Traditionally, when classifying services attending to their delivery requirements, they have been primarily divided between Real and Non Real Time services. Usually, Real Time (RT) services have been considered to impose a strict delay requirement on the end-to-end communication. With this delay the network nodes involved in the Real Time connection are to transfer the packets of the RT flow within a maximum tolerable delay. Due to these severe delay constraints, the error correction possibilities of this type of communication are very limited, and, while link level retransmissions have shown difficultly in being feasible for Real Time services, the end-to-end recovery mechanisms have been completely dismissed. This implies that the Real Time flow delivered at the terminating entity is subject to transmission errors, thus, the final application must be able to cope with the errors.
On the other hand, Non Real Time traffic has been commonly considered as error sensitive, though with less demanding delay constraints than RT traffic. These characteristics of NRT traffic allow for link and end-to-end level recovery mechanisms that enable error free delivery of the payload. Another typical difference between RT and NRT services is that the receiving end of RT applications usually consumes the information before the end of the data transaction.
The 3GPP QoS specification [14] also follows the RT/NRT classification by identifying the conversational and streaming traffic with Real Time services, whereas the interactive
However, as soon as certain delay delivery requirements are imposed on the background and mainly interactive users, their Real Time services becomes diffused [15].
Network gaming is usually considered as a good example of this situation [3]. While this service responds to a typical interactive pattern, it may have stringent delay requirements that could only be satisfied by a UMTS bearer of the conversational class.
Chapter 3
High Speed Downlink Packet Access 3.1 HSDPA Overview
As commented in Chapter 1, the evolution of the mobile communication market brings demands for both larger system capacity and higher data rates. To boost the support for the packet switched services, the 3GPP has standardized in the Release 5 a new technology denominated High Speed Downlink Packet Access (HSDPA) that represents an evolution of the WCDMA radio interface. HSDPA appears as an umbrella of features whose combination improves the network capacity, and increases the peak data rates up to a theoretical 10 Mbps for downlink packet traffic. These technological enhancements can allow operators to enable new high data rate services, improve the QoS of existing services, and achieve a lower cost per delivered data bit.
The HSDPA concept relies on a new transport channel, the High Speed Downlink Shared Channel (HS-DSCH), where a large amount of power and code resources are assigned to a single user at a certain TTI in a time and/or code multiplex manner. The time-shared nature of the HS-DSCH provides significant trunking benefits over DCH for surges of high data rate traffic [16]. In addition, HSDPA uses Adaptive Modulation and Coding, fast Physical Layer Hybrid ARQ, and fast Packet Scheduling. These features are tightly coupled and permit a per- TTI adaptation of the transmission parameters to the instantaneous variations of the radio channel quality.
The objective of the present chapter is not to give a complete description of the HSDPA concept and its performance. Rather, it will provide a general overview of HSDPA that will be required to achieve a full comprehension for the analysis/design of the Packet Scheduler functionality.
The Packet Scheduler is a key element of the overall HSDPA concept and the central scope of the following chapters and therefore, will be treated in detail there. This way, this chapter will not cover specific HSDPA aspects that are not the subject of analysis in the rest of this thesis like for example the UE capabilities. In the rest of the thesis it is assumed that the penetration of HSDPA capable terminals in the network is 100%. Other overviews of the HSDPA concept are given in [17], [18], [19].
A general description of the specific features included and excluded in HSDPA compared to basic WCDMA technology, an overview of the major architectural modifications introduced by HSDPA will be presented in the following section.
3.2 HSDPA Architecture
Figure 3.1 plots the fundamental features to be included and excluded in the HS-DSCH of HSDPA. For this new transport channel, two of the most main features of the
WCDMA technology such as closed loop power control and variable spreading factor have been deactivated.
In WCDMA, fast power control stabilizes the received signal quality (EsNo) by increasing the transmission power during the fades of the received signal level. This causes peaks in the transmission power and a subsequent power rise, which reduces the total network capacity. However, delay tolerant traffic may be served only under
favorable radio channel conditions, avoiding the transmission during the inefficient signal fading periods. Moreover, the operation of power control imposes the need of certain
Figure 3.1: Fundamental Features To Be Included And Excluded in the HS-DSCH of HSDPA.
headroom in the total Node B transmission power to accommodate its variations. The elimination of power control avoids the aforementioned power rise as well as the cell transmission power headroom. But due to the exclusion of power control, HSDPA requires other link adaptation mechanisms to adapt the transmitted signal parameters to the continuously varying channel conditions.
One of these techniques is denominated Adaptive Modulation and Coding (AMC).
With it, the modulation and the coding rate are adapted to the instantaneous channel quality instead of adjusting the power. The transmission of multiple Walsh codes is also used in the link adaptation process. Since the combination of these two mechanisms already plays the link adaptation role in HSDPA, the variable spreading factor is deactivated as its long-term adjustment to the average propagation conditions is not required anymore.
As a closed power control is not present, the channel quality variations must be minimized across the TTI, which it is accomplished by reducing its duration from the minimum 10 ms in WCDMA down to 2 ms. The fast Hybrid ARQ technique is added, which rapidly retransmits the missing transport blocks and combines the soft information from the original transmission with any subsequent retransmission before the decoding process. The network may include additional redundant information that is incrementally transmitted in subsequent retransmissions (i.e. Incremental Redundancy).
To obtain recent channel quality information that permits the link adaptation and the Packet Scheduling entities to track the user’s instantaneous radio conditions, the MAC functionality in charge of the HS-DSCH channel is moved from the RNC to the Node B. The fast channel quality information allows the Packet Scheduler to serve the user only when his conditions are favourable. This fast Packet Scheduling and the time-shared nature of the HS-DSCH enable a form of multiuser selection diversity with
important benefits for the cell throughput. The move of the scheduler to the Node B is a major architecture modification compared to the Release 99 architecture.
Unlike all the transport channels belonging to the Release 99 architecture, which are terminated at the RNC, the HS-DSCH is directly terminated at the Node B. With the purpose of controlling this channel, the MAC layer controlling the resources of this channel (so called MAC-hs) is directly located in the Node B (see Figure 3.2, Figure 3.3), thereby allowing the acquisition of recent channel quality reports that enable the tracking of the instantaneous signal quality for low speed mobiles. This location of the MAC-hs in the Node B also enables to execution of the HARQ protocol from the physical layer, which permits faster retransmissions.
Figure 3.2: Radio Interface Protocol Architecture of the HS-DSCH Transport Channel [20].
Figure 3.3: Node B Packet Scheduler Operation Procedure.
More specifically, the MAC-hs layer [21] is in charge of handling the HARQ functionality of every HSDPA user, distributing the HS-DSCH resources between all the MAC-d flows according to their priority (i.e. Packet Scheduling), and selecting the appropriate transport format for every TTI (i.e. link adaptation). The radio interface layers above the MAC are not modified from the Release 99 architecture because HSDPA is intended for transport of logical channels. Nonetheless, the RLC can only operate in either acknowledged or unacknowledged mode, but not in transparent mode due to ciphering [17]. This is because in transparent mode the ciphering is done in the MAC-d7, not in the RLC layer, and MAC-c/sh and MAC-hs do not support ciphering [21].
The MAC-hs also stores the user data to be transmitted across the air interface, which imposes some constraints on the minimum buffering capabilities of the Node B.
The move of the data queues to the Node B creates the need of a flow control mechanism (HS-DSCH Frame Protocol) that aims at keeping the buffers full. The HS-DSCH FP
handles the data transport from the serving RNC to the controlling RNC (if the Iur interface is involved) and between the controlling RNC and the Node B. The design of such flow control is a non-trivial task, because this functionality in cooperation with the Packet Scheduler is to ultimately regulate the user’s perceived service, which must fulfill the QoS attributes according to the user’s subscription (e.g. the guaranteed bit rate or the transfer delay for streaming bearers or the traffic handling priority and the allocation/
retention priority for interactive users).
Chapter 4
Models of Scheduling Algorithms 4.1 Scheduling Overview
Wireless channels are characterized by their fast and random variations in time, making it a challenge to ensure fairness among users with different channel conditions and quality of service (QoS) requirements. Fast packet scheduling is the main component of High Speed Downlink Packet Access (HSDPA) [23] which aims at tracking the variations of the channels. In each Transmission Time Interval (TTI), the function of the scheduler is to select the user or users for which packets will be transmitted and determine their rates from a finite set of values depending on the Modulation and channel Coding Schemes (MCS), for the purpose of increasing the system’s performance both in terms of throughput and fairness. One of the known algorithms that attempt to achieve a reasonable throughput- fairness tradeoff is the Proportional Fairness (PF) algorithm [24], which is implemented in HDR (High Data Rate) networks [25]. This is introduced to compromise between a fair data rate, and the total data rate for each user. The PF algorithm was shown to achieve better performance than the maximum carrier-to-interference power ratio (CIR) method in the presence of a high number of users [26].
However, with PF scheduling, the user whose channel exhibits the highest variance gets privilege over the other users, which yields unfairness in the service of communicating users, especially in the case of HSDPA where it has a finite set of discrete rate values.
Table II Summaries the common Packet Scheduling Methods.
One of the proposed solutions for the above-mentioned problem is the modified PF algorithm which is also called the Data Rate Control (DRC) Exponent rule [27]. However, even with this algorithm, fairness is not guaranteed at all times but rather in specific cases only. In this thesis, a new scheduling algorithm is proposed, called Adaptive Proportional Fairness (APF) that resolves the PF unfairness problem in a more general way. In this algorithm, an updating module is introduced to track the data rate allocated to each user and update the exponent parameters in order to achieve fairness among the different users based on their required QoS. Unlike the DRC Exponent rule [27], herein for each user, an exponent parameter is used, which is possible of changing the proportional data rate for each user without changing those of other users.
Table II. Summary of the common used Packet Scheduling Methods [1].
4.2 Proportional Fairness Scheduling Method
The user selection criterion according to the Proportional Fairness (PF) method is given by [24]
where j is the index of the selected user for the next TTI, N is the total number of users, ri
is the instantaneous data rate the UE i can support under its current channel conditions, and R i is the average achieved rate defined as the average data rate effectively received by user i. These rates are updated, at each TTI, according to the following rule:
The PF method gives fair allocation of data rates between users when they experience the same channel conditions. In real systems, due to the different fading properties the users experience, channels are heterogeneous. In this case, the PF algorithm fails in allocating to users fair data rate that is proportional to their mean rate.
Indeed, it was shown in [27], [29] that the UE with more channel variability gets a higher data rate. In order to provide fairness between users depending only on their average data channel rate (r), a solution to the problem was proposed in [28] where by adding an
exponent term to ri, which indicates the channel condition in the PF policy (1), the so-called DRC Exponent rule allows the allocated time control to the users with better channel quality compared to the ones with bad conditions. In the DRC Exponent method, the selection criteria is hence given by
(3)
where n is the exponent parameter, introduced to manage the relationship between the average data rates (Ri) of the users with different channel conditions. However, the control parameter n takes a fixed value for all users, two problems arise with this approach. First, the control parameter being fixed in time, it does not adapt to the time-varying radio conditions of each UE. Second, as this parameter takes a unique value for all users, it is not possible to fix a value for n that ensures fairness between all users at the same time.
Next these characteristics are explained in more detail and why they are important to the design of wireless scheduling policies. In wireless networks, the channel conditions of mobile users are time-varying. Radio propagation can be roughly characterized by three nearly independent phenomena: path-loss variation, slow log-normal shadowing, and fast multipath-fading. Path losses vary with the movement of mobile stations. Slow log-normal shadowing and fast multipath-fading are varying with different time-scales.
Furthermore, a user receives interference from other transmissions, which is time-varying; and background noise is also constantly varying. Hence, mobile users perceive time-varying channel conditions. SINR (signal to interference plus noise ratio) is a commonly used measure of channel conditions. Fig. 4.1 shows the time varying SINR of a mobile user. Other measures include BER (Bit Error Rate) and FER (Frame Error Rate).
As channel conditions are time-varying, users experience time-varying service quality and/or quantity. For voice users, better channel conditions may result in better voice quality. For packet data service, better channel conditions (or higher SINR) can be used to provide higher data rates using adaptation techniques.
Fig.4.1. User's time-varying SINR
Research shows that cellular spectral efficiency (in terms of b/s/Hz/sector) can be increased by a factor of two or more if users with better links are served at higher data rates [32]. Procedures to exploit this are already in place for the entire major cellular
standards: adaptive modulation and coding schemes are implemented in the TDMA standards, and variable spreading and coding are implemented in the CDMA standards.
In general, a user is served with better quality and/or a higher data rate when the channel condition is better. Hence, good scheduling schemes should be able to exploit the variability of channel conditions to achieve higher utilization of wireless resources.
The performance (e.g., throughput) of a user depends on the channel condition it experiences; hence, different performance is expected when the same resource (e.g., radio frequency) is assigned to different users. For example, consider a cell with two users.
Suppose that user 1 has a good channel, e.g., it is close to the base station. User 2 is at the edge of the cell, where the path-loss is significant and the user experiences large interference from adjacent cells. If the same amount of resource (power, time-slots, etc.) is assigned, it is likely that the throughput of user 1 will be much larger than that of user 2.
Different assignments of the wireless resource will affect the system performance, hence, resource allocation and scheduling policies are critical in wireless networks. In this Thesis, A study Adaptive Proportional Fairness scheduling is presented. Earlier, the unique features of wireless networks are described. Then an important question is: under such conditions, what should be the basic features of a scheduling policy? Consider a few users that share the same resource. The users have a constantly varying channel condition, which implies constant varying performance. The scheduling policy decides which user should transmit during a given time interval. Intuitively, assign resource is wanted to users experiencing "good" channel conditions so that the resources can be used efficiently.
At the same time, also provide some form of fairness or QoS guarantees to all users are needed. For example, allowing only users close to the base station to transmit with high transmission power may result in very high system throughput, but may starve other users.
This basic dilemma motivates this work:
To improve wireless resource efficiency by exploiting time-varying channel conditions while at the same time controlling the level of fairness/QoS among users. Fairness criteria may have different implications in wireless networks.
In wireline networks, when a certain amount of resource is assigned to a user, it is equivalent to granting the user a certain amount of throughput/performance value.
However, the situation is different in wireless networks, where the amount of resource and the performance values are not directly related (though correlated).
Hence, A study for two kinds of fairness is depicted: temporal vs. utilitarian.
Temporal fairness means that each user gets a fair share of network resource, and utilitarian fairness means that each user gets a certain share of the overall system capacity.
Further, it is considered both long-term fairness and short-term fairness in this thesis. The basic idea of previous SIR scheduling is to let users transmit in "good" channel conditions, and thus a natural question is how long a user is willing to wait for these
"good" conditions. Hence, there exists tradeoff between scheduling performance gain and short-term performance. In addition to fairness, consider a long-term QoS metric: each user has a specific data-rate requirement for the system. Because the capacity of a wireless system is not fixed, it is not always an easy task to determine the feasibility of
the requirements of all users. deep study for these issues in more detail is analyzed in this thesis.
Interference management is also a crucial component of efficient spectrum utilization in wireless systems because interference ultimately limits the system capacity.
Power allocation is a traditional interference management mechanism. It has been well studied and widely used in wireless systems to maintain desired link quality, minimize power consumption, and alleviate interference to others [30]. Further, because users experience time-varying and location-dependent channel conditions in wireless environments, users can be scheduled so that a user can exploit more of its good channel conditions and avoid (as far as possible) bad times, at least for applications (e.g., data service) that are not time-critical. Hence, a joint scheduling and power-allocation scheme
Power allocation is a traditional interference management mechanism. It has been well studied and widely used in wireless systems to maintain desired link quality, minimize power consumption, and alleviate interference to others [30]. Further, because users experience time-varying and location-dependent channel conditions in wireless environments, users can be scheduled so that a user can exploit more of its good channel conditions and avoid (as far as possible) bad times, at least for applications (e.g., data service) that are not time-critical. Hence, a joint scheduling and power-allocation scheme