Aiming at the problem of statistical delay assurance in high-speed IEEE 802.11 wireless local area network (WLAN), a novel frame aggregation algorithm of proportional-integral-differential (PID) controller is proposed. By constructing an effective capacity model, the aggregation algorithm does not need to take any channel information into account. Only using the average queue level metric, it can achieve the statistical delay guarantee of QoS. The proposed PID controller is implemented at the access point to determine the time margin allocated to each link. The simulation results show that the proposed aggregation algorithm based on PID is superior to the earliest due-first-service algorithm with the largest aggregation size and the deadline-based aggregation algorithm, which improves the channel utilization and the maximum number of wireless stations. With the increasing popularity of network broadcasting applications, such as Youku and live network television, high-quality video delivery is becoming more and more important for service providers, equipment and network providers [1?2]. In addition, it is expected that by 2019, due to the rapid development of smartphones, consumer video traffic will constitute 80% of all Internet consumption traffic. As a result, most video traffic will eventually be delivered through wireless links, especially through WiFi networks, such as the Wireless Local Area Network (WLAN) industry, which has responded to this demand by introducing the IE802.11n and the latest IE802.11ac standards [4?5]. However, due to the contention overhead and the slow speed of sending long headers, the high data rate available in MIMO is not well utilized. According to the standard of IE802.11, there are two types of aggregation: aggregated media access control service data unit (A?MSDU) and aggregated media access control protocol data unit (A?MPDU). In order to obtain the best performance, the channel conditions should be considered while setting the aggregation size. For example, if the bit rate is low, then a large transmission opportunity (TXOP), that is, a large aggregation size, will occupy limited wireless resources [6?7]. Similarly, transmission of small packets at high data rates will not make full use of MIMO links. In this paper, the theory of effective capacity is used to set appropriate TXOP in fluctuating channels.
This paper proposes a frame aggregation algorithm based on effective capacity, which can provide statistical delay guarantee in high-speed wireless LAN. The aggregation scheme proposed in this paper does not use physical layer or channel state information, but it can find an appropriate aggregation size in complex channels. The proposed method based on effective capacity only depends on the queue level index. This aggregation algorithm is implemented on Enhanced Distributed Channel Access (EDCA) and is supported by all WiFi devices. In addition, it can be fully implemented at the transmitter end, such as downlink traffic using access point (AP), thermostatic element without any changes to the IEEE 802.11 MAC. Compared with the earliest due first (EDF) algorithm and deadline-based aggregation algorithm, the proposed aggregation scheme is better. EDF algorithm needs to schedule the largest aggregation size in terms of maximum number of wireless stations (STA) and channel efficiency. In the formula, [gamma(mu)] denotes the utilization rate of links, [theta(mu)] denotes the quality of service index of links, [Dmax] denotes a given statistical delay limit, [epsilon] denotes the initial set probability value, and the timeout probability denotes [PrD(t)>Dmax]. Statistical QoS guarantees [gamma (mu), theta (mu)] are expressed in the form of [Dmax, epsilon]. Formula [C(t)] denotes the cumulative random channel service distribution. The statistical delay is determined by the QoS index. In the formula, [S] denotes the average remaining service time of sampling transmission, [Q] denotes the average queue length. The [theta,] in the solution (1) requires that the time-out probability be less than or equal to the target value of [e], i.e. [PrD(t)>Dmax<e]. The given statistical QoS requirements can be expressed in the form of delay constraints and target timeout probability [Dmax, e]. It can be seen that all parameters, including the probability of non-empty link queue [gamma], the average queue length [Q], and the average residual service time [S], can be easily and accurately estimated at the access point. Suppose an independent 802.11n basic service set (BSS) has many wireless stations and an access point to forward traffic to wireless stations [10?11]. Each frame arrives at each wireless station and needs to be queued at the access point and sent out in the form of aggregation. According to the aggregation scheme proposed in this paper, the size of the aggregation scheme is determined by the beacon of the access point. Suppose a downlink [l] [l = 1,2,… L], the necessary condition for the specified QoS is [Dl, epsilon]. During each beacon interval BI, the link specifies a time-width of [Tl]. The optimization problem is still non-convex. Therefore, by designing the PID controller, this paper uses heuristic method to solve the optimization problems in formula (7). It is concluded that recording the error of [beta-0] is an appropriate choice for the construction of a PID-based controller. The schematic diagram of the PID controller in a given link is shown in Figure 1. The access point often controls [beta l,] and at the end of each beacon interval, it can calculate the error value, [el=beta l-0] applies to all links. Then the time duration of each link is updated for the next beacon interval. According to formula (13), the accumulated sum of the appropriate gain [kP,] [kI] and [kD] is applied to the error respectively by applying the PID control rule. The performance of the proposed PID-based aggregation algorithm is compared with that of the other two algorithms, that is, the earliest due priority (EDF) algorithm with maximum aggregation [10] and the deadline-based aggregation algorithm proposed in reference [6].
The two algorithms are abbreviated as “earliest due first” and “deadline” respectively. In ns? 3 Simulation environment, suppose that there is an independent 802.11n basic service unit, including one access point and 10 wireless stations, which receives all traffic from the access point downstream. This scheme is implemented entirely at the access point, and it does not modify any function of 802.11 media access control layer. When evaluating the proposed PID aggregation algorithm, channel utilization, end-to-end average delay time and timeout probability are analyzed. Table 1 lists the main simulation parameters. When the number of wireless stations varies from 1 to 10 and the target timeout probability is 1%, the proposed PID aggregation algorithm (referred to as PID control) and the earliest due priority algorithm and deadline algorithm are compared in Figure 2 to Figure 4. Fig. 2 shows the channel utilization, and shows that the proposed PID aggregation algorithm has the best performance. Especially when there are four wireless stations, the PID can satisfy the required timeout probability, while maintaining the channel utilization rate close to 50%, while the earliest due-first algorithm and deadline algorithm use more channel time, 80% and 75% respectively. When there are eight wireless stations, the maximum utilization rate of PID is 98%, while the earliest due first algorithm and deadline algorithm reach almost 100% and 95% respectively in seven wireless stations.
The low utilization of deadline algorithm is due to its large queue at access point, which results in packet loss.
As shown in Figure 3, compared with the earliest due-first algorithm and deadline algorithm, the proposed PID control algorithm serves more wireless sites and can satisfy the statistical delay guarantee [5 s, 1%]. Especially when the target value of [e] is 1%, the deadline algorithm has the minimum capacity to support only five wireless stations, while the earliest due-first algorithm can support more than one wireless station. However, PID can support up to eight wireless sites, providing about 30% more capacity than the earliest due-first algorithm. This is due to better channel utilization, as shown in Figure 2 above. In addition, when the number of wireless stations varies from 1 to 10 and the target timeout probability is 1%, the average delay of each aggregation algorithm is shown in Figure 4. The deadline-based aggregation algorithm maintains a delay limit of slightly less than 5 s when the wireless station reaches 6. However, the average delay will soon increase after that.
The earliest due-first algorithm can provide very small delay when the number of wireless stations is small, which indicates that earliest due-first algorithm over-supplies data packets, which also explains why it has a slightly lower channel efficiency than the deadline algorithm. However, the average delay of PID is between the earliest due priority algorithm and the deadline algorithm, but it is still far below the target delay limit, which is acceptable. Next, we evaluate how to implement the proposed PID aggregation algorithm according to different QoS requirements [D, e]. Figure 5 shows the effect of delay bounds on channel utilization for each wireless station. Due to the relaxation of delay bounds, the proposed PID aggregator can make more efficient use of the channel by assigning a smaller time margin to each wireless station. For example, when there is only one wireless station and the delay limit varies from 1 to 20 seconds, the time margin of each wireless station is reduced from 15.38 MS to 11.7 ms, which is nearly 25%. This is an important improvement if there are many wireless stations. For example, when there are only six wireless stations and the delay limit varies from 1 to 20 s, the channel utilization can be reduced by 11%.
In this paper, a new frame aggregation algorithm is proposed, which provides statistical delay guarantee for high data rate WLAN. The aggregation algorithm in this paper does not need to consider any channel information, because it only uses the average queue level measure and combines the concept of effective capacity to achieve statistical delay guarantee. The method in this paper is based on the PID controller implemented at the access point to determine the time grace allotted to each link.
The simulation results show that the proposed aggregation algorithm based on PID is superior to the earliest due-first-service algorithm with the largest aggregation size and the deadline-based aggregation algorithm.