Xin Sun， Yuan Yang and Zhengyu Song

（School of Electronic and Information Engineering, Beijing Jiaotong University,Beijing 100044, China）

Abstract: An improved delay priority resource scheduling algorithm with low packet loss rate for multimedia broadcast multicast service (MBMS) in long term evolution (LTE) systems is proposed in this paper. Real-time services in LTE systems require lower delay and packet loss rate.However, it is difficult to meet the QoS requirements of real-time services using the current MBMS resource scheduling algorithm. The proposed algorithm in this paper jointly considers user delay information and real-time channel conditions. By introducing the user delay information, the lower delay and fairness of users are guaranteed. Meanwhile, by considering the channel conditions of users, the packet loss rate can be effectively reduced, improving the system throughput. Simulation results show that under the premise of ensuring the delay requirements of real-time services, the proposed algorithm achieves a lower packet loss rate compared to other existing algorithms. Furthermore, it can achieve a good balance between system throughput and user fairness.

Key words: multimedia broadcast multicast service (MBMS)； quality of service； resource scheduling；real time services

Long term evolution (LTE) networks are able to support high system capacity and data rates for users. However, with the rapid development of real-time traffic such as real-time mobile video transmission, LTE unicast service has been unable to meet the quality of service (QoS) requirements[1-2]. To address this issue, the Third Generation Partnership Project (3GPP) has proposed the multimedia broadcast multicast service (MBMS) in LTE Release 6[3]. MBMS is a point-to-multipoint service in which data is transmitted from a single source entity to multiple recipients. Multiple users with the same QoS requirements share the common network resources, which substantially improves the transmission efficiency and system capacity[4-6].

Generally, the real-time services in wireless networks require lower delay and packet loss rate, as well as sufficient bandwidth to ensure the QoS requirements, where real-time MBMS resource scheduling algorithm plays an important role[7]. There have been some existing algorithms proposed in the literature. For example, the multicast proportional fair (MPF) algorithm is proposed in Ref. [8], where the tradeoff between system throughput and user fairness is realized, but the user delay is not considered. In Ref. [9], the modified largest weighted delay first (M-LWDF)algorithm is proposed. Although the M-LWDF algorithm is able to improve the delay performance, it is difficult to guarantee a scheduling opportunity for users with poor channel conditions. Meanwhile, the complexity of the MLWDF algorithm is higher. Besides, the delay priority scheduling (DPS) algorithm is proposed in Ref.[10], which has lower computational complexity and can meet the QoS requirements of users. However, as the number of users increases,the packet loss rate becomes higher and the system throughput decreases significantly.

In view of these problems, in this paper, we propose an improved delay priority scheduling(IDPS) algorithm, which jointly considers the user delay information and the real-time downlink channel conditions. Simulation results demonstrate that the proposed IDPS algorithm effectively ensures lower user delay and packet loss rate for real-time services. It can also achieve a good balance between system throughput and user fairness.

1 System Model

In this paper, a single cell of 5 MHz bandwidth made up of 25 physical resource blocks(PRBs) and a 2 GHz carrier frequency is considered. The eNodeB is located at the center of the cell, and users are randomly distributed in the cellular coverage. Multiple users with the same QoS requirement form a multicast group sharing the common resources. The system model is as shown in Fig.1.

In this system, each user experiences a different instantaneous downlink signal to interference plus noise ratio (SINR) at each transmission time interval (TTI) and on each PRB due to the frequency-selective fading nature of multipath propagation and time-selective fading nature due to the Doppler shift[11]. Thus, users are required to report their channel quality indicators (CQIs) to the eNodeB according to their SINR at each TTI. The eNodeB can adjust and update the user’s channel status information in a timely manner according to the reported CQI information, then adopt an adaptive modulation and coding (AMC) technique to select the appropriate coding modulation scheme. Hence, the achievable data rate of user n at the t-th TTI on PRB i can be given by

Fig.1 System model

2 Proposed IDPS Algorithm

In this section, we describe the proposed IDPS algorithm in detail. In order to meet the QoS requirements of real-time services, based on the DPS algorithm, the IDPS algorithm is proposed by taking into account the delay and channel conditions of users. To be specific, the proposed IDPS algorithm can be divided into the following three steps.

In Step 1, the weighting parameter for each user is calculated in each TTI, which can be expressed as

where Tnk(t) is the average throughput of user n in group k until the t-th TTI, τnis the time-lapse threshold for user n , τn(t) is the duration of time that user n has been waiting until the t-th TTI, a and b are constant weighting coefficients, Γn(t) is the average spectrum efficiency for user n at the t-th TTI on all PRBs , which is defined as

where NPRBis the number of PRBs, φin(t) is the spectrum efficiency obtained for user n on the ith PRB according to the MCS and SINR mapping table in Ref.[12]. Spectrum efficiency is determined by the CQI reported by the user. Thus,the channel condition of user n at the t-th TTI can be characterized by Γn(t).

In Step 2, the feasible data rate assigned to group k at the t-th TTI can be calculated as

where φk,t(y) is defined as

In Eq.(5), Skis the total number of users in group k, rnk(t) is the data rate control (DRC) of user n in group k. If the eNodeB sends at a rate higher than the DRC rate of a user, then the user cannot receive any data.

where G is the number of multicast groups.

In Eq.(2), a and b are constants, where a affects the weight of users’ channel conditions, and b affects the weight of users’ delay. If a is larger,the weight of users’ channel conditions is larger in Eq.(2). Then, the IDPS algorithm gives higher priority to users with good channel conditions to accept the scheduling, which will help to improve the system throughput, but the fairness between users will be reduced. By mathematical analysis and simulation verifications, from the viewpoint of balancing the system throughput and user fairness, weight values a = 5 and b = 2 are selected in this paper.

3 Simulation Results

To evaluate the performance of our proposed IDPS algorithm, in this section, we conduct simulations on the packet loss rate, system throughput, average user delay and user fairness,where MPF, M-LDWF and DPS algorithms are also taken into account for comparison. In the simulations, users are randomly distributed in the cell. All the services are real-time video streaming transmissions, and the delay threshold is set to 20ms. In order to simplify the grouping of users in a cell, it is stipulated that each sector in the cell is an independent multicast group.There are three multicast groups in a cell. The other simulation parameters are listed in Tab. 1.

Tab. 1 Simulation parameters for MBMS in LTE systems

In Fig.2, the average packet loss rate versus the number of users is demonstrated. It can be seen that as the number of users increases, the packet loss rates of the four algorithms all grow.This is because with the increase of the number of users, it is difficult to guarantee all users’ QoS requirements due to the limited PRB resources.Thus, more packages are discarded and the packet loss rate increases. Specifically, if the number of users is less than 50, the PRB resources in the system can guarantee the real-time QoS requirements of all users. Hence, the packet loss rates of the four algorithms are all close to 0. When the number of users is between 50 and 70, the packet loss rates of the MPF and M-LWDF algorithms become higher, while the DPS and IDPS algorithms can still obtain lower packet loss rates. When the number of users is more than 70, the packet loss rate of DPS algorithm increases sharply, while that of the proposed IDPS algorithm increases slowly and is still lower than the other three algorithms. This phenomenon can be well explained. Under the same simulation parameters, when the number of users increases, there are more users with the same priority. Since the IDPS algorithm gives higher priority to users with good channel conditions to accept the scheduling, it can make full use of highquality channels to send more packets with limited PRB resources.

In Fig.3, system throughout versus the number of users is illustrated. It is observed that as the number of users increases, the throughput of all four algorithms grows continuously. In more detail, when the number of users is less than 40, the system throughputs of the four algorithms are almost the same. This is because the PRBs in the system are sufficient to support all users’ requirements. However, as the number of users increases continuously, the throughput of DPS and IDPS algorithms begins to be lower than MPF and M-LWDF algorithms. Compared with Fig.2, although the DPS and IDPS algorithms can guarantee lower packet loss rates when the number of users is more than 60, they are actually achieved at the expense of throughput. When the number of users is larger than 70,the IDPS algorithm gives higher priority to the users with good channel conditions as opposed to the DPS algorithm. Thus, the proposed IDPS algorithm obtains higher throughput.

Fig.3 System throughput vs. the number of users

In Fig.4, the average user delay versus the number of users is shown. It is shown that the MPF and M-LWDF algorithms achieve higher delay than the DPS and IDPS algorithms. This is because the MPF and M-LWDF algorithms do not consider user delay in the resource scheduling. In addition, when the number of users is less than 70, the average user delay of the DPS algorithm is small, but as the number of users increases, the average user delay is increased dramatically. This is because the DPS algorithm may choose users with poor channel quality to schedule, which leads to a sharp rise in the user delay. On the contrary, the proposed IDPS algorithm gives higher priority to users with better channel quality for scheduling when the user delay is the same. Thus, it can significantly reduce the average user delay.

Fig.4 Average user delay vs. the number of users

Fig.5 Jain’s fairness index vs. the number of users

In Fig.5, fairness performance versus the number of users is demonstrated, where the fairness is measured by Jain’s fairness index. It is found that when the number of users in the system is small, the fairness indexes of the four algorithms are very close. With the increase in the number of users, the fairness indexes of MLWDF and MPF algorithms drop sharply, while the DPS and IDPS algorithms decrease gradually. Additionally, the user fairness of IDPS algorithm is slightly lower compared with the DPS algorithm.

4 Conclusion

In this paper, the MBMS resource scheduling algorithms for real-time services in LTE systems were investigated, and an improved delay priority scheduling algorithm, named IDPS, has been proposed. For the proposed IDPS algorithm,when the transmission rate and multicast group are selected to be scheduled, the user delay information and channel conditions are jointly considered in order to reduce the average user delay and packet loss rate. Simulation results showed that compared to the existing MPF, LWDF and DPS algorithms, the proposed IDPS algorithm achieves lower average user delay and packet loss rate. Meanwhile, the IDPS algorithm attains a good balance between system throughput and user fairness.