Enhanced Dedicated Channel Scheduling Optimization in WCDMA *
Dimitrios Skoutas, **Dimitrios Komnakos, *Demosthenes Vouyioukas and *Angelos Rouskas
*Department of Information and Communication Systems Engineering University of the Aegean Samos, GR-83200, GREECE e-mail:{d.skoutas, dvouyiou, arouskas}@aegean.gr
Abstract— Enhanced Uplink operation of WCDMA-UMTS systems utilizes fast Node B scheduling and a short Transmission Time Interval (TTI) of 2ms. The scheduling decisions are based on scheduling information (SI) sent by each UE to the Node B. The main drawback of the current scheduling procedure is that the SI does not include any information regarding the delay sensitivity of each traffic flow, as well as the actual delay that the buffered data have experienced. In this paper, we propose the inclusion of such information in SI by introducing a modification of the respective 3GPP specifications. Furthermore, we propose a Dynamic Priority Scheduler (DPS), which can utilize this new information. The performance of the proposed scheme is evaluated by means of system level simulations. Keywords-Enhanced Uplink; Packet Scheduling; QoS; WCDMA
I. INTRODUCTION The evolving technology for 3G is called High Speed Packet Access (HSPA) which is actually a packet based cellular system deployed on top of the current WCDMA networks. HSPA is an evolution of WCDMA-UMTS technology, achieving greater bit rates and reduced delays. Responsible for the standardization of HSPA is the 3GPP organization. The commercial utilization of this technology is rather new, since new HSDPA (downlink) networks are launched by European providers continuously, while the first Enhanced Uplink (HSUPA) networks were implemented only during the last year. Enhanced Uplink incorporates a significant number of innovative features, such as Adaptive Modulation and Coding, short Transmission Time Interval (2ms), fast Hybrid Automatic Repeat Request (HARQ), customized schedulers for the proper manipulation and routing of the data, as well as the possibility for a Multiple Input Multiple Output (MIMO) add-on. The 3GPP release 7 [1], also known as HSPA+, expects to increase the downlink peak bit rate to 28.8 Mbps and the uplink peak bit rate to 11.5 Mbps, being the outrider for 3GPP long-term evolution (LTE). This ability of HSPA to serve simultaneously a sufficient number of mobile terminals running multiple demanding applications while keeping the end to end delay low, makes it an ideal candidate for a vast number of new services.
**School of Electrical and Computer Engineering, National Technical University of Athens Zographou, GR-15780, GREECE e-mail:
[email protected]
Focusing in HSUPA, this increasing demand for high-speed services requires fast and reliable scheduling algorithms along with robust QoS control mechanisms. However according to 3GPP, the SI transmitted by the UE to the Node B includes information regarding only the amount of the buffered data and the required transmission power. As a result, the scheduling framework, proposed by 3GPP, does not provide the required information to the scheduler to dynamically adapt to the delay requirements of the various services. In this paper, we propose the inclusion of such information in SI by introducing a modification of the respective 3GPP specifications. Additionally, a new scheduling scheme is proposed, namely Dynamic Priority Scheduler (DPS), which takes into account the delay sensitivity of traffic flows as well as the actual delay of buffered data of each traffic flow. The evaluation of the proposed scheduling scheme is performed throughout uplink scheduling scenarios and for different types of services, showing significant improvements in average and maximum packet delay and latency. The rest of the paper is organized as follows; Section II illustrates the scheduling framework as it is currently defined by 3GPP. At Section III the scheduling problem under study is presented together with an overview of previously published related works. In Section IV, we describe the proposed modifications of the scheduling information sent by the UE to the Node B. Based on these modifications Section V addresses the Node-B scheduling problem in a step-by-step procedure. The proposed scheduler is evaluated through simulations described in Section VI where results are also presented. Finally, Section VII concludes our work. II.
SYSTEM MODEL
Although 3GPP uses the term Enhanced Dedicated Channel (E-DCH) the term High Speed Uplink Packet Access (HSUPA) is widely used in the wireless industry in order to designate EDCH as a counterpart of the High Speed Downlink Packet Access (HSDPA). Both HSDPA and E-DCH utilize fast Node B scheduling which minimizes the latency of the scheduling decisions and a short TTI of 2ms. On the other hand, E-DCH is a dedicated channel which does not support adaptive
modulation while it supports variable spreading factor as well as fast HARQ and fast power control [4], [5], [6]. A number of new channels are required for the E-DCH operation. The Enhanced Dedicated Physical Data Channel (EDPDCH) is used to carry the data while the required signaling is carried on the Enhanced Dedicated Physical Control Channel (E-DPCCH). Only one E-DCH is allowed per UE however multiple services can be multiplexed into a single E-DCH. Scheduling in Enhanced Uplink is based on grant requests and allocations. The grant allocations are performed by absolute grants and relative grants. Accordingly, the scheduler can control the transmission power and rate of each UE. As the number of UEs in a system increases, the amount of allocated resource per UE decreases and the amount of interference increases causing degradation of the system throughput. A. The Node B scheduler The Node B scheduling procedure is illustrated in Figure 1. The resource that the Node B scheduler has to share is the uplink noise rise. The Node B is able to measure the level of interference in the cell which is caused by the users in the cell plus the interference from the neighboring cells and the thermal noise. The total resulting interference should not exceed a target noise rise threshold beyond which overload occurs. Therefore, the scheduler should determine the UE's power allocation (Serving Grant) for the uplink and consequently the uplink transmission rate in such a way that the total noise rise is below the required threshold.
The Node B scheduler may use SI reports in addition to the happy bit. The SI reports are sent by the UE to the Node B, either periodically or triggered by data arriving at an empty buffer provided that the UE had no permission to transmit during the previous frame [4], [5], [6]. Each SI report consists of the following data: a)
The Highest priority Logical channel ID (HLID) with buffered data on 4 bits. If there are more than one logical channels of high priority then the channel with the more buffered data is reported.
b) The Total E-DCH Buffer Status (TEBS) on 5 bits. c)
The Highest priority Logical channel Buffer Status (HLBS) on 4 bits denoted as a fraction of TEBS.
d) The UE Power Headroom (UPH) denoting the ability of the UE to increase its own transmission power. The length of UPH is 5 bits. e)
Serving and Neighbor Cell Path Loss (SNPL): This information may be used by the Node-B in order to assist the intercell interference estimation. The length of SNPL is 5 bits.
2) Serving Grant Based on the SI, the scheduler prioritizes connections determining which traffic flows have higher priority. Subsequently, based on the interference measurements, the scheduler allocates transmission power (Serving Grant-SG) to each UE. Then, the UE according to the allocated Serving Grant selects the proper Transport Format Combination (TFC) which corresponds to the maximum possible bit rate. III.
Figure 1. Node B scheduling procedure
Additionally, the scheduler should also take into account the Quality of Service (QoS) requirements of the uplink traffic flows. In order to preserve the QoS the scheduler requires feedback information from the UEs. 1) Scheduling Information For every E-DCH transmission, the UE transmits the “happy bit” in order to indicate whether it requires more resources or not. A UE declares to be unhappy when the three following conditions are true [5]: a)
The UE is able to transmit as much data as allowed by the current Serving Grant; and b) it has sufficient available power to transmit at higher data rate; and c) Total buffer status would require more than Happy_Bit_Delay_Condition ms to be transmitted with the current Serving_Grant × the ratio of active processes to the total number of processes.
RELATED WORK AND PROBLEM STATEMENT
There are two types of CDMA uplink scheduling algorithms: code-domain scheduling and time-domain scheduling [4]. The code-domain scheduling allocates a limited radio resource to many UEs continuously, inducing guaranteed QoS in terms of delay for real-time traffic. The time-domain scheduling can utilize multi-user diversity and improve system throughput by an opportunistic transmission considering channel status. A combined approach of time and code division scheduling algorithms for WCDMA uplink is presented in [3]. Three scheduling algorithms namely Blind Fair Throughput scheduler (BFT), Maximize Transmit Power Efficiency scheduler (MTPE) and Channel-State Aware Fair Throughput scheduler (CSAFT) have been studied. While the first follows the Round Robin scheduling discipline the other two schedulers utilize an Uplink Channel Quality Indicator. An improved hybrid allocation scheme, presented in [7], aims to combine the advantages of time and code domain scheduling. The UE service priority according to this scheme is calculated based on the respective packet drop rate and an adaptive weight factor which is depended on the characteristics of the traffic (traffic class).
A. Problem Statement The transmission of HLID and HLBS informs the scheduler only about the logical channel with the highest priority among the traffic flows of a UE and the corresponding amount of buffered data. Assuming that the logical channel IDs are mapped to different service flows, the Node B scheduler is able to identify the service of each logical channel. Hence, the scheduler could serve for example VoIP connections before FTP connections.
each flow. This value is already known to the UE and thus requires no further signaling. TABLE I.
Empty buffer
THE PROPOSED MODIFICATIONS
Our goal is to overcome the scheduling constraints posed from the current 3GPP specifications and thus we propose some modifications of the SI sent by the UE to the Node B. The aim of these modifications is to include in the scheduling process a criterion which combines the delay sensitivity of each service with the respective queuing delay. First, we introduce the Logical Channel Priority (LCP) variable which is calculated at the UE for all the E-DCH traffic flows. LCP, as shown at equation (1), is defined as the ratio of the Head Of Line (HOL) packet delay of the flow to the respective packet delay threshold of the flow:
LCP =
HOL _ packet _ delay flow _ delay _ threshold
(1)
The LCP data is then used as the measure for prioritizing connections. A. HLID modification We propose the following modification regarding HLID: The UE, instead of transmitting the 4 bits of HLID, transmits the Higher LCP (HLCP) which denotes the higher priority flow among all the flows of the UE. The HLCP can also be mapped to 4 bits according to the Table I. Consequently, the scheduler will be able to directly compare the high priority flows of different UEs based on a criterion which considers the delay sensitivity of the services, as well as the buffered packets delay status. B. Happy bit modification Furthermore we propose a modification at the third condition of the “happy bit” described in Section II as follows: “The buffered data of each traffic flow would require more time than the respective delay threshold to be transmitted with the current Serving Grant and the current LCP prioritization.” As a result, instead of using the predefined and unchanging value of Happy_Bit_Delay_Condition, we introduce the use of a variable which dynamically adapts to the requirements of
HLCP Transmitted (4bits) 0000
HLCP
It is now obvious that all the scheduling schemes for EDCH are restricted by the 3GPP specifications and there is no way to become aware of the actual delay that the buffered data have experienced or the delay sensitivity of each service. Thus, an E-DCH scheduler cannot effectively provide QoS differentiation among services of the same traffic class under the current specifications. IV.
HLCP MAPPING
If If If
0,00 0,07 0,14
≤ HLCP < ≤ HLCP < ≤ HLCP
