Intercell Radio Resource Management through Network Coordination for IMT-Advanced Systems

Recent developments in the area of wireless communications have witnessed a remarkable proliferation of wireless technologies. While 3G and Beyond 3G cellular systems are still being deployed, the IMT-advanced standards have set a goal for an evolutionary growth towards 4G wireless networks. The potential candidate technologies for 4G networks include 802.16m (WiMAX) and LTE-Advanced (3GPP). These next-generation systems have adopted orthogonal frequency division multiple access (OFDMA) as the air-interface technology. In OFDMA systems, as neighboring cells can reuse the same frequency, intercell interference is an important problem that needs to be solved. Due to intercell interference, cell-edge users may suffer low data rates (or high error rates) even when the most robust modulation and coding techniques are used.

To enhance the performance of cell-edge users in OFDMA systems, frequency reuse techniques between neighboring cells, have been developed. A simple technique is to exploit a large frequency reuse factor such as 3. With a reuse factor of one, all cells use all frequencies in which case the intercell interference would be the highest. but cell throughput is greatly reduced because each cell now can use only 1/3 of the total available frequencies (channels). An enhanced reuse technique is to use two reuse factors simultaneously; for example, cell-edge users are supported by a reuse factor of 3 while the others use a reuse factor of 1 [1]. These methods, however, limit flexibility in terms of radio-resource utilization and cell deployment. For flexibility, dynamic channel allocation has been widely examined (see [2, 3] and references therein). In [2], dynamic channel allocation is realized at both a radio network controller (RNC) and at a BS which assign channels at the super-frame level and frame-level, respectively. Dynamic channel allocation is also referred to as dynamic fractional frequency reuse (FFR) [4], and it is known that FFR can enhance cell-edge throughput by about 15% [5] but at the expense of a reduced average cell throughput.

Another technique to enhance cell-edge user performance is macrodiversity. With macro-diversity, multiple BSs can serve a user, thus making the link condition of cell-edge users more reliable and robust [6]. Such a technique has already been adopted in the IEEE 802.16 family of standards but the current macro-diversity schemes limit the signaling formats (or space-time codes) that can be employed by the coordinated BSs. An emerging technique that is still in its infancy but has nevertheless generated enough interest, is that of network MIMO. The concept of network MIMO (a.k.a. multi-BS MIMO) to perform joint MIMO transmission and reception between multiple coordinated BSs and multiple users over the same radio resources [7, 8]. Network MIMO can be further divided into closed-loop network MIMO, where coordinated BSs have access to and exploit fast changing channel state information to serve multiple users, or open loop network MIMO where only average channel quality information is used to serve the users. Note that open-loop network MIMO subsumes macro-diversity schemes as special cases. In this paper we present our solution to classify users by assuming a simple macro-diversity scheme but our techniques can be readily extended to other more general open-loop network MIMO schemes.

Both dynamic FFR and macro-diversity require coordinated RRM between neighboring BSs within the network. We use RRM to refer to both radio resource management and radio resource managers. If dynamic FFR is deployed in the system, designated neighboring BSs of a cell to which a certain cell-edge user is attached should avoid concurrent transmission over the same set of channels. If macro-diversity is used, one or more neighboring BSs should serve a certain cell-edge user at the same time; this means concurrent transmissions over the same set of channels from multiple BSs to the same user is required. To handle such requirements, the system needs network-level coordination of radio resources through a cross-layer approach between the network level and the radio level. In this paper, we consider two different backhaul network architecture options (hierarchical and flat) and develop a framework for network-assisted RRM within these architectures so as to maximize network throughput and enhance edge-user throughput.

In a conventional cellular architecture, network-level coordination is easier to achieve as radio resources are managed by an upper-level entity such as an RNC. For example, as seen in [2], techniques such as dynamic FFR and macro-diversity can be managed more easily at an RNC since they required coordination among multiple BSs. However, there are other techniques such as channel-aware opportunistic scheduling [9‐11] and closed-loop MIMO operations that require real-time channel feedback from users [12]. Wireless channel characteristics, specifically for mobile users, fluctuate due to channel fading, and thus it is required that channel feedback be delivered within one slot/frame (5 msec in WiMAX systems and 1.67 msec in cdma2000 EV-DO/HDR systems) duration [11]. In this case, if RRM is implemented at a central entity, the two-hop delivery of feedback information from a mobile station to the upper-level entity through the BS may make this information outdated and not usable.

In HDR and HSDPA systems [13, 14], where opportunistic scheduling based on channel feedback plays a key role in enhancing cell throughput; user and channel scheduling is performed at the BSs. These systems do not implement either dynamic FFR or macro-diversity both of which require network-level coordination and are at odds with implementing RRM functions at the BSs. In upcoming 4G systems where techniques such as dynamic FFR and macro-diversity are expected to be deployed together with conventional techniques such as opportunistic scheduling and closed-loop MIMO, it becomes imperative that a framework for RRM be developed to enable such coexistence.

The classification of cell-edge and cell-interior users are not based purely on geographic location as in conventional frequency reuse techniques. We classify users as cell-edge or cell-interior users with the goal of maximizing network throughput subject to the condition that the user throughput does not decrease upon switching; for example, a user at the edge of a cell may still get classified as a cell-interior user if such classification leads to higher network throughput with no attendant loss in the user throughput or conversely if such classification increases the user throughput without a loss in the network throughput, this can happen due to the multiuser diversity gain which is obtained when channel-dependent scheduling is employed to serve cell interior users. Furthermore, we also show that as compared to a switching scheme which only aims to maximize the network throughput, our classification scheme results in a better cell-edge performance without a loss in network throughput. Within the proposed framework, we address three main problems: ( ) initial user classification, ( ) strategy to switch users from one class to another subsequently, and ( ) radio-resource reservation in neighboring cells for users who are classified as cell-edge users. We develop solutions for these problems in situations where dynamic FFR and macro-diversity mechanisms are available to cell-edge users. As discussed earlier, these mechanisms require radio resources from multiple neighboring cells.

The remainder of this paper is organized as follows. In Section 2, we provide a brief overview of the current framework for RRM support within a conventional 3G cellular network architecture, which is hierarchical, as well as a Beyond 3G system such as 802.16e (WiMAX) which is flat. In Section 3, we describe the proposed two-level framework for RRM and show its applicability within both a hierarchical architecture and a flat architecture; 4G systems are expected to implement the latter. The framework uses an upper-level RRM and a lower-level RRM to handle cell-edge and cell-interior users, respectively. In Section 4, we introduce metrics for describing a mechanism for classifying users as cell-edge and cell-interior users, and explain initial user classification. In Section 5, we provide algorithms for switching a user between these two classes. In Section 7, simulation results demonstrate that our classification algorithms improve cell-edge performance. Section 8 provides conclusions from our work.

Traditionally, 2.5G and 3G cellular networks have employed a hierarchical backhaul structure. Recently, Beyond 3G and 4G systems such as 802.16e and 802.16m (WiMAX) have been evolving more towards a flat architecture. The WiMAX standard has defined three different profiles for an access service network (ASN) which is the backhaul network that connects multiple BSs to an ASN gateway [15]. In Profile A, which is now defunct, most radio resources are managed by the ASN gateway as in traditional cellular networks and thus has a hierarchical structure. In Profile B, the functionalities of a BS and an ASN gateway are colocated on the same platform/solution, which makes the architecture flat. Profile C also defines a flat architecture where all the RRM functions are performed at the BS.

Figure 1 presents the WiMAX ASN models for Profiles A and C. In both cases, the interface (named R6) between an ASN gateway and a BS is explicitly defined. In both these profiles, a BS implements a RRA (radio resource agent), the difference being where the RRC (radio resource controller) is located. A RRA collects information on radio resources and supervises the MAC and PHY functions including power control. A RRC collects radio resource information from RRAs and performs RRM. In Profile A, the RRC is located at the ASN gateway whereas in Profile C, it is co-located with the RRA at the BS; Profile C does support a RRC relay at the ASN gateways to relay RRM messages that allows for inter-profile RRM signaling between ASNs that are of type Profile A, B, or C.

In the next section, we will use the WiMAX ASN architecture for Profiles A and C to motivate our two-level RRM framework for a hierarchical architecture and a flat architecture, respectively.

In this section, we introduce metrics for classifying a user as a cell-edge user to be assigned to an upper RRC and also describe the initial classification of a new user.

We now derive the condition for reclassifying users and switching them from upper RRC to lower RRC or vice versa. Users that do not satisfy these conditions will, by default, not be reclassified. The objective behind reclassifying users is to maximize the sum throughput over all the users in the network covered by an ASN gateway (or a set of BSs deployed for cooperation) subject to a minimum throughput guarantee for each user. In particular, each user is allocated to either a lower RRC or an upper RRC to meet the following objective:

where is the set of BSs within the domain. Further, this reclassification is also subject to the condition that the switching (reclassified) user's throughput must not decrease.

We are now ready to propose our reclassification strategy in which a user is allowed to switch only if both its own throughput as well as the system throughput do not decrease and at-least one of them strictly increases.

The overhead of RRC switch is the exchange of signaling messages for switch request and response between two RRCs. If the algorithms are triggered more frequently, the classification will probably be more accurate, but the overhead will be higher. The overhead is related to how frequently , , , and in (14) change.

One factor that affects a user's capacity, that is, and as well as and , is user mobility, because the capacity of fast-moving users may vary in a small-time scale. If they are able to compute and via a long-term average, they will not suffer from frequent RRC switch. An alternative is to make fast-moving users always be managed by the upper RRC regardless of whether they are classified as cell-interior or cell-edge users.

Meanwhile, a factor that affects or is addition or deletion of a user in the cell, which may trigger switching of other users. This is explained by simultaneous switching. In case more than two users trigger switching simultaneously, the conditions of a permissible switching is also derived. We refer to the set of all users for which reclassification is permissible as the permissible set. We can adopt a sequential approach where the user from the permissible set which offers the highest throughput gain upon switching is selected and reclassified. The permissible set is then recomputed before switching the next selected user. In each step, the procedure is the same as the reclassification algorithm stated earlier. This approach will converge to a state for which the permissible set is empty, because at each step, the system throughput strictly increases upon switching.

We evaluated the performance in an OFDMA-based wireless network by simulation experiments, emulating mobile WiMAX systems with parameters listed in Table 1. We consider a single omnidirectional antenna at each transmitter and each receiver. In our simulator, users are uniformly distributed in a hexagonal cell and BSs of 6 first-tier and 12 second-tier neighboring cells generate intercell interference to those users. Our channel model follows path loss with an exponent of 4, Gaussian shadowing with zero mean and variance of 8 dB, and Rayleigh fading. We use the Jakes' model [20] to generate frequency-selective Rayleigh fading followed by the Doppler effect with the maximum velocity of 3 Km/hr. To serve cell-interior users, BSs either adopt a round-robin (RR) scheduling algorithm or a multichannel proportional fair (PF) scheduling algorithm [21] that guarantees minimum throughput (150 Kbps for all users in our setting) [10]. It is assumed that the channel coefficients are perfectly known at the BS and the data rate is then determined by the Shannon capacity. In our simulation, each user measures the two strongest 's from their neighboring BSs, and the serving BS is able to coordinate with one or two of those neighboring BSs. The cell performance was computed during the simulation time of 60 seconds, after each user's RRC had been completely determined according to our algorithm.

Our simulation results show that users are appropriately classified into cell-edge and cell-interior types by our algorithms. We confirmed that (i) the first cases in Propositions 1 and 2 are generally observed, (ii) FFR and macro-diversity (MD) increase cell-edge throughput by up to 15% when without a loss in system throughput, and (iii) more users switch to the cell-edge type when the neighboring cell is lightly loaded.

Figure 6 plots the ratio of edge users in the cell. The cell-edge users are now divided into the users coordinated by two BSs (FFR-2 and MD-2) and users coordinated by three BSs (FFR-3 and MD-3). Here, PF scheduling for cell-interior users is only plotted, because RR scheduling has the same tendency. In any cases, users can take advantage of FFR-2 or MD-2 more than FFR-3 or MD-3. Interestingly, the ratio of users supported by MD-3 is very small unlike MD-2, which means that macro-diversity by three BSs is not so beneficial in enhancing the throughput of cell-edge users.

Figure 7 shows the distribution of cell-edge users' and in the case of Figure 5, when macro-diversity by at most two BSs is employed and . The black area represents and of those users by simulation results and two lines represent the threshold given by (14). In this experiment, the other cases except the first one in Propositions 1 and 2 are rarely observed; for instance, the ratio of such cases is only 0.5% among all users at and it approaches zero as increases above 0.2. Therefore, as expected, the first cases can be regarded as the simplified solution in general. Furthermore, Figure 8 shows the possible location of cell-edge users in a hexagonal cell, when a user type is classified according to (26).

The average cell-edge throughput and system throughput (i.e., cell throughput in this simulation) are presented in Figure 9 when . Both "PF + FFR" and "PF + MD" represent the cases where cell-interior users are supported by the PF scheduling and cell-edge users are supported by FFR or MD. The proposed algorithm shows better cell-edge throughput, compared to the relaxed switching ("Relaxed") mentioned in Remark 2. Also, our algorithm is compared to a simple mechanism (represented by "Fixed") where RRC switch is determined by a fixed threshold, (3 dB or 7 dB). Here, is given by the neighboring BS that interferes most dominantly.

In the case of Figure 9, we obtained throughput improvement of cell-edge users, as shown in Figure 10. Compared to the case of no upper RRC, the proposed one improves cell-edge throughput by 13.0% and 14.3% for FFR and MD, respectively, without a loss in system throughput, while the relaxed case improves it only by 4.2%. But the "Fixed" algorithm degrades those users' throughput. In summary, the proposed algorithm achieves the best cell-edge throughput without losing system throughput. We omit "PF + FFR" for the fixed and relaxed switching because it results in a slightly inferior cell-edge performance to "PF + MD".

Figure 11 shows the throughput comparison between FFR and MD as a function of , when is averaged over six neighboring cells. Throughput improvement decreases as increases, because less users are allowed to switch to the upper RRC. When PF is used for cell-interior users, there is little difference between FFR and MD. In contrast, when RR scheduling is employed for cell-interior users (see "RR + FFR" and "RR + MD"), it is shown that MD is better than FFR in improving the overall throughput.

The effect of is demonstrated in Figure 12 that plots as a function of . Here, also includes the fraction of resource allocated to cell-edge users who are located in six neighboring cells. As discussed in Proposition 3, users are less likely to switch to the upper RRC as increases. To obtain this result, we imposed no upper limit on (i.e., ). When RR scheduling is employed for cell-interior users, they do not take advantage of opportunistic scheduling, and thus it drives more users to switch to the upper RRC. Therefore, in case of RR scheduling is much greater than that of PF scheduling.

We have proposed a new RRM framework for wide-area wireless data networks that manages radio resources of cell-interior and cell-edge users separately. We believe that our framework can be employed with many recent approaches that require network coordination to improve cell-edge throughput, including fractional frequency reuse, macro-diversity, and various other forms of network MIMO techniques applicable to cell-edge users, although we focused on fractional frequency reuse and macro-diversity in this work. The work presented in this paper has been limited to downlink data transmission; RRM schemes for uplink in conjunction with downlink would be one avenue for future work.