تجزیه و تحلیل عملکرد و مدل سازی خطاها و تلفات در LAN های 802.11b برای نرخ بیت بالا در چندرسانه ای زمان واقعی

Inherent error-resilient nature of multimedia content renders two high-level options for wireless multimedia application design. One option is to employ (semi-) reliable wireless Medium Access Control (MAC) functions in conjunction with the traditional User Datagram Protocol (UDP). The other option is to employ a less-reliable MAC and transport layer protocol stack that passes corrupted packets to the application layer, which consequently achieves a “higher throughput”. This “higher throughput” traffic, however, could contain many “useless” corrupted packets. In this paper, we address key questions regarding the viability of the above two options for the support of high-bit-rate wireless multimedia applications over 802.11b LANs. First, we study the level of throughput improvements realized by the less-reliable protocol stack at 2, 5.5 and 11 Mbps data rates using actual measurements that mimic realistic home or business settings. Second, we analyze and model the error patterns within the “higher throughput” corrupted packets to evaluate their potential impact on multimedia applications. Third, we compare the amount of overhead that is needed at the application layer to achieve different levels of lost- and corrupted-packet recovery for the two (reliable and less-reliable) protocol stack scenarios. Major conclusions of our study include: (1) Either protocol-stack is viable at 2 Mbps while neither of them is viable at 11 Mbps under realistic settings; and (2) Some benefits of the “higher throughput” corrupted packets can be realized at 5.5 Mbps when combined with a joint erasure-error protection algorithm at the application layer.

The advent of wireless networks has advanced real-time multimedia communication into a novel realm. The wireless medium, due to its inherent vulnerability and physical characteristics, is more error prone than most of the contemporary (wired) media. Nevertheless, migration of technologies from the wired to wireless domain is currently underway. One of the key challenges in this context is the delivery of real-time content over the wireless medium. The promise of real-time multimedia over the wired Ethernet is often attributed to the simplicity and pragmatism of User Datagram Protocol (UDP)/IP protocol suite. However, this protocol suite was designed for considerably low error-rate environments as opposed to wireless networks. In this regard, a positive aspect of real-time applications is their inherent tolerance to a certain level of corrupted and/or dropped packets. This characteristic of real-time applications motivated the development of new methods and protocols for wireless networks. For example, a new transport layer protocol which is tailored for real-time applications over error-prone networks has been proposed recently [1], [2] and [3]. This protocol, known as UDP Lite, relies on the premise that multimedia applications can tolerate, and therefore should, receive corrupted packets. Naturally, this UDP Lite-based framework makes it necessary for the Medium Access Control (MAC) layer to forward corrupted packets to the higher layers. Consequently, developers of wireless multimedia applications are faced with two (high-level) options: (1) Employing current wireless MAC functions, which include dropping of corrupted packets and attempting to recover these packets through retransmissions, in conjunction with the traditional UDP protocol; or (2) Allowing the MAC layer to pass corrupted packets to a UDP Lite-type transport layer which in turn relies on the application layer to handle the corrupted data. A key advantage of the second option is the increase in the, potentially corrupted, throughput of the end-to-end transport layer packets. However, these “high-throughput” packets contain corrupted data. Therefore, and based on the above two options, three key questions can be raised: (1) How much improvement in the “throughput” is actually being gained by using the new UDP Lite-based framework? (2) How badly corrupted these “high-throughput” packets are? In other words, how much and what type of error patterns are observed in these packets? (3) As a consequence of the first two questions, what level of application layer loss-protection, error-concealment and/or error-correction would be required to achieve acceptable quality under the two protocol-stack variants? In this paper, we address the above questions in the context of high-bit-rate1 multimedia applications2 over 802.11b wireless LANs. At this point, it is important to outline how we intend to answer these three questions. The first question, which has been addressed partially by previous studies (see, for example, [3] and [4]), can be answered through a set of experiments and related analysis that is presented in this paper. Our analysis of packet throughput is conducted at the 2, 5.5 and 11 Mbps data rates using actual measurements that mimic realistic home or business settings.3 Answering the second question requires a thorough analysis and some modeling of the error patterns at the MAC layer (i.e., errors that are not corrected by the physical layer). In addition to presenting our experimental results and analysis for byte-level4 error patterns at the MAC layer, we propose a new hierarchical Markov-based model for representing these error patterns. As shown later in the paper, the proposed model captures the observed errors more accurately than the traditional two-state Markov chain. Addressing the third question is naturally more challenging than the first two questions, as it depends on a wide spectrum of objective and subjective evaluations of a variety of multimedia applications and corresponding error protection and concealment algorithms. For example, some applications targeted for wireless networks may take advantage of better error resilience features, such as Reversible Variable-Length Coding (RVLC) that is supported in the MPEG-4 video standard [5]. Other examples include a range of (video) error concealment algorithms and/or Forward-Error-Correction (FEC) methods [6] and [7]. Due to this wide spectrum of methods and algorithms, which can influence the answer to the above third question, we had no option but to focus on one important (sub-) question of this spectrum: What is the amount of overhead that is needed at the application layer to achieve different levels of lost- and corrupted-packet recovery for the two protocol-stack scenarios? The answer to this question provides a worst-case measure of the level of overhead needed to correct/recover a desired level of corrupted/dropped packets for both UDP and UDP Lite. As explained later, this leads to an important conclusion about the viability of multimedia applications at rates higher than 2 Mbps under realistic settings. Our decision to focus on high-bit-rate multimedia has been mainly influenced by the fact that emerging wireless LAN standards are designed to support unprecedented high-bit-rates. In particular, 802.11 LANs are finding their way into homes and businesses. Consequently, it is quite feasible that in the near future these LANs could utilize unicast and/or multicast frameworks to provide real-time television-like services to large numbers of users. As explained above, evaluating this feasibility requires a methodical understanding of the error and loss performance at different layers of the protocol stack. We perform all analysis and modeling above the physical layer. Here, we rely on the assumption that, in addition to its adherence to the constraints of the underlying standard, the physical layer is providing the best possible performance at a given desired bit-rate and under given channel conditions. Hence, our channel definition encompasses the wireless medium and the 802.11b physical layer. The remainder of this paper is organized as follows. Section 2 provides a brief overview of related work in this area. Section 3 describes the experimental setup to generate error traces at 2, 5.5 and 11 Mbps. Section 4, first, studies the throughput of the MAC layer under different loss conditions and uses this insight to conclude an appropriate link layer strategy for multimedia applications. This section, then, investigates the throughput of transport layer protocols, i.e., UDP and UDP Lite in conjunction with the new link layer strategy. A two-state Markov model for packet drop bursts is also presented in this section. Section 5 outlines a heavy-tail-based byte-level burst behavior and proposes two Markov-based models to approximate the burstiness. Section 6 compares the amount of redundancy required by UDP and UDP Lite under the previously mentioned loss conditions. Section 7 summarizes some key conclusions of this paper.

In this paper, we studied the feasibility of two high-level options for the delivery of multimedia content over 802.11b wireless LANs. First, we presented our measurements of MAC layer error and loss traces over an 802.11b network at 2, 5.5 and 11 Mbps rates. We utilized these traces to address our first concern (i.e., how much improvement in the “throughput” is actually being gained by using the new UDP Lite-based framework?) and evaluated the transport layer throughput with different combinations of transport- and link layer variants. More specifically, we compared overall (error-free and corrupted) end-to-end throughput provided by UDP and UDP Lite. In addition, we employed a simple two-state Markov model to approximate the packet drop bursts. We showed that this model provided promising results at all three bit-rates. Furthermore, we conclude that UDP Lite renders a significant throughput improvement by relaying corrupted packets to the application layer. This led us to the next question, i.e., how much and what type of error patterns are observed in these packets? Since the errors in the traces are largely bursty in nature, we again used the two-state Markov models and derived the respective transitional probabilities. Our analysis revealed that traditional Markov models were not always adequate for capturing 802.11b error and loss patterns since the collected traces showed a variety of burst and random error patterns. We also evaluated the byte-error-burst probability density function and showed that its tail can be characterized using a Pareto distribution. This provides further insight into the type and number of errors observed at the aforementioned bit-rates. Therefore, we proposed a hMM with high-level states, each of which includes (i.e., embedded within it) traditional Markov chains. The high-level states were identified based on the “level of burstiness” within different segments of the observed traces. Our analysis revealed that the hMM framework provided a better approximate of byte-level bursts. Lastly, we observed that, UDP Lite improves the overall end-to-end throughput. However, due to the non-stationary and high-bursty nature of errors, the quality of perceived media was largely unaltered when compared with a standard UDP-based protocol stack. Hence, the nature of such bursty errors demands FEC protection at the application layer regardless of the underlying transport layer protocol in order to deliver high-bit-rate multimedia. Therefore, we focused on comparing the FEC overhead required by UDP and UDP Lite. This led to one of the key findings of our study, and that is, the amount of FEC overhead required by UDP Lite is considerably less than traditional UDP. Hence UDP Lite provides improvement in bandwidth utilization (i.e., the amount of redundancy required) in order to deliver lossless multimedia. For example, UDP Lite at 5.5 Mbps uses an FEC overhead of (approximately) 2 Mbps to recover packet drops and corruptions. Therefore, an effective bandwidth utilization of more than 3.5 Mbps was achieved. This illustrates the viability of supporting high-bit-rate and high-quality multimedia applications at rates higher than 2 Mbps under realistic conditions.