A Radio Resource Management Concept with Local Centralization

Miguel Berg

Stefan Pettersson

Summary

We propose a radio resource management (RRM) concept based on local centralization for future wireless communication networks. The concept integrates dynamic channel allocation and quality-based power control with an allocation feasibility check. Centralized networks have a potential to achieve higher resource utilization and spectrum efficiency than networks with distributed RRM, but for large networks, the computational complexity may be overwhelming. Local centralization can reduce this complexity problem and, to a large extent, keep the benefits of centralization. Here, the network is divided into several base station clusters, each controlled by a central unit responsible for resource allocations. Resource allocation is centralized within each cluster and performed in a distributed manner between clusters. The RRM decisions are based on a link gain matrix obtained from downlink path loss measurements on all base stations within a cluster. Prior to every resource allocation, a feasibility check is performed to ensure the link quality of the new and the already admitted users. In this paper, we show that our RRM concept outperforms distributed allocation schemes in both indoor and outdoor scenarios. We also show that sector antennas can regain the capacity lost with multiple central units.

1.Introduction

In the future, a variety of different services will be provided through wireless networks. The services now available on the Internet, such as WWW-browsing, electronic mail, file transfer, audio and video broadcasts, will be offered to mobile users and not only to those connected to a fixed network. The demand for wireless access to these services comes mainly from the success of cellular telephony. The freedom that users have gained from cellular telephony is irreversible and will continue to grow.

Different wireless services require different Quality of Service (QoS), such as delay, data rate and error rate. Several constraints must be met to provide a certain service with satisfaction. The maximum delay is much more important in an interactive real-time video session than in an e-mail transfer. For one-way non-interactive video, the delay jitter is of more importance than the maximum delay. We can tolerate a constant delay in a video download but variations in the delay are problematic. In a large file transfer, we are more interested in the average data rate than the peak data rate that can be offered. There is also a difference in error sensitivity. A video session can tolerate minor errors but a file transfer cannot. These differences in QoS must be handled in multi-service communication systems.

Most of the early research efforts in wireless communication were made on individual radio links from one transmitter to one receiver. The main problem was to send signals over an unreliable radio link. The signal that arrives at the receiver is greatly distorted due to time-varying and frequency-selective fading. The radio waves are scattered from reflecting obstacles and arrive at the receiver with different delays, producing multi-path fading. These problems can largely be handled after the introduction of digital communication. Digital signal processing has shifted the focus from increasing the capacity on single radio links, limited by gaussian noise and bandwidth, to capabilities of multi-user networks. Now, the problem is more of how to combat the interference produced by other users using the same channel in a nearby area. This co-channel interference is more limiting for the capacity than thermal noise in the receivers, especially in indoor communication with a dense architecture.

To deal with multi-user and multi-service wireless communication networks, effective radio resource management is essential. Dynamic channel allocation (DCA) and power control are two very effective techniques to mitigate the capacity limiting co-channel interference. It is generally considered that DCA should be performed in a distributed manner due to complexity issues. DCA can also be performed centralized. The allocation decisions are then made centrally and the mobile terminals are prohibited from making the decisions. The system capacity can be increased with smart and centralized channel allocation schemes, although they generally require more measurements and signaling compared to distributed schemes.

In this paper, we propose a locally centralized radio resource management concept. Local centralization can increase the capacity and still keep the complexity at relatively low level. Similar work has been done in [10], [15] and [16] but they assumed that neighboring bunches or clusters used different channel sets. Another difference from our work is how the central knowledge is used in RRM decisions. Recently, Qiu et al [21] proposed a framework similar to our bunch concept with the link gain matrix as a basis for the RRM decisions. One big difference is that they use link adaptation instead of transmitter power control. Further, their approach could be seen as a “dynamic bunch” in the sense that every one of their BSs act like a CU, exchanging RRM related information with its neighbors and then performs the necessary calculations. The advantage is that inter-bunch problems are reduced since the MS requesting resources is always in the center of the bunch. On the other hand, every BS needs computational capacity.

Early work on the concept focused on the standardization of the next generation of mobile telephone system. Later work has also considered our concept suitable for WLANs like HiperLAN/2.

We have organized the paper as follows. In section 2, the RRM concept is described and the background is presented. In section 3, the different components of the RRM architecture are presented. Section 4 addresses issues of implementation. Proposals of how to implements our concept is shown and some concept weaknesses are discussed along with proposals on how to avoid or reduce them. The numerical evaluation and its assumptions are presented in section 5.

2.The locally centralized RRM concept

Our locally centralized RRM concept (a.k.a. the bunch concept) was first presented in [4] and consists of a Central Unit (CU) connected to a bunch (cluster) of Remote Antenna Units (RAUs), as shown in Figure 1. In this paper, the notation bunch is used both for the central unit with its connected RAUs and also for the area that the RAUs cover. We hope that the context makes the meaning clear.

Within a bunch, we assume that high-speed communication links are available between the RAUs and the CU. This makes communication “cheap” and enables the use of advanced RRM algorithms, utilizing all the available information in the bunch. Between bunches, communication capacity is probably more limited and expensive, which means that inter-bunch resource management must be effective without frequent information exchange. Thus, it is suitable to centralize resource management within a bunch and use decentralized RRM or interference averaging (e.g. spread spectrum) to handle interference between bunches. It would of course also be possible to assign different channel sets to neighboring bunches but this limits the number of available channels in the whole bunch, not just at the border where problems with uncontrolled interference are much more likely. A further disadvantage with different channel sets is that it requires planning in advance.

To perform efficient RRM within a bunch, the CU uses central knowledge about the system. In our implementation of the bunch concept, the main idea is to make a feasibility check in the CU prior to a resource allocation. The new allocation is feasible only if the required link quality for the new and the already allocated co-channel users can be achieved. This feasibility check is based on knowledge of the path gains (inverse path loss) between the user terminals and the RAUs, and on knowledge of the transmitter powers within the bunch. Our proposed RRM architecture also includes Signal-to-Interference-Ratio (SIR) based channel allocation and transmitter power control combined with the feasibility check. Further, the existing co-channel users are notified of any necessary transmitter power updates before the new user is admitted. Intra-bunch handovers are treated in a similar way as new allocations. A difference is that this type of handover will get a higher priority than a new allocation. Inter-bunch handover is treated on a higher level and not considered in the concept.

The idea of using path gains to calculate the feasibility of allocations was originally proposed by Nettleton [17]. The implementation of our bunch concept extends Nettleton’s work by combining the feasibility check with SIR-based power control, where initial transmitter powers are calculated by the CU during the feasibility check. The idea of protecting the existing links and users is not new; in [1]-[3], interactive methods for this purpose are proposed and in [9] a method similar to our feasibility check is proposed but they do not consider local centralization to reduce the computational complexity.

Our RRM architecture for the bunch concept was first presented at the MMT conference 1997 [4] and is based on ideas from [8], [10], [11], and [14]. The work was a joint effort within the FRAMES project, which was part of the European Community ACTS research program. The overall objective of FRAMES was to define, to develop and to evaluate a wideband and efficient multiple access scheme which fulfils the Universal Mobile Telecommunication System (UMTS) requirements. The project was very successful; two of the three radio interfaces developed in FRAMES survived with slight modifications and now form the basis of the WCDMA (UTRA FDD) and TD-CDMA (UTRA TDD) standards recommended by the Third Generation Partnership Project (3GPP) [1].

One of our partners in FRAMES introduced some of the ideas and terminology regarding the bunch concept [10]. A major difference between their and our work is the assumed central knowledge in the CU. The work in [10], [15] and [16] assume a central structure matrix instead of our link gain matrix. The structure matrix could be described as a compatibility matrix and contains information of the cell overlap in the system. The overlapping areas of the cells are called zones and the mobile location determines which zone it is in. A zone is therefore an area covered by one or several RAUs. The information in the structure matrix tells if a zone is covered, interfered or non-interfered by a specific RAU. The different levels of coverage are then used in the dynamic resource allocation. Combining such a scheme with power control complicates matters considerably but the authors presented a solution for a system with two different power levels.

3.Resource management architecture

All the intra-bunch radio resource management algorithms such as channel allocation, link adaptation, power control, and handover should be tightly integrated. The reason for this is that we strongly believe that resource management is much more efficient if the different algorithms co-operate tightly contrary to the conventional subdivision of RRM functions. The central unit is responsible for resource allocation, reallocation, and deallocation. All allocation requests are passed through a queue sorted in priority order. Reallocation is performed by deallocating the resource, generating a new resource request in the queue and finally allocating it with some constraints that depend on the reason for deallocation. The most important functions in the RRM scheme are:

  • A Priority Queue through which all requests are passed. Initial priorities are assumed to be supplied by a higher level based on traffic type. Reallocations in a bunch have a higher priority than new allocations in order to avoid dropping of existing users.
  • The Generic Allocation algorithm, which handles all resource requests including the ones for new users and for users needing more capacity. If the allocation fails, the request is put back in the queue.
  • The Generic Deallocation algorithm, which is responsible for releasing resources in case of problems (quality warnings from measurements), and when otherwise needed, e.g. upon handover or service termination and in case of too good quality. Resources that are removed in case of problems are put in the priority queue.
  • The Power Control (PC).
  • Inter-bunch Resource Management

3.1.The Priority Queue

Resource requests are placed in a priority queue. Here we assume that Admission Control supplies each user request with a priority. Requests from the already assigned users normally have a higher priority than new requests, i.e. reallocations and requests for upgrading the link quality/transmission rate are usually served before new requests but this is a system parameter that the operator can change. Resources that are put in the queue by the GD have a status field containing the reason for deallocation, e.g. bad quality. It also contains eventual constraints on the wanted allocation, e.g. a wish list on preferred channels or RAUs.

We also need a timeout-mechanism for the requests in the queue. Each request has a time limit that must not be exceeded while waiting for service. If the request cannot be allocated within this time in the queue, that request is dropped. This means that the user request with the lowest priority in the system is dropped. The priorities can be dynamic and altered with time if necessary.

3.2.Generic Allocation

Generic Allocation (GA) is responsible for allocation of resources and allocates resources according to the requests in the queue. It includes initial selection of transmission modes and uses central knowledge to allocate the resources. If the queue is not empty, the resource allocation procedure is triggered on the following events:

  • Change in the queue
  • A timer has expired since the last allocation attempt
  • De-allocation has occurred

We divide GA into four sub-algorithms: RAU Selection, which finds the most suitable RAU (e.g. with the lowest path loss); Channel Selection, which selects resources to try; Feasibility Check, which calculates if the selected resources can achieve an acceptable quality (SIR) without disturbing the existing users; Initial Power, which decides what transmitter power the new and existing connections should have when resource allocations are performed.

3.2.1.RAU Selection

The first part in the allocation process is the selection of the Remote Antenna Unit, RAU. The mobiles measure on the beacon channels prior to the first contact with the system. These values are reported and are used in the selection process. In this study, we consider only the case when the terminals are connected to the strongest RAU although other algorithms are possible. A terminal must be served by different RAUs when moving from one area to another within the bunch. An intra-bunch handover of the connection is then performed in some way. The handover decision in made centrally with the use of measured data reported by the mobile terminals. In a system with a dense indoor architecture, the intra-bunch handover rate may become a problem if not dealt with properly. However, handover is not considered in this paper and left as a topic for future studies.

3.2.2.Channel Selection

The Channel Selection determines the order in which to test the available channels in an RAU for feasibility. More or less complex algorithms can be used to find a channel. They can for instance be based on measurements, centralized calculations or just picked at random. With enough computational capacity available, it would be possible to test all channels for feasibility and select the best one according to some criterion. If no feasible channels are available at the selected RAU, the resource request is either blocked or the central unit could try to allocate the MS to another RAU.

3.2.3.Feasibility Check

The task of the Feasibility Check is to protect the existing links and users in the system from a quality drop when a new resource is allocated. This is done by calculating the effect of an allocation before it is actually performed. Data of the current system situation are continuously gathered and stored centrally. From these data, link gains are calculated from all RAUs to every terminal in the system and the result is stored in the link gain matrix (see section 3.5). This matrix together with transmitter power levels and SIR targets are used in the feasibility check calculations to ensure that the requested resource can be granted. A channel is feasible if both existing and new links can meet their specified SIR targets.

3.2.4.Initial Power

If a channel was found feasible, we determine the new transmitter powers for all co-channel users in order to ensure that their quality needs are met. The users that are affected by the allocation are notified of this and the transmitter powers can be adjusted to the calculated levels. Instead of transmitting individual power updates to all affected users, we can save a large amount of signaling by broadcasting a common “boost factor” to them [1], [3]. All co-channel powers are then raised to maintain SIR despite the increased interference and the new user gets an initial power that is sufficiently low not to interfere with the previously admitted.