Skip to content

SMACS Research Group

Sections
Personal tools
You are here: home » research

Research topics of the SMACS Research Group

Research of SMACS focusses on the statistical analysis of buffers as well as on the study of ARQ retransmission protocols.

Prof. Dr. Herwig BRUNEEL leads the SMACS Research Group which operates in the area of stochastic modeling. The main applications of the work done by this group are situated in the performance assessment of digital communication systems and networks. The research activities of the SMACS Research Group can be more or less divided in two parts. The major topic is the statistical analysis of buffers for the storage of digital information, by means of discrete-time queueing models. The second (minor) topic is the design, analysis and optimization of ARQ retransmission protocols. Both subjects are explained in more detail below.

Statistical analysis of buffers

In digital communication networks, buffers are used for the temporary storage of information units which cannot be transmitted instantaneously to their destination when they are generated or when they arrive at a given point in the network. Thus buffers are encountered in multiplexers, concentrators, switching elements, resequencing units, rate adapters, (leaky-bucket) traffic shapers, etc., or, more generally, in any subsystem of a network where the generation of information is not well adapted to the transport capabilities of the system, or where some form of competition exists between the various information units with respect to the use of the available resources. A deep understanding of the behavior of these buffers and the mechanisms that lead to this behavior is of crucial importance to network designers, because the performance of the network may be very closely related to it. For instance, information units may get lost whenever a buffer is fully occupied at the time of their arrival at this buffer, or they may experience undesirable delays or delay variations in buffers, and so on.

Usually, information sequences are transmitted in digital communication networks according to the synchronous transmission mode. This means that the digital information is chopped in packets of fixed length, that time is divided in slots of constant length (such that one slot suffices for the transmission of one packet), and that packet transmissions must necessarily begin (and end) at slot boundaries, that is, at a discrete sequence of time points. Therefore, discrete-time queueing models are very natural tools for the statistical analysis of the behavior of buffers in digital communication networks. In these models, the arrival stream of digital information into the buffer is commonly characterized by specifying the numbers of arriving packets or messages (where each message may be composed of a variable number of fixed-length packets) in the consecutive slots. In basic models, these numbers of arrivals are assumed to be independent and identically distributed discrete random variables, and the corresponding arrival process is referred to as an uncorrelated arrival process. In some applications, however, such descriptions are not sufficient to characterize the possibly very complicated arrival streams that may occur (for instance, in integrated-services digital networks which carry a great variety of information types and services). Therefore, more advanced models will allow the numbers of arrivals in consecutive slots to be nonindependent. Such models are referred to as correlated arrival processes.

The channels (servers) which are being used for the transmission of information units from a buffer are not always permanently available. For instance, they may be subject to failures or transmissions may be sometimes unreliable. Also, a channel may have to be shared with other types of information streams having a (momentarily) higher priority than the information in the buffer. In order to be able to deal with these kinds of situations, more complicated queueing models are required in which the possibility of (random) server interruptions is incorporated.

The SMACS Research Group has a long history of (internationally recognized) experience with the development of fundamental discrete-time queueing models and associated techniques of analysis which can be used for the performance evaluation of various types of buffers in terms of such performance measures as the probability distributions of the buffer contents (expressed in packets or messages) and the packet or message delay, the loss probabilities of packets, the busy and idle periods of the system, etc. Specifically, a great variety of analytical techniques have been developed by SMACS for the analysis of discrete-time queueing models with one or multiple servers, correlated or uncorrelated arrival processes, with or without server interruptions, etc. Although this research work is essentially of a very fundamental nature, it has been the object of increasing interest on the part of the telecommunications industry during the last several years because of its direct applicability in the design and performance assessment of various subsystems of Broadband Integrated-Services Digital Networks (B-ISDNs), such as ATM multiplexers and switches. The SMACS Research Group is currently involved in various major research projects in this domain, either sponsored by governmental organizations (Flemish, Belgian or European) or in cooperation with industrial partners.

Study of ARQ retransmission protocols

The communication channels used to transmit digital information from a transmitting unit to one or more receivers are usually subject to noise, which implies that transmission errors are, in general, inevitable. One way to deal with this is to use a so-called ARQ retransmission protocol (ARQ = Automatic Repeat reQuest). Whenever errors are detected by a receiver in a packet of information, a negative acknowledgement signal is forwarded to the transmitter. Upon reception of this signal, the transmitter automatically sends a new copy of the erroneous packet. This procedure is repeated possibly several times, until a correct version of the erroneous packet is received by the receiver. Various ARQ protocols may differ in the way the transmissions and retransmissions of packets are organized or in the way the packets are reconstructed at the receiver side (i.e., the demodulation procedure being used). Their performance is measured in terms of the throughput efficiency - which is a measure for the fraction of time the communications channel is effectively being used for the transmission of correct information - and the delay characteristics of the packets.

The SMACS Research Group has a well-established reputation and long-term experience with the design, analysis and optimization of a great variety of ARQ protocols. A variety of high-performance ARQ schemes for channels with high error rates and long propagation delays (such as satellite channels) have been developed in the group.

Perspectives

Due to the more and more increasing demand of users for the integration of voice, video and data communication, the development of multiservice networks is an important research topic in the telecommunication community at this moment. Especially, the real-time transport of multimedia services over packet-based networks (based on ATM (Asynchronous Transfer Mode) or IP (Internet Protocol)) is a hot topic nowadays. In order to be able to assess the QOS (quality of service) (in terms of packet loss, delay, jitter) in the context of transporting voice and video over packet-based networks, new types of buffer models must be investigated, which are substantially different from those analyzed in the past. Typical features that arise when considering for instance VOIP (Voice over IP) or VTOA (Voice and Telephony over ATM) networks are variable service times combined with multiple priority classes, complex correlated arrival processes, scheduling and discard strategies, service interruptions, etc. All these features need to be incorporated in the existing buffer models. This circumstance will result in very complicated queueing models and turns the development of appropriate new queueing analysis techniques into an enormous challenge for the years to come.