Robust Blind Channel Estimation for Bursty Impulsive Noise with a Constrained EM Approach

Chin-Hung Chen ⋆, Ivana Nikoloska ⋆, Wim van Houtum ⋆†, Yan Wu †, Boris Karanov ‡, Alex Alvarado ⋆ · 2025

Read Robust Blind Channel Estimation for Bursty Impulsive Noise with a Constrained EM Approach by Chin-Hung Chen ⋆, Ivana Nikoloska ⋆, Wim van Houtum ⋆†, Yan Wu †, Boris Karanov ‡, Alex Alvarado ⋆ (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-03685.txt.

Introduction

Chin-Hung Chen and co-authors sharpen blind channel guessing for wireless links hit by bursts of impulsive noise. Their 2025 paper studies crackle from power devices, which can swamp sensitive receivers. Detectors need channel knowledge, yet pilots may be absent.

The guesser runs an expectation-maximization loop shaped to a burst-noise model with a trellis skeleton. The constrained version ties some state spreads together and mirrors the switch matrix. That cuts guess time nearly in half and lifts accuracy over the standard loop. Stress tests with wrong channel facts show both versions hold up, though higher doubt slows settling. The authors frame the constraint as free speed for burst channels.

Catalog record

CREATORChin-Hung Chen ⋆, Ivana Nikoloska ⋆, Wim van Houtum ⋆†, Yan Wu †, Boris Karanov ‡, Alex Alvarado ⋆
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-03685

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Robust Blind Channel Estimation for Bursty Impulsive Noise with a Constrained EM Approach Chin-Hung Chen ⋆, Ivana Nikoloska ⋆, Wim van Houtum ⋆†, Yan Wu †, Boris Karanov ‡, Alex Alvarado ⋆ Abstract Impulsive noise (IN) commonly generated by power devices can severely degrade the performance of high-sensitivity wireless receivers. Accurate channel state information (CSI) knowledge is essential for designing optimal maximum a posteriori detectors. This paper examines blind channel estimation methods based on the expectation-maximization (EM) algorithm tailored for scenarios impacted by bursty IN, which can be described by the Markov-Middleton model. We propose a constrained EM algorithm that exploits the trellis structure of the IN model and the transmitted binary phase-shift keying (BPSK) symbols. By enforcing shared variance among specific trellis states and symmetry in the transition matrix, the proposed constrained EM algorithm adapted for the bursty IN channel has an almost two times faster convergence rate and better estimation performance than the standard EM approach. We comprehensively evaluate the robustness of both standard and constrained EM estimators under different types of CSI uncertainties. The results indicate that the final estimations of both EM estimators are robust enough to mismatch Markov-Middleton model parameters. However, as the level of CSI uncertainty increases, the convergence rate decreases. I Introduction This paper examines blind estimation techniques for channels affected by bursty impulsive noise (IN) generated by power devices, such as power lines and DC-DC converters. It is well established that these electronic components, found in power substations and electric vehicles (e.g., [ 1 , 2 ] ), emit consecutive high-energy spikes, which can seriously degrade the performance of high-sensitivity wireless communication receivers [ 3 , 4 , 5 ] . Therefore, several studies have concentrated on modeling bursty IN to develop robust receivers equipped with specialized detectors. The well-known bursty IN models including the two-state Markov-Gaussian model [ 6 ] and the finite-state Markov-Middleton model [ 7 ] adopted the hidden-Markov model (HMM) to address the correlation between IN samples. Our previous study [ 2 ] also discovered that a Markov-Middleton model effectively represents the EV-induced interference observed in field tests. Conventional detectors for handling IN, such as clipping [ 10 ] and blanking [ 11 ] , are widely adopted due to their simple implementation and low computational complexity. However, their performance is inherently suboptimal. In contrast, the design of optimal IN detectors, which utilize the statistical characteristics of the underlying channel and employ maximum a posteriori (MAP) detection, has been extensively investigated across various IN models and transmission formats [ 6 , 7 , 8 , 9 ] . In [ 12 ] , a thorough comparison of various IN detectors is presented. Optimal MAP detectors can be implemented using the Bahl-Cocke-Jelinek-Raviv (BCJR) algorithm [ 13 ] to compute posterior probabilities; however, these detectors are highly sensitive to channel state information (CSI) [ 6 , 7 ] . Therefore, accurate parameter estimation is crucial for the optimal IN detectors. Channel estimators that use the expectation maximization (EM) algorithm have received considerable attention in the field of wireless communications, largely due to their ability to iteratively improve parameter estimates without depending on training data. This advantage has made the EM algorithm a powerful method for blind channel estimation in dynamic and noisy environments. In recent decades, the EM estimator has become a mature topic for wireless channels affected by intersymbol interference (ISI) [ 14 , 15 , 16 , 17 , 18 ] . Although similar EM-based methods have been adapted for IN channels, most studies focus on memoryless IN, where noise samples are independent, and the state transition probability is considered trivial [ 19 , 21 , 20 ] . In [ 22 ] and [ 23 ] , an approximate message-passing algorithm was proposed to estimate the correlated transmitted symbols and the bursty IN in parallel through a loopy factor graph. Recently, [ 24 ] introduced an EM-based estimator tailored for bursty IN channels with uncorrelated transmitted symbols. This implementation combines transmitted symbols and IN into one loop-free factor graph, facilitating the estimation of channel likelihood and state transition probabilities. Our study builds upon the work of [ 24 ] by proposing a constrained EM algorithm that enforces shared variance among specific trellis states and ensures symmetry in the transition probabilities. Compared to the standard EM algorithm adopted in [ 24 ] , we demonstrate that the constrained EM estimator for the bursty IN channels is more robust and converges faster. Moreover, we perform a thorough robustness analysis to evaluate the estimation performance of both standard and constrained EM estimators across various initialization errors, reflecting the different levels of CSI uncertainties that can occur in practice. This paper is structured as follows: Sec. II summarizes the transmission over the Markov-Middleton model. Sec. III illustrates the standard and constrained EM algorithms. In Sec. IV , we present the numerical simulation for the considered EM estimators. Finally, Sec. V summarizes the paper. II System Model In this paper, we assume a binary phase-shift keying (BPSK) modulation, where a symbol mapper maps the binary bits u_{t}\in\{0,1\} to a BPSK symbol x_{t}\in\{-1,+1\} . The symbol time step is denoted by t . For transmission over a Markov-Middleton IN channel, the received observation sequence \mathbf{y}_{1}^{T} of length T can be described by \mathbf{y}_{1}^{T}=\mathbf{x}_{1}^{T}+\mathbf{n}_{1}^{T} , where we use boldface letters to denote a sequence, with subscripts and superscripts denoting the start and end of the sequence. The statistical properties of noise samples n_{t} are completely determined by the noise state realizations at time t , which is defined as w_{t}\in\mathcal{W}=\{0,1,\cdots,W-1\} , where w_{t}=0 represents the background nose state, and W is the total number of noise states. The generic probability density function (PDF) of the Markov-Middleton noise sample n_{t} can be written as the finite Gaussian mixture model [ 7 ] as \displaystyle p(n_{t}) \displaystyle=\sum_{j\in\mathcal{W}}P(w_{t}=j)\cdot p(n_{t}|w_{t}=j) \displaystyle=\sum_{j\in\mathcal{W}}\frac{P(w_{t}=j)}{\sqrt{2\pi\sigma^{2}_{w_{j}}}}\exp{\left(-\frac{n_{t}^{2}}{2\sigma^{2}_{w_{j}}}\right)}, (1) where we use p(\cdot) to denote a PDF for continuous random variables and P(\cdot) to denote a probability mass function for discrete random variables. In ( 1 ), the probability of being in state j is written as \displaystyle P(w_{t}=j)=\frac{e^{-A}A^{j}/j!}{\sum_{i\in\mathcal{W}}e^{-A}A^{i}/i!}, (2) where A indicates the impulsive index; as A increases, the amount of noise in the IN states ( j>0 ) increases. The noise variance \sigma^{2}_{w_{j}} at a specific state w_{t}=j is defined as \displaystyle\sigma^{2}_{w_{j}}=\left(1+\frac{j\Lambda}{A}\right)\sigma^{2}_{w_{0}}, where \sigma^{2}_{w_{0}} is the background noise variance and \Lambda=\sigma_{I}^{2}/\sigma_{B}^{2} is the impulsive-to-background average noise power ratio. 1 1 1 In the original paper [ 7 ] , the power ratio is defined as \Gamma=\sigma_{B}^{2}/\sigma_{I}^{2} . To describe the burstiness (memory) of the noise observed in practice, [ 7 ] introduces a parameter r\in[0,1) to establish a correlation between consecutive noise samples. Specifically, the time-invariant state transition matrix of the Markov-Middleton noise model is constructed as \displaystyle P_{w_{ij}} \displaystyle=P(w_{t}=j|w_{t-1}=i) \displaystyle=\begin{cases}r+(1-r)\cdot P(w_{t}=j),\quad i=j\\ (1-r)\cdot P(w_{t}=j),\quad\text{otherwise}.\end{cases} (3) From ( 3 ), we can see that the larger the correlation parameter r , the more likely the following noise sample will stay in the same state. That is, the burst behavior is more prominent. In Figs. and , we show two-state ( W=2 ) Markov-Middleton noise realizations for two different parameter sets. Fig. has an impulsive state probability P(w_{t}=1)\approx 0.09 , whereas Fig. is featured

Read the complete text