A Fast Blind Impulse Detector for Bernoulli-Gaussian Noise in Underspread Channel

Abstract

Impulsive noises widely existing in various channels can significantly degrade the performance and reliability of communication systems. The Bernoulli-Gaussian (BG) model is practical to characterize noises in this category. To estimate the BG model parameters from noise measurements, a precise impulse detection is essential. In this paper, we propose a novel blind impulse detector, which is proven to be fast and accurate for BG noise in underspread communication channels. I. INTRODUCTIONImpulsive noises are widely observed in various communication systems, including ultra wide-band (UWB) systems [1], wireless local area networks (WLAN) [2], digital subscriber line (DSL) networks [3] and power line communication (PLC) systems [4]. Due to their non-stationary nature and high peak power, they can significantly degrade the performance and reliability of communication systems. Such impacts can become critical in urban and industrial environments where 1) frequent mechanical switching operations and vibrations are present to produce dense impulsive noises, and 2) ultra-high reliability and ultra-low latency are expected in short-range wired/wireless communication applications. Various use cases of this kind have been addressed in scopes of both the Fifth Generation (5G) mobile networks [5] and new advanced industrial communication solutions [6]. In these contexts, techniques of modeling impulsive noises are required as a major tool to define channel models and thereby develop robust communication systems.The Bernoulli-Gaussian (BG) model has been widely applied on impulsive noises. Compared to other common impulsive noise models such like the Middleton's Class-A (MCA) model [7] and the symmetric α-stable (SαS) model [8], the BG model stands out with its compatibility to different noise bandwidths, while simultaneously exhibiting a heavy-tailed probability density function (PDF) with a simple closed-form expression. Moreover, it can be easily extended to the Markov-Gaussian model to characterize noise bursts [9].However, despite these superiorities, the deployment of BG model in communications and signal processing has been limited by the insufficient study on its parameter estimation, or more specifically, on the impulse detection. Unlike the MCA/SαS models that describe the overall statistics of the mixed noise, the BG model separates impulses from the

Parameter estimation of impulsive noise for channel coded communication systems

“…The relation to the parameters of the B–G model is given by The relations () and () ensure that the probabilities of the impulse occurrence of both models are the same and the total noise powers of both models are equal, that is, () equals to (). According to previous studies [2, 25 33, 34], the value of is from 10 −4 to 10 −2 and Γ is from 10 −1 to 10 2 for PLC systems. Nevertheless, we apply wider range of and Γ to test the robustness of the parameter estimation.…”

Bit error probability of the M‐QAM scheme subject to multilevel double gated additive white Gaussian noise and η‐μ, κ‐μ, or α‐μ fading

“…Works on 5G systems or wireless sensor networks have addressed impulsive noise in communication systems . In the works of Cheffena and Shhab et al, for example, it is shown that the performance of those systems can be affected by the impulsive noise characterized by Gaussian mixtures, such as the type of noise considered in the present paper.…”

Merging the Bernoulli–Gaussian and symmetric α-stable models for impulsive noises in narrowband power line channels

“…Here we invoke the Narrowband Regression method [17] to cancel periodically fluctuating narrowband interferers from the downsampled measurements. Then we invoke the blind BG impulse detector reported in [16] on the cleaned results to distinguish the spikes from the background noise. An instance of noise preprocessing result is depicted in Fig.…”

“…1. The simple BG model is supposed to be applied on impulsive noises in underspread channels, where the impulse width is significantly shorter than the sampling interval, so that the multi-path channel fading can be ignored [16]. Under a very high sampling rate such as 80 MSPS, such approximation does not hold any more, and the noise must be first de-convoluted from an unknown observation matrix before fitted with the BG model, which would complicate the task.…”