Abstract-The purpose of this paper is to examine (i) some of the models commonly used to represent fading, and (ii) the information-theoretic metrics most commonly used to evaluate performance over those models. We raise the question of whether these models and metrics remain adequate in light of the advances that wireless systems have undergone over the last two decades. Weaknesses are pointed out, and ideas on possible fixes are put forth.
I. MOTIVATIONMultipath fading is one of the most essential attributes of wireless channels and, as such, the characterization of its impact on fundamental communication limits has been the object of much scrutiny (cf.[1] and references therein). This characterization is difficult because the physical mechanisms that underlie fading are intricate, and thus not easily modeled in a way that facilitates analysis and insights. Recognizing this, a few canonical settings that offer a compromise between realism and tractability have become established over time. These settings, involving models for the fading, operational assumptions, and metrics, have served the research community very well for years. The question addressed in this paper is whether these settings remain adequate in light of the rapid advances experienced by wireless systems in their characteristics and operating conditions.
II. CANONICAL SETTINGSIn the settings object of our interest, fading is modeled as a stochastic process. The marginal modeling is not problematic: the central limit theorem applied to the sum of a large number of multipath components yields a theoretical Rayleigh distribution for the amplitude, and experimental measurements have repeatedly confirmed the validity of this distribution [2].It is the modeling of the fading dynamics that presents the most complications. From an information-theoretic vantage, what is of essence is the selectivity of the fading over each codeword and the two most common canonical settings idealize this selectivity in limiting senses:• Ergodic setting. The fading varies ergodically over the span of each codeword.• Quasi-static setting. The fading is fixed over the span of each codeword, and varies only from codeword to codeword. Both these settings represent an admittedly drastic simplification of reality, deemed necessary in order to enable rigorous mathematical treatment. We next elaborate on these two settings, with emphasis on the operational assumptions that have to do with channel-state information (CSI) availability at the transmitter.
A. Ergodic SettingIn this setting, each codeword is subject to the entire distribution of the fading. This entails a high degree of selectivity, which can be modeled as continuous but more often is modeled simply as a succession of blocks [1]. While artificial, a block-fading structure further simplifies the setting while retaining its essence.The ergodic setting has a well-defined capacity in the Shannon sense [1], the so-called "ergodic capacity", which entails an expectation over the fading distribution. The ergodic capacity can...