Channel Adaptive Techniques in the Presence of Channel Prediction Inaccuracy 1 Ana Aguiar, Holger Karl, Adam Wolisz TU Berlin Einsteinufer 25, 10587 Berlin, Germany {aaguiar|karl|wolisz}@tkn.tu-berlin.de fon: +49.30.31423817, fax: +49.30.31423818 Abstract : Adapting transmission parameters to the future channel state is an appealing approach to improve efficiency in wireless communication. Adaptation requires predicting the channel state. Current channel-adaptive techniques as- sume that the prediction is perfect. In this paper, we claim that neglecting the prediction error can lead to poor perfor- mance results, possibly even worse than without prediction at all. We have simulated the behaviour of adaptive modulation and adaptive scheduling, as well as a combination of both for in- accurate channel prediction. The results show that prediction inaccuracy can degrade the performance of wireless commu- nications when channel adaptive techniques are used. The sensitivity to inaccuracy is, however, dependent on several factors, like the adaptive technique used or the average re- ceived signal-to-noise level at the receiver. Keywords: prediction accuracy, channel-adaptive techniques, simulation 1 Introduction The widespread use of wireless communications has increased performance and reliability requirements. As the wireless channel is a very error-prone medium, error mitigation and control techniques have to be used. How- ever, the error characteristic is time-varying and a static policy leads to low error performance or low resource use. An appropriate solution to mitigate the effects of time-varying errors is to adapt transmission parameters to the channel state. Such channel adaptive techniques (CAT) — e.g., adapting the used modulation or transmission power or delaying the transmission of packets — assume that in- formation about the future channel state exists and is available. In fact, several methods exist which can pro- vide the sender with an estimate or prediction of the channel behaviour (namely, its attenuation) [3, 7]. How- ever, none of these methods is perfect: the prediction results are inaccurate and do not correspond exactly to the future behaviour of the channel [6]. The influence of this inaccuracy on the performance of channel adaptive techniques evidently has a large impact on the suitability of such techniques and should hence be carefully studied — a study that is largely missing so far. We developed a framework which enables this kind of study for different channel-adaptive techniques and combinations of them [1]. In this paper, we present the 1 This work has been supported by Siemens AG, ICM, and the Por- tuguese Foundation for Science and Technology and the European So- cial Fund under the scholarship nr. SFRH/BC/4601/2001. results obtained for adaptive modulation and adaptive scheduling [2], as well as the combinations of adaptive scheduling with adaptive modulation. The results show that the impact of prediction inaccuracy has to be care- fully taken into account when designing and using CATs. The rest of the paper is organised as follows: Section 2 describes our approach to capture prediction inaccura- cies in a simulation framework; Section 3 discusses the channel adaptive techniques that we investigate. The re- sults of this investigation are presented and discussed in Section 4; Section 5 concludes. 2 Modelling Prediction Inaccuracy There are a number of different approaches to pre- dict channel behaviour [3, 7]. As all of them are fairly complex and resource-intensive, it would be very diffi- cult and time-consuming to actually implement, e.g., in a simulation model, every existing solution to channel prediction and then evaluate their effects on the channel- adaptive techniques. In particular, simulating such a model with actual prediction algorithms inside would not be feasible at acceptable runtimes and at acceptable levels of statistical confidence. Hence, we use a model for prediction inaccuracy instead of the actual prediction algorithms. This model can capture and is adapted to the inaccuracy of a channel predictor, without depending on the prediction algorithms themselves. This prediction model (Section 2.2) is based on an analog channel model (Section 2.1) and is used in the context of a system model (Section 2.2). The resulting simulation framework is de- scribed in more detail in [1]. 2.1 Channel Model The channel behaviour is expressed by the attenua- tion in dB, a k , at the discrete time instant k. We use a Rayleigh fading channel model to generate a k . The granularity of the samples depends on the bandwidth of the Rayleigh process and, consequently, on the speed of the receiver (which determines the maximum Doppler frequency). The value of the attenuation is then used, together with the transmitted signal power and the noise power (both in dBm), which is assumed to be additive, white and Gaussian (AWGN), to calculate the signal to noise ratio (SNR) for each sample. Depending on the used modu- lation scheme, the received signal-to-noise ratio is con- verted to energy-per-bit E b /N ; this value is then used to calculate the probability of a bit error, from a curve relating E b /N to BER for the used modulation [5].