QoS-Aware Load Indicators for Intelligent Cell Selection Du Ho Kang, Youngkyu Choi † , Dongmyoung Kim, and Sunghyun Choi School of Electrical Engineering and INMC Seoul National University, Seoul, Korea † System Lab., Telecommunication Network Business SAMSUNG Electronics, Suwon, Korea Email: dhkang@mwnl.snu.ac.kr, youngkyu.choi@ieee.org, dmkim@mwnl.snu.ac.kr, and schoi@snu.ac.kr Abstract—Load balancing in cellular networks is an important technique to mitigate the imbalanced usage of network-wide air-resources. Specifically, the load information notified by the network to users over the air, i.e., load indicator, helps a user intel- ligently select a base station (BS) by considering the network load along with the channel quality information obtained by its own local measurement. Our key argument is that the load indicator should be more detailed than overall radio resource utilization since the air resources in packet-based cellular systems are shared according to scheduling discipline among heterogeneous traffic flows with different quality-of-service (QoS) requirements. Accordingly, we propose QoS-aware load indicators when traffic flows with and without rate constraints are mixed. Then, we discuss two BS selection algorithms, which are devised based on the proposed QoS-aware load indicators. Simulation result shows that our proposed schemes indeed achieve more intelligent BS selections. Index Terms—load indicator; load balancing; association con- trol I. I NTRODUCTION The usage of radio resources in wireless cellular networks is often imbalanced among base stations (BSs) due to irregular mobility pattern and uneven user distribution [1]. When a user is simply attached to the BS with the strongest received signal strength, geographically imbalanced user distribution conse- quently results in an inefficient utilization of the network-wide resources. In order to deal with this issue, load balancing via association control has been studied [2–4]. While a centralized coordination in [2, 3] yields an almost optimal performance, it is practically hard to implement both due to the huge computation complexity of a controlling entity and signaling overhead needed to acquire the channel gains between each user and neighboring BSs. For these reasons, a network-assisted distributed association control is a more reasonable approach. It makes each user se- lect the BS intelligently in a distributed manner based not only on channel quality measurement but also on auxiliary informa- tion about network load, i.e., a load indicator, provided by the network. For the distributed association control, broadcasting This work was supported by the IT R&D program of MKE/KEIT (KI001810, Development of cooperative operation profiles in multicell wire- less systems). the load information to users over the air is defined in recently amended IEEE 802.16 Wireless Metropolitan Area Networks (WMAN) [5] where averaged available downlink/uplink radio resources are periodically broadcasted. However, the broadcast information infers the outwardly appeared resource usage, i.e., the portion of the resource occupied by all admitted traffic flows in an aggregated sense without differentiating the ratio of radio resource occupied to meet the QoS constraints with that assigned to best-effort traffic. Namely, while in circuit-switched networks one minus fractional ratio of resource utilization exactly represents the normalized available resource, 100% of overall resource uti- lization in packet-based networks does not necessarily imply a zero rate for incoming flows since some portion of resources occupied by existing flows can be shared with the new flows according to the scheduling policy. In fact, we observe in packet-switched networks that some applications such as voice over IP (VoIP), which is configured, for example, to use unso- licited grant service (UGS) scheduling type in IEEE 802.16, require resource reservation equivalently as in circuit-switched networks. In contrast, others classified as elastic traffic, e.g., flows served by best effort (BE) scheduling type, can still attain a meaningful utility even when the rate of existing service flows is reduced by allowing a new service flow to participate in resource sharing. In other words, the resource occupied by each admitted flow is either exclusive or bearable for shared use of resource with a new flow depending on corresponding quality-of-service (QoS) requirement. To reflect these features to load of packet-based network, differentiated resource utilization should be represented. Recently, the notion of differentiated load indicators starts to appear in IEEE 802.11 [6] standards. For instance, in IEEE 802.11k [7], available amount of medium time corresponding the different traffic types called Access Categories is delivered to stations (STA) via beacons or probe response frames. However, when the air resource is fully utilized, an incoming STA might have a difficulty in estimating its expected rate since achieved rate depends not only on the transmission rate, but also on the number of contending STAs. Another ongoing task group IEEE 802.11v [8] introduces in its draft another load metric, i.e., the number of associated STAs