Overview

Providing quality of service (QoS) guarantees is an important objective in the design of the next-generation wireless networks.  In this project, we address the QoS provisioning problem from the network perspective.  We propose and develop a link-layer channel model termed the effective capacity (EC) model.  The effective capacity model captures the effect of channel fading on the queueing behavior of the link, using a computationally simple yet accurate model, and thus, is the critical device we need to design efficient QoS provisioning mechanisms.  We call such an approach to channel modeling and QoS provisioning, as effective capacity approach.

With the effective capacity approach, we obtain link-layer QoS measures for various scenarios: flat-fading channels, frequency-selective fading channels, multi-link wireless networks, variable-bit-rate sources, packetized traffic, and wireless channels with non-negligible propagation delay.  The link-layer QoS measures are characterized by a data rate, delay bound, and delay-bound violation probability triplet. Armed with the EC channel model, we develop simple and efficient schemes for admission control, resource allocation, and scheduling, which can yield substantial capacity gain.

We study some of the practical aspects of effective capacity approach, namely, the effect of modulation and channel coding, and robustness against non-stationary channel gain processes.  We show how to quantify the effect of practical modulation and channel coding on the effective capacity approach to QoS provisioning. We identify the fundamental trade-off between power and time diversity in QoS provisioning over fading channels, and propose a novel time-diversity dependent power control scheme to leverage time diversity.

Background

1. Channel Capacity for a Fading Channel

There are four capacity notions for a fading channel:

1) Shannon ergodic capacity: defined by the expectation of log(1+SNR) where the expectation is w.r.t. the channel gain.

2) Outage capacity:  for a time interval of length T, the outage capacity is the maximum (constant) data rate R achievable with a violation probability P.   Hence, R is a function of P and therefore outage capacity is actually a CDF (cumulative distribution function) of the data rate R.

3) Delay limited capacity: the maximum (constant) data rate R achievable with delay bound Dmax and zero delay-bound-violation probability.  It is defined in

S. Hanly and D. Tse, "Multi-access fading channels: part II: delay-limited capacities," IEEE Trans. on Information Theory, vol. 44, no. 7, pp. 2816-2831, Nov. 1998.

4) Probabilistic delay constrained capacity:  the maximum (constant) data rate R achievable with delay bound Dmax and delay-bound-violation probability PD.   The probabilistic delay constrained capacity is specified by the triplet {R, Dmax, PD}, which is a surface in 3D space.  We call it Pareto frontier or Pareto surface.  Note that Shannon's rate distortion function is also a Pareto frontier (in 2D space).   Effective capacity defined in the following paper is one step forward toward deriving probabilistic delay constrained capacity.

Dapeng Wu and Rohit Negi, "Effective Capacity: A Wireless Link Model for Support of Quality of Service," IEEE Transactions on Wireless Communications, vol. 2, no. 4, pp. 630-643, July 2003. [pdf]

2. Motivation

Historically, the physical-layer designers used signal-to-noise ratio (SNR) and bit error ratio (BER) to quantify
the performance of wireless communication systems since the wireless communication systems used to be circuit-based and be designed to support constant-bit-rate voice traffic; hence, queueing theory and queueing performance are not the concerns of physical-layer designers.  In contrast, the network designers use data rate, delay bound, delay-bound violation probability, or packet loss ratio to quantify the performance of packet-based networking devices and networks;
since wired packet-based networks such as Internet are built on inherently reliable, low noise communication channels, the network designers are not interested in SNR, BER, communication theory, and information theory as the physical-layer designers.

With the emergence of broadband packet-based wireless networks and increasing demand of multimedia information on the Internet, wireless multimedia services are foreseen to become widely deployed in the next decade.  To support multimedia transmission over wireless channels, it is important to consider both the physical-layer QoS (e.g., SNR) and the networking-layer QoS (e.g., delay performance) since both physical-layer bit errors and networking-layer buffer overflow can cause errors, which negatively affect the upper-layer multimedia applications.

However, due to the separation of the two camps (i.e., communication theorists and queueing theorists) in the past, there has been a big gap between the physical-layer QoS and the networking-layer QoS. For example, even if the physical-layer designers provide the SNR and BER performance of a wireless communication system, the network designers have no idea of what the delay performance the wireless communication system will achieve. Obviously, the mapping from the physical-layer QoS to the networking-layer QoS is lacking.  This project is intended to develop novel approaches and theories to this problem.  The techniques developed are expected to be applicable to various emerging wireless communication systems such as 3G/4G, WiMax, mesh networks, and satellite data networks.

  1. Approaches:
  2. Why is large deviations theory applicable to QoS provisioning with small delay bound requirements?

Wireless fading channels are different from wired channels.   Rayleigh fading with a specific Doppler spectrum, log(1+SNR), make the delay bound violation probability having exponential form.  Systematic studies are needed.

  1. Methods: Simulation-based optimization (Abhijit Gosavi), computational probability (Neuts), computational operations research, experimental mathematics (Stephen Wolfram), stochastic approximation theory (G. Yin).
  2. Multi-objective optimization, Pareto optimal, Gradient method 

Project Investigators

Research Results (coming up soon)

Questions and Answers:

  1. I am a student and my research is about radio resource allocation. Recently, I am reading your paper, named Downlink Scheduling in a cellular network for quality-of-service assurance, and I am very interested in it and find it has been cited by many paper. there is some question that I can't understand and I was hoping to get help from you. The problems is as followsㄩin paper, you said that in reference 5 you use a simple and efficient algorithm to estimate the QoS exponent function by direct measurement of queueing behavior resulting from channel capacity process r(t). But when we consider admission control, we can't get the measurement of queueing behavior before we allocate the channel to the data source. In this condition, how does the scheduler execute the admission controlㄛthat is how we check whether the QoS exponent of a channel is larger than the user-required one?

Answer:  It takes a few seconds to dial a phone number.    During this dialing period, a cell phone can measure the channel condition; there is a common pilot channel (CPICH) in 3G WCDMA system; measure the channel gain of CPICH.  Then using the channel gain to calculate instantaneous Shannon capacity and apply it to a virtual queue; then collect statistics for computing effective capacity.   So during the dialing period, the effective capacity can be estimated and admission control can be done.

  1. I'm reading your paper, called Downlink Scheduling in a Cellular Network for Quality-of-Service Assurance. As far as I know, this paper is cited by many Academic articles and thanks for your contribution to this field. I'm very interested in it. However there is a problem that I can't understand and I need your help. The problem is as followingㄩ
    In your admission-control and resource-allocation scheme, scheduler needs to assure that the resources allocated to a user is sufficient to provide QoS assurance. Assuming the K&H/RR scheduling algorithm is used and there are 2 users and 2 channels, the fractions of the frames in the two channels, which every user can get, are different in every scheduling process. My question is how to calculate the QoS exponent $\theta$ for the case where User 1 always has a higher channel gain than User 2 for all the frames/slots.

Answer:  The QoS exponent $\theta$ is a performance measure with respect to the whole sample space, rather than w.r.t. a special event that User 1 always has a higher channel gain than User 2.   In other words, all events will contribute to the value of QoS exponent $\theta$.

  1. For estimation of QoS exponent, you use a simple algorithm because it is hard to use the existing physical-layer channel model to estimate the QoS exponent. Is the precondition of the estimation algorithm that wireless channel condition ,SNR, must follow Ricean or Rayleigh distribution? But for CDMA system we use SNIR rather than SNR to denote the channel condition, dose the channel condition, SINR, follow a distribution?

Answer:  How to define effective capacity for an interference-limited system such as CDMA is still an open problem.  We are still working on it.

Publications

Related Work

CDF scheduling (maximum quantile scheduling)

﹛

Useful Links

﹛