Part II Physical Layer for Downlink 121
12.2 General Considerations for Resource Allocation Strategies
The generic function of a resource scheduler, as shown for the downlink case in Figure 12.2, is to schedule data to a set of UEs on a shared set of physical resources.
It is worth noting that the algorithm used by the resource scheduler is also tightly coupled with the Adaptive Modulation and Coding (AMC) scheme and the retransmission protocol (Hybrid Automatic Repeat reQuest, HARQ, see Sections 4.4 and 10.3.2.5). This is due firstly to the fact that, in addition to dynamic physical resource allocation, the channel measurements are also used to adapt the Modulation and Coding Scheme (MCS) (i.e.
transmission spectral-efficiency) as explained in Section 10.2. Secondly, the queue dynamics, which impact the throughput and delay characteristics of the link seen by the application, depend heavily on the HARQ protocol and transport block sizes. Moreover, the combination of channel coding and retransmissions provided by HARQ enables the spectral efficiency of an individual transmission in one subframe to be traded offagainst the number of subframes in which retransmissions take place. Well-designed practical scheduling algorithms will necessarily consider all these aspects.
In general, scheduling algorithms can make use of two types of measurement information to inform the scheduling decisions, namely Channel State Information (CSI) and traffic measurements (volume and priority). These are obtained either by direct measurements at
the eNodeB, or via feedback signalling channels, or a combination of both. The amount of feedback used is an important consideration, since the availability of accurate CSI and traffic information helps to maximize the data rate in one direction at the expense of more overhead in the other. This fundamental trade-off, which is common to all feedback-based resource scheduling schemes, is particularly important in Frequency Division Duplex (FDD) operation where uplink-downlink reciprocity of the radio channels cannot be assumed (see Chapter 20).
For Time Division Duplex (TDD) systems, coherence between the uplink and downlink channels may be used to assist the scheduling algorithm, as discussed in Section 23.5.
Figure 12.2: Generic view of a wideband resource scheduler.
Based on the available measurement information, the eNodeB resource scheduler must manage the differing requirements of all the UEs in the cells under its control to ensure that sufficient radio transmission resources are allocated to each UE within acceptable latencies to meet their QoS requirements in a spectrally efficient way. The details of this process are not standardized as it is largely internal to the eNodeB, allowing for vendor-specific algorithms to be developed which can be optimized for specific scenarios in conjunction with network operators. However, the key inputs available to the resource scheduling process are common.
In general, two extremes of scheduling algorithm may be identified: opportunistic schedulingandfair scheduling. The former is typically designed to maximize the sum of the transmitted data rates to all users by exploiting the fact that different users experience different channel gains and hence will experience good channel conditions at different times and frequencies. A fundamental characteristic of mobile radio channels is the fading effects
arising from the mobility of the UEs in a multipath propagation environment, and from variations in the surrounding environment itself (see Chapter 20). In [1–3] it is shown that, for a multi-user system, significantly more information can be transmitted across a fading channel than a non-fading channel for the same average signal power at the receiver. This principle is known as multi-user diversity. With proper dynamic scheduling, allocating the channel at each given time instant only to the user with the best channel condition in a particular part of the spectrum can yield a considerable increase in the total throughput as the number of active users becomes large.
The main issue arising from opportunistic resource allocation schemes is the difficulty of ensuring fairness and the required QoS. Users’ data cannot always wait until the channel conditions are sufficiently favourable for transmission, especially in slowly varying channels.
Furthermore, as explained in Chapter 1, it is important that network operators can provide reliable wide area coverage, including to stationary users near the cell edge – not just to the users which happen to experience good channel conditions by virtue of their proximity to the eNodeB.
The second extreme of scheduling algorithm, fair scheduling, therefore pays more attention to latency and achieving a minimum data rate for each user than to the total data rate achieved. This is particularly important for real-time applications such as Voice- over-IP (VoIP) or video-conferencing, where a certain minimum rate must be guaranteed independently of the channel state.
In practice, most scheduling algorithms fall between the two extremes outlined above, including elements of both to deliver the required mix of QoS. A variety of metrics can be used to quantify the degree of fairness provided by a scheduling algorithm, the general objective being to avoid heavily penalizing the cell-edge users in an attempt to give high throughputs to the users with good channel conditions. One example is based on the Cumulative Distribution Function (CDF) of the throughput of all users, whereby a system may be considered sufficiently fair if the CDF of the throughput lies to the right-hand side of a particular line, such as that shown in Figure 12.3. Another example is theJain index[4], which gives an indication of the variation in throughput between users. It is calculated as:
J= T2
T2+var(T), 0≤J≤1, (12.1)
whereT and var(T) are the mean and variance respectively of the average user throughputs.
The more similar the average user throughputs (var(T) → 0), the higher the value ofJ.
Other factors also need to be taken into account, especially the fact that, in a coordinated deployment, individual cells cannot be considered in isolation – nor even the individual set of cells controlled by a single eNodeB. The eNodeBs should take into account the interference generated by co-channel cells, which can be a severe limiting factor, especially for cell-edge users. Similarly, the performance of the system as a whole can be enhanced if each eNodeB also takes into account the impact of the transmissions of its own cells on the neighbouring cells. These inter-cell aspects, and the corresponding inter-eNodeB signalling mechanisms provided in LTE, are discussed in detail in Section 12.5.
criterion Fairness
CDF of perưuser throughput
Perưuser throughput relative to peak 0
0.5 1
0% 10% 50% 100%
Figure 12.3: An example of a metric based on the throughput CDF over all users for scheduler fairness evaluation (10–50 metric).