Channel capacity of Middleton's class A interference channel

Abstract

Middleton's class A interference model has properties that make it possible to represent a wide class of interference signals. By choice of model parameters, interference signals ranging from pure Gaussian distributed to pure impulsive interference can be modelled. These properties make the model very useful for a large variety of applications. However, an expression for the channel capacity of the class A interference channel has not yet been published. The channel capacity of this model is derived, and numerical examples are given for some useful sets of model parameters.Introduction: The performance of communication systems is impaired by interference, or noise, from both external and internal sources. A common approach in the analysis of interference impact is to model the interference as having a Gaussian amplitude distribution with flat power spectral density (additive white Gaussian noise, AWGN). The advantage of the AWGN model is its simplicity for analysis and that a large amount of results can be found in the literature, most notably Shannon's channel capacity results for the AWGN channel [1]. However, the AWGN model does not cover the behaviour of a large class of commonly occurring interference signals, especially impulsive interference. A more accurate interference model is the Middleton's class A interference model [2]. This model has the advantage that it can represent a number of interference signals with arbitrary impulsiveness content. By choice of model parameters, we can model a large class of interference ranging from purely Gaussian distributed noise to highly impulsive interference. A communication channel experiencing class A interference is referred to as an additive white class-A noise (AWCN) channel [3]. The drawback is that the analysis of the AWCN channel is not as simple as in the AWGN case. This is probably the reason why the model has not gained as much popularity as the simpler AWGN model. However, the interest for the class A model is increasing and it is more and more often seen in interference analyses.Areas where the AWCN channel model is applicable include dynamic spectrum access (DSA) for cognitive radios (CRs) where other communication systems act as impulsive interferers and IEEE 802.11 g, an orthogonal frequency division multiplexing (OFDM) system operating in the 2.4 GHz ISM band where microwave ovens can create impulsive interference. To maximise the spectrum utilisation in, for example, an OFDM system, waterfilling can be used to efficiently distribute signals over parallel channels. However, waterfilling requires that the channel capacity is known. Such an expression has not yet been published. In [4], an upper bound of the channel capacity is calculated for bivariate Middleton class A interference, which is an extension of the original Middleton's class A interference. However, that work is based on a different approach, which results in an upper bound of the capacity. Our present work is based on the fact that Middleton's class A model can be viewed as a Markov chai...

On the suitability of the Middleton class A noise model for narrowband PLC

“…The first one is its simplicity, since it has only two parameters: the impulsive index, A, which characterizes the impulsiveness of the noise, and the Gaussian to impulsive noise power ratio, Γ. The second reason is that the performance of communication systems impaired by this type of noise has been largely studied [10][11] [12]. Hence, A and Γ can be used to develop a communication-oriented classification of NB-PLC noise, which would be particularly useful in the context of AMI deployment.…”

Estimating information rates of Bernoulli-Gaussian impulsive noise channels in Rayleigh fading

“…From Lemma 1, it can be seen that the error of approximation in (11) will not exceed ε for γ ≥ γ max . Next, in the range γ ∈ [0, γ max ), we will apply the PWLCF algorithm in [18] to approximate the differential entropy curve h GI (γ).…”

“…Impulsive noise is also observed in wireless fading channels, such as urban and indoor radio channels [4], cognitive radio [5], [6], as well as open spectrum access schemes [7]. Among various statistical models of the impulsive noise such as Bernoulli-Gaussian (BG) noise [8], [9], Gaussian-Mixture noise [10] and Middleton Class-A, Class-B, Class-C [11], the BG noise is of practical interest, due to its accuracy it approximating the impulsive behavior for various communication channels [9], [12], [13].…”

“…Up to date, most information-theoretic studies on impulsive noise focused on static channels and the estimation of capacities/information rates was mostly based on Monte Carlo simulations [9], [11], [14], [15]. For instance, over a static impulsive noise channel, it was shown in [14], [15] that a Gaussian input distribution can be considered as a good choice for impulsive noise channels.…”

Incremental Selective Decode-and-Forward Relaying for Power Line Communication

“…As the channel is corrupted by background noise with probability p 1 = (1 − p), and background and impulsive noise with probability p 2 = p, the instantaneous channel capacity can be expressed as [15]…”