HETEROGENEOUS LINK ASYMMETRY IN TDD MODE CELLULAR SYSTEMS Magnus Lindström Radio Communication Systems Department of Signals, Sensors and Systems Royal Institute of Technology (KTH) SE- 44, STOCKHOLM, Sweden e-mail: magnus.lindstrom@radio.kth.se Abstract Time Division Duplex systems provide great flexibility that can be used to implement asymmetrical links. Adverse interference conditions easily arise, however. Especially if different asymmetries are required in neighbouring cells. This paper assesses the performance of TDD mode systems in evironments that require heterogeneous link asymmetry. In particular the Soft Switching Point and Disjoint Base station Set algorithms are considered. Results indicate that gradual asymmetry variations can be handled with almost no performance loss, but random variations result in capacity loss. I. BACKGROUND In future wireless applications an increasing fraction of the traffic is expected to appear in the downlink (fixed-to-mobile link). This requirement matches poorly with the system design and spectrum allocation for current and future systems which employ symmetric Frequency Division Duplex (FDD). TDMA/TDD systems allow for a more flexible resource allocation by moving the boundary between uplink and downlink time-slots. In [], the performances of a few slot allocation strategies were evaluated and compared. It was shown that capacity could be increased if the uplinkto-downlink boundary, a.k.a. the switching point, is not fixed. Similar results were presented in [2]. The key problem caused by moving the boundary is the introduction of Inter-Mobile Interference (IMI) and Inter-Base station Interference (IBI). In [3] the Soft Switching Point (SSP) strategy originally proposed in [] was further improved with the introduction of a heuristic IMI avoidance scheme, the Disjoint Base Station Set (DBS) method. The asymmetry of the traffic offered to the system is likely to vary over the coverage area. For instance, voice traffic could dominate in some areas whereas event driven data transfers (i.e. web browsing) could dominate in other areas. The presented dynamic resource allocation schemes provide a means to support asymmetries which vary over the coverage area, but heterogeneous asymmetry gives rise to IMI and IBI which lowers the quality of the affected links, possibly resulting in outage. Granted, supporting only one asymmetry, A G, throughout the system would efficiently avoid IBI and IMI. On the other hand, ignoring the variations would cause excessive blocking in parts of the system where the asymmetry of the offered traffic matches poorly with the asymmetry supported by the system. In this paper we evaluate and assess the performance gains of the Soft Switching Point strategy with DBS in environments with gradually and randomly changing link asymmetries, respectively. The results are compared with the corresponding results of a system where the switching point is fixed in the same position throughout the system.
.9.9 asymmetry, A F.8.7.6 asymmetry, A H.8.7.6 asymmetry, A R.8.6.5.5.4 (a) Flat asymmetry (b) Hot Spot asymmetry (c) Random asymmetry Figure : Illustration of different asymmetry scenarios. The surface represent the expected local asymmetry as a function of the location. The vertical lines represent the locations of the base stations. II. ASYMMETRY SCENARIOS To facilitate the description of different asymmetry scenarios, we define offered asymmetry, A, as the ratio of the traffic offered in the downlink to the total offered traffic; i.e. A = ω DL ω, where ω DL is the offered downlink traffic and ω is the total offered traffic. Three asymmetry scenarios are considered: A flat asymmetry, where the offered asymmetry, A F,i, is equal in all cells i, Figure (a); a hot spot asymmetry, where the asymmetry, A H,i, changes gradually over the cells to reach its maximum, A max, in the centre of the service area, Figure (b) and a random asymmetry, where the expected asymmetries of different cells, A R,i, are mutually independent, Figure (c). Assume that the cellular system is deployed in the square area: x < y < Then, the three scenarios can be defined more formally as follows: A F,i = A G { A max x 3 A H,i = 2 (A max A G ) A max y 3 2 (A max A G ) A R,i U(2A G A max, A max ), if x y, if x < y. where A G is the global mean asymmetry of the offered traffic. The mean asymmetry of the system is kept at a constant A G =.75 for all scenarios. The peak asymmetries for the hot spot and random scenarios are varied from A max =.75 to A max =, that is, from 75% downlink traffic to % downlink traffic. III. SSP AND DBS The Soft Switching Point (SSP) algorithm is a resource allocation algorithm for TDD mode systems that was first presented in [] as a generalisation of the Movable Boundary (MB) scheme proposed in [4]. When a resource unit is requested the TDMA frame is sequentially searched for a free time-slot capable of providing sufficient quality. Depending on the link direction, the search starts either from the beginning or from the end of the frame. The search is aborted when either: a good slot is found or the entire frame has been searched. A slot is considered good if the integrated feasibility check finds that all links can be pro-
vided with a sufficiently high signal-to-interference ratio. The SSP algorithm provides for multiple switching points, whereas the MB scheme does not. BS-to-MS link gains and IB link gains are assumed known through intelligent measuring. The IM link gains, on the other hand are considered infeasible to measure. Suppose there are N mobile stations, then there are N(N )/2 link gains to measure, i.e. the number of link gains grow with the square of the number of mobiles. Instead a heuristic method, dubbed the Disjoint Base station Set (DBS) method and presented in [3], is used to avoid IMI. For each mobile i, a set of only the n strongest base stations is considered. That is, Mi n = { Bi,..., Bi n }, where B j i is the jth strongest base station as seen by mobile i. If the sets of two mobiles, k and l, are disjoint, the corresponding inter-mobile link gain, denoted G IM k,l, is approximated with the largest cross-set inter-base station link gain, ( ) G IB i,j, G IM k,l = max i M n k, j M m l where G IB i,j is the link gain between base stations i and j. If the sets of the two mobiles are not disjoint, the mobiles must only share a frequency-time-slot if both are transmitting or if both are receiving (Figure 2). The idea behind the above method is that when two mobiles are near each other their link gains toward neighbouring base stations can be expected to be highly correlated and thus the probability that the two mobiles will have overlapping sets increases rapidly with the size of the sets. Since mobiles with overlapping sets are not allowed to share time-slots for different link directions, less IM interference is introduced. The three resource allocation schemes studied in this paper are: The SSP algorithm with DBS, the SSP algorithm with perfect IM link gain estimates (Perfect SSP) and MS MS2 MS3 Figure 2: The strongest base stations sets for three mobile stations. The sets of MS and MS2 are not disjoint. Thus MS is not allowed to transmit in a time-slot where MS2 is receiving or vice versa. Since MS and MS3 have disjoint sets, we approximate the MS-to-MS3 link gain with the greatest inter-set base station gain (dashed arrow). a Common Switching Point (CSP) algorithm, where the switching point is matched to the average asymmetry and the same in all cells. The Perfect SSP algorithm is a hypothetic algorithm which can completely avoid making resource assignments that would yield insufficient SIRs. It serves as an upper bound on the performance of the studied type of algorithms. IV. SYSTEM MODELS The resource allocation schemes are evaluated in a Manhattan type environment with asymmetrical traffic. Mobile stations are assumed to belong to outdoor users only and to be uniformly distributed on the streets in a by block Manhattan environment. Blocks are 2 by 2 meters and separated by streets that are 3 meters wide. The area is serviced by 55 BSs which are placed in a regular pattern along the streets. The BSs are placed halfway between intersections, thus increasing the isolation between BSs compared to an in-intersections deployment. We define the relative traffic load of the system, ω c, as
the average number of users in the system divided by the number of cells and the number of channels available, i.e. terminals/channel/cell. Further, asymmetry is modelled by assuming that terminals are active either in uplink or in downlink, not both, with probabilities p and p respectively. The recursive path loss model of [5] is used. In the two slope LOS model signals decay proportional to the square of the distance for distances, r, less than 3 meters. Beyond 3 m, the attenuation is proportional to the fourth power of r. Only shadow fading is considered and assumed log-normal with a standard deviation of 8 db. Mobile stations connect to the strongest BS. Constant received power control is exercised with a target of 98 dbmw and a maximum transmission power of.76 dbw. The receiver noise level is 8 dbmw. Outage occurs if the SIR drops below 7 db. The access mode is TDMA/TDD with 6 time-slots per frame. The resource allocation algorithms are evaluated in terms of a) the user blocking rate ν bl, i.e. the rate at which terminals are unable to find a free time-slot, b) the user outage rate ν out, i.e. the rate at which users that are assigned channels fail to reach the required SIR, and c) capacity in terms of users/channel/cell. Outage, possibly leading to dropping, is generally considered more annoying than blocking. Therefore we define a (user) Grade-of-Annoyance (GoA) measure, giving outage ten times the weight of blocking, GoA = ν bl + ( ν bl ) ν out. The capacity of the system is defined as the load at which either the uplink GoA or the downlink GoA is 5%, whichever occurs first. Thus, outage and blocking are effectively limited to a maximum of.5% and 5%, respectively. relative capacity.9.8.7.6.5.4.3 hotspot scenario random scenario Perfect SSP SSP with DBS CSP.2.7.75.8.85.9.95 A max Figure 3: Relative capacity as a function of peak asymmetry, A max. Solid lines represent the hot spot asymmetry scenario. Dashed lines show the result for the scenario with random asymmetry. The overall offered asymmetry, A G, is.75. and BSs are located between intersections. V. RESULTS Results show that any one of the studied resource allocation strategies that allows for arbitrary switching points, provides significantly higher capacity than the scheme with a global common switching point, Figure 3. This is largely due to the fact that with a fixed switching point there is an over provisioning of resources to the low-volume link direction. That comes at the expense of a corresponding lack of resources for the high volume link direction. Results for the flat asymmetry scenario are not plotted separately since they are implicitly given by the hot spot and random asymmetry scenarios. For a peak or maximum asymmetry of.75, all cells have a local asymmetry equal to the global mean. This corresponds to a flat asymmetry scenario. Ideally, results for all scenarios should coincide at this point. The slight deviations observed in Figure 3 are due to the simulation approach taken to the evaluation. The dynamic schemes gracefully cope with the introduction of heterogeneous asymmetry. Going from a flat asymmetry distribution, i.e. maximum asymme-
try equal to the global mean asymmetry, to a hot spot distribution where there is only downlink traffic in the centre of the system and an approximately equal mix of uplink and downlink traffic at the perimeter, hardly degrades performance. As expected, the degradation is more pronounced when asymmetry changes rapidly between cells than when the change appears gradually over a few cells. With the CSP algorithm, the capacity drops approximately twice as fast as with the SSP-with-DBS algorithm as asymmetry variations increase. Since the CSP algorithm yields a much lower capacity than does the SSP-with-DBS algorithm, this is particularly problematic. The uplink and downlink blocking characteristics at the capacity limit are shown in Figure 4 for the random asymmetry scenario. For the uplink, SSP with DBS offers the lowest blocking rate. A blocking rate lower than that of the reference scheme with perfect knowledge about all inter-mobile link gains must, of-course, come at the expense of higher outage. The outage, however, stays rather constant at an acceptable level of about.3%. In the downlink, the CSP system exhibits a significantly lower blocking rate than the other systems. Owing to the absence of IBI and IMI in the CSP system, the outage rate is negligible. Therefore, the low downlink blocking rate clearly indicates that the CSP system is severely limited by uplink blocking. More generally, it was found in [] that the performance of the CSP algorithm is limited by blocking in the low-volume link direction. The dynamic schemes better balance the quality of the low- and high-volume link directions. The results for the hot spot scenario are similar. VI. CONCLUSIONS Heterogeneous asymmetry reduces the capacity of TDD mode cellular systems. Gradual changes can, however, be handled with almost no degradation by the Soft Switching Point algorithm in conjunction with the Disjoint Base station Set method. Random changes are harder to cope with and especially so if a common switching point is enforced globally. blocking rate 2 uplink downlink Perfect SSP SSP with DBS CSP 3.7.75.8.85.9.95 A max Figure 4: Blocking, at the capacity limit, as a function of peak asymmetry, A max, for the random asymmetry scenario. Solid lines represent the uplink. Dashed lines show the result for the downlink. The three allocation strategies are distinguished by different markers. REFERENCES () M. Lindström and J. Zander, Dynamic link asymmetry in bunched wireless networks, in Proc. IEEE VTC 999 Fall, vol., pp. 352 356, 999. (2) W. S. Leon and D. G. Jeong, Comparison of time slot allocation strategies for CDMA/TDD systems, in Proc. IEEE VTC Spring, vol. 2, pp. 86 9,. (3) M. Lindström, Improved TDD resource allocation through link gain estimation,. To appear at 3Gwireless, San Fransisco, U.S.A., May 3-June 2, 2. (4) L. Chen et al., A dynamic channel assignment algorithm for asymmetric traffic in voice/data integrated TDMA/TDD mobile radio, in International Conference on Information, Communications and Signal Processing (ICICS), pp. 25 29, 997. (5) J.-E. Berg, A recursive method for street microcell path loss calculations, in Proc. IEEE PIMRC 95, pp. 4 43, 995.