations (BS) of the macro- and microcells are assumed to be outdoor, also all the macrocell mobile stations (MS) and 50% of the microcell MSs. The remaining microcell MSs as well as all picocell BSs and MSs are assumed to be indoor. Although a real radio network layout does not have a regular hexagonal cell layout like in our model, we may assume that a real reuse factor averaged over a couple of urban cells is higher than in a 1/3 or 3/9 macrocell reuse scheme with an additional microcell layer with a reuse factor of 5 and thus allows only for a lower traffic density. Since a reasonable FH gain occurs beginning at 4 FH frequencies, we allocate 4 frequencies to each sector cell in the 3/9 reuse scheme. In the case of 1/3 reuse, we allocate these 36 frequencies to only 3 sector cells per cluster and assume synthesized FH as well as partial loading. The resulting total number of 36 macrocell plus 10 microcell frequencies is a possible set a network operator may have available for a dense urban area. While the macro- and microcells apply FH following the standard GSM algorithm, the picocells use a different FH algorithm instead, which is also a part of the GSM CTS specification [9] and was proposed in [10]. Table 1 summarizes these assumptions. Each layer is modeled with an independent Poisson call arrival process. However, macrocells are usually intended to carry overflow traffic from the microcells. We assume that the microcells have a busy hour load of 80% and that the blocked calls (16%) are served by the macrocells. Thus, with respect to the macro- and microcell areas, the macrocells have to serve 3.9 Erlang overflow traffic from the microcells (table 2). For the picocells we do not consider such an overflow, because they should serve indoor areas that may have bad macro- and microcell coverage. For our interference analysis, the logical channels BCCH and SDCCH (broadcast and stand-alone dedicated control channels, respectively) are not separately modeled. For representing their interference, we load them like a traffic channel (TCH). The continuous transmission on all time slots of the BCCH frequency (with dummy bursts if not occupied by a TCH) is not modeled, however, without power control and at the high average load values of 72 and 80% we expect no considerable difference. Nevertheless, for the capacity calculation we considered that 2 and, in case of 6 transceivers (TRX) per cell, 3 time slots in a macro- or microcell are reserved for BCCH and SDCCH signalling. Each picocell in our model can serve one user at a time, so that the assumed traffic value of 0.1 Erlang per user leads to 10% average load in such a single-TCH picocell. For our investigation, mainly the average TCH load is of interest, which directly causes the interference to be analyzed. How this load is generated in the picocell layer can also be interpreted in other ways. For example, a group of 8 closely situated picocells, which use the same frequencies, cause similar interference in our model like one picocell that can allocate up to 8 time slots of a single transceiver according to
INCREASED CAPACITY
473
Table 1 Radio network configuration. Macrocell layer • hexagonal layout • inter-BS-distance: DM = 500 m • 3 sectors per BS site • 1/3-reuse with FH over 12 frequencies or 3/9-reuse with FH over 4 frequencies per sector • sector antenna diagram: 3 dB-beamwidth of 60◦ , maximum gain of 21 dBi, and front-to-back ratio of 40 dB • BS transmit power: 49 dBm EIRP (i.e., at least 37 dBm within the 120◦ sector) • MS transmit power: 33 dBm EIRP
3/9 reuse cluster:
Microcell layer • Manhattan grid layout • BSs at the intersections of the grid; inter-BS-distance: Dm = 200 m • 5-reuse scheme with FH over 2 frequencies per cell • omnidirectional BS antennas • micro-BS and -MS transmit power: 27 dBm EIRP
5 reuse:
Picocell layer • picocells, i.e., circles with 40 m diameter, at arbitrary positions, uniformly distributed over the investigation area • pico-BS and -MS at arbitrary positions, uniformly distributed within a picocell • pico-BS and -MS transmit power: 17 dBm EIRP • Total Frequency Hopping following the Lempel–Greenberger/Nonrepeating algorithm [10] Table 2 Layer-specific traffic parameters. Macrocells 1/3-reuse #frequencies per layer #TCH per cell Planning criterion Offered traffic per cell Average call duration Blocking probability Average load Carried traffic per area Spectral capacity
Macrocells 3/9-reuse
Microcells
36 36 10 45 (6 TRXs) 30 (4 TRXs) 14 (2 TRXs) 2% blocking probability 80% average load 50% partial loadinga 22.5 Erlang per cellb 21.9 Erlang per cellb 13.4 Erlang per cell 90 s 90 s 90 s 0 2% 16%c 50% 72% 80% 312 Erlang/km2 b 297 Erlang/km2 b 280 Erlang/km2 1/3 reuse: 64 Erlang/km2 /MHz, 3/9 reuse: 63 Erlang/km2 /MHz (for 36+10 freq.)
Picocells 3, . . . , 18 out of 46 or all 46 1 (like an MS hardware) 3000 users/km2 0.1 Erlang per user 120 s 0 10% 300 Erlang/km2 33Erlang/km2 /MHz (for 46 freq.)
a In relation to the 6 transceivers, i.e., 25% partial loading of the 12 frequencies per sector. b Including 3.9 Erlang per macrocell area, i.e., 54 Erlang/km2 , overflow traffic from the microcells. c The blocked traffic (2.1 Erlang/km2 per microcell area) is served by the macrocells as overflow traffic.
the GSM CTS specification [6]. Due to the trunking gain, of course, a group of 8 TCHs in one base station could be heavier loaded than the 10% of a single TCH. In comparison to the high number of individual base stations in the private CTS use case a radio network operator therefore could spare base stations while serving the same amount of traffic, but only as far as the traffic demand is high enough within the small coverage area of an indoor picocell. Handoff (due to the best server criterion) within the macro- and microcell layers is modeled, but not between different layers. Under those considerations, the simulated load is equivalent to the layer-specific traffic parameters as given in table 2.
The carried traffic in relation to the area and the spectrum can also be considered as the spectral capacity. This has been calculated in table 2 for the two-layer GSM radio network consisting of microcells and either 1/3 or 3/9 reuse macrocells on the one side and for the picocells on the other side. In both cases, the related number of frequencies is 46 (9.2 MHz). 3.2. Propagation models For the macrocell path loss calculation over the distance d in meters we use the Walfish–Ikegami model [11] with a base station antenna height hBS = 25 m, and a building height
474
J. DEISSNER AND G.P. FETTWEIS
hroof = 15 m: LMM = 7.7 + 38 lg d [m].
(1)
For d < 21 m, LMM is replaced by the free-space path loss. For the relation between the macrocell-BSs and the picocell-MSs an indoor penetration loss Lindoor = 12 dB is added: LMp = LMM + 12 dB.
(2)
In the microcell propagation we distinguish the line-of-sight (LOS) case with a two-slope model [11] 27 + 26 lg d [m] for d 500 m LmmLOS = (3) 10.8 + 40 lg d [m] for d > 500 m from the non-line-of-sight (NLOS) case. In the latter and for the relations between the pico-BS and the macro- or microMS as well as between the micro-BS and pico-MS we apply the Walfish–Ikegami model with hBS = 4 m LmmNLOS = −7.5 + 49 lg d [m].
(4)
The linear term which actually arises in this model from the multi-screen diffraction loss was neglected. Thereby, we overestimate the interference by 2 dB at a distance of 100 m to 8 dB at 500 m and beyond. For d < 24 m, LmmNLOS is replaced by the free-space path loss. If the microcell-MS is indoor or if any picocell is involved, Lindoor = 12 dB is added, which represents a relatively low average for different penetration depths and materials in order not to overestimate the isolation of indoor picocells within buildings: LpM = Lmp = Lpm = LmmNLOS + 12 dB.
(5)
For the picocells we use within the same building (d 40 m) a linear attenuation model [12] Lpp = 31.5 + 20 lg d [m] + 0.9d [m]
Figure 1. Propagation models.
(6)
and to other buildings the path loss equation (4) with 2 Lindoor added as the building penetration loss. Log-normal shadowing is considered with σ = 6 dB for indoor and microcell propagation and σ = 8 dB otherwise. Figure 1 gives an overview of all these propagation models. A complete definition of the applied propagation models can be found in [13]. 3.3. Adaptive frequency allocation In our model we determine for each frequency of the candidate list the mean value of the uplink and downlink received power levels of all time slots within the TDMA frame. These power level measurements include the path loss value only, no shadowing. According to our simulated time of 500 s, we measure only once at the simulation setup. For each picocell user, the N least interfered frequencies are used for Total Frequency Hopping (T-FH).
Figure 2. Investigation area for 1/3 macrocell reuse.
3.4. Interference analysis We assume that all BSs and MSs are slot-synchronous and we evaluate co-channel interference only. For 1/3 reuse, we investigate 3 tiers of interfering macrocells (figure 2), for 3/9 reuse one interfering tier. In both cases, the microcell area is a square with a side length of 1667 m and the picocell area a circle with a diameter of 1667 m. However, for the macrocell layer, the carrier-toco-channel-interference ratio (C/I) samples are only taken within the central cluster and for the micro- and picocell layers within a circle with a radius of 500 m. From the simulated C/I samples we derive the C/I distributions and apply the outage probability as performance criterion, which we define as: Poutage = P [C/I < 9 dB].
(7)
Each simulation setup we run with 5 different random generator seeds. Thereby we achieve 95% confidence intervals of 0.5% at 1% outage probability, ±2% at outage probabilities of 8%, and 3% at outage probabilities of 15%. On this basis we present a comparative study for a highdensity hierarchical GSM network with worst case assumptions with respect to the capacity of the picocells.
INCREASED CAPACITY
475
Figure 3. Outage probability in the macro- and microcell layers for 3/9 reuse1 .
Figure 4. Outage probability in the macro- and microcell layers for 1/3 reuse1 .
4. Simulation results 4.1. The allocation of the frequencies to be reused For both macrocell reuse structures, 3/9 with hard blocking and 1/3 with partial loading, we analyze the mutual interference between the picocells and the other layers in several scenarios either with static frequency lists (complete or reduced) or when using adaptive frequency allocation (AFA). In these scenarios the picocells reuse (a) all 46 frequencies of both the macro- and the microcell layer, (b) 18 frequencies according to a reduced static frequency list, which contains all 10 microcell frequencies, but only those macrocell frequencies which are not used in the potential serving cell and its neighbors (in the case of 3/9 reuse only), (c) the 10 microcell frequencies only, (d) 8 frequencies according to a reduced static frequency list, which contains no microcell frequencies and, as
in case b, only those macrocell frequencies which are not used in the potential serving cell and its neighbors (in the case of 3/9 reuse only), and (e)–(j) N = 18 down to N = 3 frequencies determined out of all 46 frequencies by measurements according to the AFA algorithm.
4.1.1. Interference within the macro- and microcell layers Figures 3 and 4 show the respective outage probabilities in the macro- and microcell uplink (UL; left pillar of each pair) and downlink (DL; right pillar), respectively. The reasons for the considerably higher macrocell outage probabilities for 1/3 reuse, which are mainly caused by the macrocell layer itself, are the denser cell layout and the use of 6 (instead of 4) transceivers per macrocell-BS. The benefit from partial loading would only be able to surface by additionally 1 Note, that in case (c) only the micro- and picocells interfere each other,
however, in case (d) only the macro- and picocells.
476
J. DEISSNER AND G.P. FETTWEIS
Figure 5. Outage probability in the picocell layer for 3/9 reuse2 .
considering power control and DTX, which were not modeled. The large difference in the microcell up- and downlink is due to the positions of the micro-BSs at the crossings of the Manhattan grid and the propagation model that assumes a considerably smaller path loss within the street corridors (grid lines) than around the corners and across the builtup area between the streets. So, in our simulated setup, a microcell-BS “sees” possible interfering microcell-MSs in four directions, whereas vice versa only in two directions. 4.1.2. Results for total frequency hopping The reuse of all 46 frequencies for Total Frequency Hopping leads only to a small increase of the outage probability in the macro- and microcell layers (at most 0.8% on the microcell uplink), figures 3 and 4. Also on the picocell uplink a good outage probability can be achieved (4.9% for 3/9 reuse and 5.3% for 1/3 reuse in the macrocells). However, in the assumed network configurations the outage probability on the picocell downlink is not acceptable (16.3% for 3/9 reuse and 15.1% for 1/3 reuse in the macrocells), figures 5 and 6. This means, that the interference averaging gain by the use of T-FH alone is not sufficient for the picocell downlink. Of course, especially a lower macro-BS transmit power and the consideration of power control and DTX would improve the results (see section 4.2). However, we aim at a solution for the picocells that also works under the condition of high interference levels from the macro- and microcell layers. 2 Note, that in case (c) only the micro- and picocells interfere each other,
however, in case (d) only the macro- and picocells.
Figure 6. Outage probability in the picocell layer for 1/3 reuse2 .
4.1.3. Results for T-FH with reduced static frequency lists The use of static frequency lists for T-FH, which contain a reduced number of frequencies with respect to the network configuration, improve the performance of the picocell downlink already to some extent. The results for the reuse of only 8 macrocell frequencies indicate that the influence of the picocells on the macrocells is negligible, see case (d) in figure 3. This is a real capacity reserve, which should be exploited for the picocells. However, the outage probability on the picocell downlink remains with 11.5% still unacceptably high, figure 5.
INCREASED CAPACITY
477
By including all the frequencies of the microcells the outage probability on the picocell downlink can slightly be improved, but only to the debit of the microcell as well as the picocell uplink performance, see case (b) in figures 3 and 5. The results for case (c), where only the microcell frequencies are reused, further emphasize the need for also excluding the strongestly interfered microcell frequencies, although the micro-BSs and -MSs only transmit with 27 dBm EIRP. Of course, the exclusion of locally used microcell frequencies would further improve the picocell performance. However, in real irregular cell layouts the improvement in the interference on the macro- or microcell layer by reduced static frequency lists would only be both achieved and intensified by measurements and, based on this, an adaptation to the local interference situation as the AFA algorithm eventually provides.
figurations at, approximately, 6 frequencies for 3/9 reuse. This value is shifted towards 4 frequencies for 1/3 reuse. Obviously, in the denser macrocell layout, less frequencies meet the low interference level, which is required for an effective reuse in the picocells. It is remarkable that, although the macro- and microcellcaused interference is in the cases (e) and (f) much higher for the denser 1/3 reuse, approximately the same picocell performance like in the 3/9 reuse can be reached in the cases (h)– (j). From these results we can see, that it is more efficient to exclude the strongest interferers than to achieve a higher interference averaging gain by using as much frequencies as possible for T-FH.
4.1.4. Results for T-FH with adaptive frequency allocation If AFA determines 18 frequencies or less for T-FH in the 3/9 macrocell reuse cases, figure 3, and 10 frequencies or less in the 1/3 reuse cases, figure 4, respectively, the impact on the macrocells from the frequency reuse in the picocells is negligible. In the 1/3 reuse cases, the picocells obviously make more use of the less interfered microcell frequencies than of the macrocell frequencies, which leads in comparison to 3/9 reuse to an increased picocell-caused outage probability in the microcells, compare especially case (f) in figures 3 and 4. Nevertheless, the microcell outages can still be decreased, e.g., on the uplink from 1.7% in case (f) to 0.3% in cases (i) and (j), figure 4. At low numbers of frequencies the picocellcaused outage probability in the microcells even reaches similar values, namely 0.2–0.4%, for both the 3/9 and the denser 1/3 macrocell reuse. In the picocells the use of AFA considerably decreases the outage probability, figures 5 and 6. According to the transmission characteristics in the macrocells and the interference conditions in the microcells, up- and downlink reach their minimal outage probability at different numbers of frequencies. On the uplink (left pillar of each pair) we find the minimum at 18 frequencies for 3/9 reuse and at 10 frequencies for 1/3 reuse, respectively. In both cases, the downlink outage probability (right pillar of each pair) reaches its minimum at 6 frequencies. This minimum is due to an optimal combination of excluding strongly interfered frequencies on the one hand and of interference averaging on the other hand. For less frequencies, the interference averaging gain decreases, because the probability of hits from other picocell users hopping over 5 or less frequencies increases. The resulting increase of the frame erasure rate will additionally be influenced by the limited frequency diversity, which slow moving or quasistationary picocell users encounter for very few hopping frequencies. If we consider up- and downlink together, we can find the optimal number of frequencies selected by the AFA algorithm for T-FH in the considered hierarchical network con-
In order to better assess the previous results, we want to discuss the influence of some other characteristics than the number of frequencies in the following, namely the picocell user density, the use of power control and DTX, and the building penetration loss. If we double the picocell user density to 6000/km2 in the 1/3 reuse case with 4 AFA frequencies (compare with case (i) in figures 4 and 6), the macrocell performance is not affected. The picocell-caused increase of the microcell outage probability is 0.7% on the downlink and even less on the uplink. The increase of the picocell outage probability by approximately 1% on both, up- and downlink, is mainly due to the higher interference within the picocell layer. Hence, these results indicate that also higher picocell user densities than 3000/km2 are possible and even 6000/km2 is not yet critical. Moreover, we expect that the picocell performance will still improve under consideration of power control and DTX. Further simulations with a 6 dB lower transmit power level of the sectorized macrocell-BSs, i.e., 43 dBm EIRP, were expected to have a similar effect in the reduction of interference like the application of power control and DTX. In the respective results for 3/9 macrocell reuse the outage probability on the picocell downlink drops from 16.3% (figure 5) to ca. 11% (not shown) in case (a) and from 10.8% (figure 5) to ca. 9% (not shown) in case (b), respectively. The picocell-caused interference in the macro- or microcells is not affected. Finally, among the propagation modeling assumptions, the building penetration loss has a strong influence on the results. For example, a 3 dB stronger isolation of picocells within buildings, i.e., Lindoor = 15 dB, would further decrease the outage probability on the picocell downlink cases (e), (f), and (h) for 3/9 reuse by ca. 3%, 2%, and 1%, respectively (see figure 5).
4.2. The influence of other characteristics than the frequency allocation
4.3. Summary Under worst-case conditions with respect to the interference from other layers to picocells in a dense urban environment,
478
the presented results indicate that AFA provides a considerable interference reduction gain in both investigated network configurations (at most a reduction of the outage probability on the sensitive picocell downlink by 12% and 11%, respectively). Moreover, approximately the same picocell outage probability values like in the macro- and microcell layers can be achieved, though those are frequency-planned and the picocells not. Since a real network with irregular cell layout is less tight than the 3/9 reuse case or at most within the range of 3/9 and 1/3 reuse, each overlaid with a 5-reuse microcell layer, we may assume that the performance results are a lower bound for the capacity of the indoor picocells. Hence, solely by reusing existing macro- and microcell frequencies, an additional indoor and hot-spot traffic of more than 300 Erlang/km2 can be carried by the picocells. This corresponds to an increase in spectral capacity by over 50%, namely 33 Erlang/km2/MHz. The influence of the picocells on the macrocells is in both reuse scenarios for less than 10 frequencies negligible. The outage probability in the microcells is increased by the picocells at most by 1.7%, in the cases, which are optimal for the picocells, only by 0.5%. In joint consideration of up- and downlink, the optimal number of frequencies selected by the AFA algorithm for TFH is for the considered hierarchical network configurations at approximately 6 frequencies for 3/9 reuse and is shifted towards 4 frequencies for 1/3 reuse. The results also show, that it is more efficient to exclude the strongest interferers than to achieve a higher interference averaging gain by using as much frequencies for T-FH as possible. Moreover, higher picocell user densities than 3000/km2 are possible and even 6000/km2 is not yet critical. And if we assume 6 dB smaller transmit powers at the macrocell-BSs, which can be compared with the case of downlink power control and DTX, the macrocell-caused outage probability at the picocell downlink can be further decreased by 5% for 46 reused frequencies and by 2% for adaptively 18 reused frequencies, respectively, without increasing the picocellcaused interference in the macro- or microcells.
5. Discussion of the results Since the configuration and the environment of cellular networks are usually very inhomogeneous, hierarchical cellular structures require reuse strategies that do not only adapt to the current local interference situation, but additionally distribute the remaining interference to as many resources as possible. Likewise the measurement effort should be kept to a minimum. For a hierarchical GSM network these requirements are fulfilled by the T-FH/AFA technique very well. In addition to that, this technique spares the planning of the picocell frequencies that are reused from the macro- and microcells. Frequency planning for these picocells would not be possible or at least incorporate a very high effort. For FDMA systems with a reuse factor greater than 1 adaptive and intelligent channel allocation techniques like
J. DEISSNER AND G.P. FETTWEIS
AFA can transform capacity reserves into a considerable additional traffic capacity solely by the reuse of spectrum already allocated to this system. The reasons for this are the different characteristics of the several layers of the hierarchical cellular network, mainly regarding to the power settings and the propagation, which overall lead to an inhomogeneous occupation of the radio resources. Moreover, in real irregular cell layouts, lacks of homogeneous coverage can appear, because the base stations can often not be placed at the optimal locations. Attenuations due to strong shadowing should not be eliminated by increasing the power, e.g., in order to provide indoor coverage. On the contrary, they should be exploited for reusing the attenuated resources. For example, dedicated indoor cells can benefit from the building penetration loss. At least partially they can reuse the resources from the outdoor cells and thus provide extra capacity without additional radio resources. Though the fixed frequency duplex offset of the GSM system prevents different frequency sets in uplink and downlink, adaptive channel allocation techniques in future cellular systems should be able to handle uplink and downlink separately. The two reasons for that are the different interference situations on uplink and downlink, as we can also see in most of the results presented in section 4, as well as different capacity and, where appropriate, quality requirements on uplink and downlink. For the reduction of the measurement effort and the speed-up of the AFA list update in new environments (especially for ad hoc installations), it is worth to incorporate information on the network structure in the AFA algorithm (such as the layer the channel belongs to and its average load as well as if it is a channel of the potential serving cell). Thereby, less promising channels, i.e., with potentially high interference, need not to be measured or may only be gradually included for the optimization of the AFA list after the initialization or an update. As frequency hopping is a special CDMA technique, we can find in our results for GSM with T-FH/AFA some similarities with results from other CDMA system analyses. For instance, as typical for CDMA systems, the downlink limits the performance of the radio network, as it is the case for the picocells in our study. Moreover, regarding the ability to avoid and distribute interference, T-FH/AFA is apparently very similar to a broadband CDMA system overlaying a GSM system with inter-layer reuse, which requires agile notch filtering according to studies in [14,15]. The increase in spectral capacity which we achieved in our study seems to be small in comparison to the results from those studies. In [14], for example, a CDMA overlay network cannot only carry 50% in addition to the GSM network traffic but nearly 300%. This requires 9 notch filters with depths of 35 dB on receive and 20 dB on transmit. Moreover, the GSM and the Broadband CDMA base stations are intended to be co-located and the overlay network does not “need” some sort of shadowing in the environment for an isolation of the cells that reuse the GSM resources
INCREASED CAPACITY
as in our study. However, we investigated a much denser GSM cell layout, where a large part of possible capacity reserves has already being exploited in the GSM macro- and microcells. In [14] a 4/12 reuse GSM cell layout with an inter-BS distance of 1.7 km and 67% traffic load is studied, which has a spectral capacity of merely 3 Erlang/km2/MHz. The CDMA overlay network increases this spectral capacity by some 7 Erlang/km2/MHz. Note that we achieved an increase by 33 Erlang/km2/MHz over an existing spectral capacity of 64 Erlang/km2/MHz. Hence, the use of the absolute spectral capacity as a measure for the density of the cell layout is strongly recommended for a comparison of the results of such kind of studies. Of course, the cost of additional base stations and new base station functionality is only worth with adequately high traffic demand. However, the quickly growing user penetration together with an increased use of data services on the basis of new techniques like the General Packet Radio Service (GPRS) and High-Speed Circuit-Switched Data (HSCSD) in GSM, which are all to be served by the limited bandwidth licensed to a network operator, indicate a rising necessity for such an optimization by means of hierarchical cellular networks. Naturally, hierarchical cellular structures become a decisive part of third generation mobile radio system concepts from their beginning. However, apart from the layer-specific optimization or the only interdependence through inter-layer handoffs, a further potential capacity improvement arises from the mutual reuse of the radio resources in the different layers on the basis of techniques that adapt themselves to the local interference situation. In order to further analyze such capacity improvements, we use simulation techniques that adequately cover the adaptivity and the dynamics, which are decisive for the expected gains [16–18]. For the deployment of hierarchical cellular structures it has also to be ensured that the reuse of the radio resources also works in rare extreme cases. Hence, in addition to the statistical analysis of the interference performance, system scenario calculations are always required, e.g., with respect to the transmitter output RF spectrum and spurious emissions as well as to the receiver blocking characteristics [19].
6. Conclusions In an interference study for a hierarchical GSM radio network we showed the capability of Total Frequency Hopping (T-FH) in conjunction with Adaptive Frequency Allocation (AFA) to transform capacity reserves into an increase of the spectral capacity from 64 Erlang/km2/MHz to 97 Erlang/km2/MHz. Thereby, a good interference performance can be achieved in all layers. Moreover, frequency planning for the indoor picocells is avoided. Such capacity reserves may still exist in very dense radio network configurations like those investigated as a worst case. Generally, they can be exploited by reuse strategies,
479
which do not only adapt to the current local interference situation, but additionally distribute the remaining interference to as many resources as possible. T-FH/AFA is an example for such a technique, which is well suited for FDMA systems and implies less measurement effort than dynamic channel allocation schemes. Moreover, it has already become a part of an extension of the GSM standard towards a Cordless Telephony System. Similarities with Broadband CDMA overlay scenarios were discussed. As in that case, the spectral capacity is a measure that enables a comparison of investigation results for very different radio network configurations. Acknowledgements The simulation software was developed within a project at Dresden University of Technology that was sponsored by the Alcatel Corporate Research Center, Stuttgart. Here, we wish to thank Ulrich Barth and Charles Skelton for providing us with the software version that was extended afterwards by Alcatel and their support. We would also like to thank Dr. Michael Hartmann, Mannesmann Mobilfunk GmbH, Dresden, for the fruitful discussion on the setup for this simulation study. References [1] M. Frullone, G. Riva, P. Grazioso and M. Barbiroli, Advanced planning criteria for cellular systems, IEEE Personal Communications (December 1996) 10–15. [2] M. Madfors, K. Wallstedt, S. Magnusson, H. Oloffson, P.-O. Backman and S. Engström, High capacity with limited spectrum in cellular systems, IEEE Communications Magazine (August 1997) 38–44. [3] M.I. Silventoinen, M. Kuusela and P.A. Ranta, Analysis of a new channel access method for Home Base Station, in: Proc. ICUPC’96, Cambridge, MA (October 1996) pp. 930–935. [4] J. Deißner, A. Noll Barreto, U. Barth and G. Fettweis, Interference analysis of a Total Frequency Hopping GSM Cordless Telephony System, in: Proc. PIMRC’98, Boston, MA (September 1998) pp. 1525– 1529. [5] First investigations of AFA performance in the CTS, TDoc SMG2 WPB 87/98, ETSI (1998). [6] Digital cellular telecommunications system (Phase 2+); GSM Cordless Telephony System (CTS), Phase 1; Lower Layers of the CTS Radio Interface; Stage 2 (GSM 03.52 version 7.0.0 Release 1998), ETSI (1999). [7] Interference performance of GSM-CTS: Simulation results, TDoc SMG2 WPB 95/98, ETSI (1998). [8] Simulation results of the interference performance of the AFA/TFH radio interface concepts for the GSM–CTS for scenarios including a microlayer in the GSM–PLMN, TDoc SMG2 WPB 174/98, ETSI (1998). [9] Digital cellular telecommunications system (Phase 2+); Multiplexing and multiple access on the radio path (GSM 05.02 version 8.0.0 Release 1998), ETSI (1999). [10] A. Noll Barreto, J. Deissner and G.P. Fettweis, A frequency hopping algorithm for cordless telephone systems, in: Proc. ICUPC’98, Florence, Italy (October 1998) pp. 1273–1277. [11] Digital cellular telecommunications system; Radio network planning aspects (GSM 03.30 version 5.0.0), ETSI (1996). [12] COST Action 231: Digital mobile radio towards future generation systems, Final report, European Communities (1999).
480
J. DEISSNER AND G.P. FETTWEIS
[13] Definition of the simulation environment for the evaluation of the interference performance of GSM–CTS radio interface concepts, TDoc SMG2 WPB 88/98, ETSI (1998). [14] D.M. Grieco and D.L. Schilling, The capacity of Broadband CDMA overlaying a GSM cellular system, in: Proc. IEEE VTC’94 (1994) pp. 31–35. [15] T. Widdowson, A CDMA overlay of the GSM network, in: Proc. IEEE PIMRC’97, Helsinki, Finland (September 1997) pp. 160–163. [16] J. Fischer, J. Deissner, G. Fettweis, D. Hunold, R. Lehnert, M. Schweigel, J. Voigt and J. Wagner, Object-oriented modeling of a generic mobile radio system for dynamic system simulation, in: Proc. of the Symposium on Performance Evaluation of Computer and Telecommunication Systems SPECTS’99, Chicago, IL (July 1999) pp. 240–247. [17] J. Deissner, G. Fettweis, J. Fischer, D. Hunold, R. Lehnert, M. Schweigel, A.Steil, J. Voigt and J. Wagner, A development platform for the design and optimization of mobile radio networks, in: Kurzvorträge der 10. GI/ITG- Fachtagung Messung, Modellierung und Bewertung von Rechen- und Kommunikationssystemen MMB’99, eds. D. Baum, N. Müller, R. Rödler, Universität Trier, Germany, Mathematik/Informatik, Forschungsbericht Nr. 99-17 (1999) pp. 129133. [18] WiNeS – Wireless Network System Simulator, TU Dresden, http://entmuc.et.tu-dresden.de/MNS/forschung. [19] System scenario calculations for GSM–CTS, TDoc SMG2 WPB 12/ 99, ETSI (1999).
Jürgen Deissner received his Dipl.-Ing. and Ph.D. degree in electrical engineering/information technology from Dresden University of Technology, Germany, in 1994 and 2001, respectively. From 1995 to 2000 he was a Research Associate at the Mannesmann Mobilfunk Chair for Mobile Communications Systems at Dresden University of Technology working on radio resource management techniques for increasing the capacity of mobile radio networks as well as modeling mobile
communications systems for interference-based capacity and quality analysis by means of radio-network-level simulations. In 2000 he worked with AT&T Labs Research, New Jersey, in the field of GPRS/EDGE modeling and analysis, and in 2001 he joined Radioplan GmbH, a startup spun out of Dresden University of Technology focusing on radio network analysis and optimization software. Jürgen Deissner is a member of the IEEE and the German VDE. E-mail: [email protected]
Gerhard Fettweis [IEEE S’82–M’90–SM’98] received his MSc/Dipl.-Ing. and Ph.D. degree in electrical engineering from the Aachen University of Technology (RWTH), Germany, in 1986 and 1990, respectively. From 1990 to 1991 he was a Visiting Scientist at the IBM Almaden Research Center in San Jose, CA, working on signal processing for disk drives. From 1991 to 1994 he was a Scientist with TCSI, Berkeley, CA, responsible for signal processor developments for mobile phones. Since September 1994 he holds the Mannesmann Mobilfunk Chair for Mobile Communications Systems at the Dresden University of Technology, Germany. He is an elected member of the SSC Society’s Administrative Commitee, and of IEEE ComSoc Board of governors, since 1999 and 1998, respectively. He has been an associate editor for IEEE Transactions on Circuits and Systems II, and now is an associate editor for IEEE Journal on Selected Areas in Communications wireless series. E-mail: [email protected]
