ReRAM crossbar-based computing-in-memory (CIM) systems offer computational efficiency but suffer from significant accuracy degradation under stuck-at faults (SAFs). Conventional approaches like retraining-based methods fail to effectively generalize across diverse SAFs ratios. To address this challenge, we propose the Mask-based Meta Learning (MML) framework, leveraging meta-learning’s multi-task generalization capability to achieve robust performance of varying SAFs scenarios. Within the MML framework, a SAFs mask-based task formulation is used to create meta-learning tasks based on SAFs masks rather than dataset. Second, we define a new meta-learning objective by integrating different SAFs masks into the meta loss. Finally, we develop a SAFs-sensitivity guided weight importance search algorithm and dynamically expands the adjustment range of crucial weights using the ReRAM array’s redundant cells to further enhance model performance. Experimental results show that our method demonstrates superior robustness and generalization performance over the state-of-the-art robust training approaches. Moreover, our method can maintain high accuracy across various SAFs ratios while the accuracy of other approaches illustrates large fluctuations.

Resistive random-access memory (ReRAM) is a promising non-volatile memory technology. However, its crossbar array structure leads to a severe problem known as sneak path interference (SPI), which is correlated and data-dependent. From an information-theoretic perspective, memory systems like ReRAM can be considered as special types of communication channels. Inspired by deep learning applications in communication systems, the detection of ReRAM channels with SPI was formulated as a learning problem recently, and a multi-layer perceptron (MLP) network was employed to mitigate SPI. However, it requires a large amount of training data to achieve satisfactory performance. In this paper, we first propose a bidirectional long short-term memory (BiLSTM) based detector for ReRAM to exploit the correlation between memory cells introduced by SPI. Moreover, a few-shot learning algorithm based on meta-transfer learning (MTL) is proposed to further improve the generalization ability of the detector. The bit error rate (BER) bound and generalization bound are also derived to verify the effectiveness of our proposed schemes. Simulation results demonstrate that the BiLSTM-based detector with MTL can dramatically reduce the required training samples by four to five orders of magnitude while improving the BER performance compared to the existing MLP-based detection scheme.

Intelligent fault diagnosis (IFD) plays a crucial role in ensuring the safe operation of industrial machinery and improving production efficiency. However, traditional supervised deep learning methods require a large amount of training data and labels, which are often located in different clients. Additionally, the cost of data labeling is high, making labels difficult to acquire. Meanwhile, differences in data distribution among clients may also hinder the model’s performance. To tackle these challenges, this paper proposes a semi-supervised federated learning framework, SSFL-DCSL, which integrates dual contrastive loss and soft labeling to address data and label scarcity for distributed clients with few labeled samples while safeguarding user privacy. It enables representation learning using unlabeled data on the client side and facilitates joint learning among clients through prototypes, thereby achieving mutual knowledge sharing and preventing local model divergence. Specifically, first, a sample weighting function based on the Laplace distribution is designed to alleviate bias caused by low confidence in pseudo labels during the semi-supervised training process. Second, a dual contrastive loss is introduced to mitigate model divergence caused by different data distributions, comprising local contrastive loss and global contrastive loss. Third, local prototypes are aggregated on the server with weighted averaging and updated with momentum to share knowledge among clients. To evaluate the proposed SSFL-DCSL framework, experiments are conducted on two publicly available datasets and a dataset collected on motors from the factory. In the most challenging task, where only 10% of the data are labeled, the proposed SSFL-DCSL can improve accuracy by 1.15% to 7.85% over state-of-the-art methods.

Accurate modeling and estimation of the threshold voltages of the flash memory can facilitate the efficient design of channel codes and detectors. However, most flash memory channel models are based on Gaussian distributions, which fail to capture certain key properties of the threshold voltages, such as their heavy-tails. To enhance the model accuracy, we first propose a piecewise student's t-distribution mixture model (PSTMM), which features degrees of freedom to control the left and right tails of the voltage distributions. We further propose an PSTMM based expectation maximization (PSTMM-EM) algorithm to estimate model parameters for flash memories by alternately computing the expected values of the missing data and maximizing the likelihood function with respect to the model parameters. Simulation results demonstrate that our proposed algorithm exhibits superior stability and can effectively extend the flash memory lifespan by 1700 program/erase (PE) cycles compared with the existing parameter estimation algorithms.

Nowadays, covert communication is envisioned as a promising and secure method of delivering private information. However, higher bit error rates, limited data rates, and vulnerability to advanced machine learning detection methods significantly challenge the application of covert communication. In this paper, we propose a multiple carrier index keying orthogonal frequency division multiplexing (MCIK-OFDM) based covert communication system aided by a chaotic modulation scheme to improve covert data rate and covertness. First, we propose a covert information embedding method by dynamically selecting the activation or deactivation of a subcarrier to embed covert bits according to a previously negotiated covert key between the transmitter and receiver. Then, the chaotic modulation scheme is developed to mask transmitted signals with generated chaotic signals. Moreover, we propose an adversarial machine learning-based (AML) perturbation algorithm to resist the eavesdropper’s detection of covert signals. Furthermore, the closed-form bit error rate (BER) and the achievable covert rate of the proposed covert communication system are derived. Numerical and simulation results demonstrate that the BER of the proposed MCIK-OFDM-based hybrid chaotic covert communication system is much lower than that of conventional chaotic communication systems. In addition, the proposed AML perturbation algorithm can more effectively protect covert communication from being detected by supervised and unsupervised machine learning methods compared to traditional algorithms.

With the rapid development of cutting-edge technologies such as software-defined networking, edge computing, and deep learning (DL), the application of the field of Industrial Internet of Things (IIoT) has been deepening, especially in the areas of fault diagnosis, defect detection, and production management, which has shown great potential. Federated learning (FL) is a collaborative model training approach that allows multiple clients to work together while maintaining data privacy. This method is particularly useful for DL methods in industrial surface defect classification, which often require a large amount of training data that can be hard to gather due to its distributed nature across various sources. However, the aggregated model in federated learning may not perform well when there is a discrepancy between the training dataset (source domain) and the testing dataset (target domain), as well as when individual users face data scarcity. To counter these challenges, we propose a novel federated transfer learning framework with multi-scale aggregation (FTL-MSA) for surface defect classification in the IIoT system. A dynamic central loss function, which takes into account both intra-instance and inter-instance contrasting, is proposed to enhance the model's accuracy. Furthermore, we introduce a multi-scale model aggregation technique for FL. This technique considers the distances between the source domain and the target domain at multiple scales, which utilizes the Jensen-Shannon distance for statistical consistency, and the cosine distance for directional consistency, thereby effectively mitigating the impacts of domain differences. Empirical validation on two public steel defect datasets shows that our FTL-MSA framework outperforms state-of-the-art methods, achieving accuracy improvements of 3.12%-12.51%.

In this article, we explore the application of hybrid automatic repeat request (HARQ) within the intelligent reflection surface-assisted nonorthogonal multiple access (IRS-NOMA) system. We investigate the closed-form expressions for the outage probability of multiple users in the HARQ-assisted IRS-NOMA system, considering scenarios with perfect successive interference cancellation (pSIC) and imperfect successive interference cancellation (ipSIC), respectively. A definite integral approximation method for multidimensional functions based on Gauss-Chebyshev quadrature (GCQ) is proposed to conduct the performance analysis of the proposed HARQ-assisted IRS-NOMA system. Based on the asymptotic outage probability, the diversity order of multiple users for HARQ-assisted IRS-NOMA is obtained. Under the analytical results, the diversity order of the mth$(m\gt 1)$user for the HARQ-assisted IRS-NOMA with ipSIC is zero, and that of the mth user for the HARQ-assisted IRS-NOMA with pSIC is in connection with the number of reflecting elements and the number of transmission rounds. The simulation results are presented to substantiate the accuracy of the analytical results. The results demonstrate that: 1) the HARQ-assisted IRS-NOMA systems can achieve a significant gain compared with the IRS-NOMA systems; 2) the HARQ-assisted IRS-NOMA outperforms the HARQ-assisted IRS-orthogonal multiple access in terms of outage probability and diversity order; and 3) the HARQ with incremental redundancy (HARQ-IR)-assisted IRS-NOMA system has a better performance than the HARQ with chase combining (HARQ-CC)-assisted IRS-NOMA system.

The NAND flash memory channel is corrupted by different types of noises, such as the data retention noise and the wear-out noise, which lead to unknown channel offset and make the flash memory channel non-stationary. In the literature, machine learning-based methods have been proposed for data detection for flash memory channels. However, these methods require a large number of training samples and labels to achieve a satisfactory performance, which is costly. Furthermore, with a large unknown channel offset, it may be impossible to obtain enough correct labels. In this paper, we reformulate the data detection for the flash memory channel as a transfer learning (TL) problem. We then propose a model-based deep TL (DTL) algorithm for flash memory channel detection. It can effectively reduce the training data size from 106samples to less than 104samples. Moreover, we propose an unsupervised domain adaptation (UDA)-based DTL algorithm using moment alignment, which can detect data without any labels. Hence, it is suitable for scenarios where the decoding of error-correcting code fails and no labels can be obtained. Finally, a UDA-based threshold detector is proposed to eliminate the need for a neural network. Both the channel raw error rate analysis and simulation results demonstrate that the proposed DTL-based detection schemes can achieve near-optimal bit error rate (BER) performance with much less training data and/or without using any labels.

With the rapid proliferation of smart mobile devices, federated learning (FL) has been widely considered for application in wireless networks for distributed model training. However, data heterogeneity, e.g., non-independently identically distributions and different sizes of training datasets among clients, poses major challenges to wireless FL. Limited communication resources complicate the implementation of fair scheduling which is required for training on heterogeneous data, and further deteriorate the overall performance. To address this issue, this paper focuses on performance analysis and optimization for wireless FL, considering data heterogeneity, combined with wireless resource allocation. Specifically, we first develop a closed-form expression for an upper bound on the FL loss function, with a particular emphasis on data heterogeneity described by a dataset size vector and a data divergence vector. Then we formulate the loss function minimization problem, under constraints on long-term energy consumption and latency, and jointly optimize client scheduling, uplink transmission power, channel allocation and the number of local epochs. Next, via the Lyapunov drift technique, we transform the optimization problem into a series of tractable problems. Extensive experiments on real-world datasets demonstrate that our method outperforms other benchmarks in terms of the learning accuracy and energy consumption.

Non-volatile memory (NVM) channels suffer from unknown offsets due to the presence of various impairments of the memory devices. Machine learning based methods have been proposed for data detection for NVMs under unknown channel offsets. However, the existing methods require a large number of training samples and labels to achieve a satisfactory data detection performance, which will result in large read latency and more power consumption. In this paper, we formulate a deep learning based data detection framework as a transfer learning problem. A deep transfer learning (DTL) based data detection scheme is proposed to reduce the number of required training samples and labels. The optimal symbol error rate is also derived as the performance benchmark by assuming that the perfect channel knowledge is known to the detector. Our experiment results demonstrate that the proposed DTL-based data detection scheme can achieve near-optimal performance with the training data size being reduced by two orders of magnitude compared with the original deep learning-based detector.

Federated self-supervised learning (FedSSL) is an emerging method in the domain of machine learning. It collaboratively learns a powerful feature extractor among multiple participants by utilizing distributed unlabeled data. However, conventional FedSSL suffers from statistical heterogeneity due to the non-independent and identically distributed (Non-IID) data among participants. In this work, we introduce a novel method to tackle the Non-IID data issue in FedSSL. First, the relation knowledge distillation is utilized to enhance the learning from global models. Then, we dynamically update the local model with divergence-aware update (DAU) method to preserve the client’s knowledge of Non-IID data. Our experimental results demonstrate that the proposed approach outperforms other methods by up to 8% on linear evaluation, verifying the effectiveness of our approach.

Abstract The error‐correcting performance of multi‐level‐cell (MLC) NAND flash memory is closely related to the block length of error‐correcting codes (ECCs) and log‐likelihood‐ratios of the read‐voltage thresholds. Driven by this issue, this paper optimizes the read‐voltage thresholds for MLC flash memory to improve the decoding performance of ECCs with finite block length. First, through the analysis of channel coding rate and decoding error probability under finite block length, the optimization problem of read‐voltage thresholds to minimize the maximum decoding error probability is formulated. Second, a cross‐iterative search algorithm to optimize read‐voltage thresholds under the perfect knowledge of flash memory channel is developed. However, it is challenging to analytically characterize the voltage distribution under the effect of data retention noise. To address this problem, a deep neural network (DNN)‐aided optimization strategy to optimize the read‐voltage thresholds is developed, where a multi‐layer perception network is employed to learn the relationship between voltage distribution and read‐voltage thresholds. Simulation results show that, compared with the existing schemes, the proposed DNN‐aided read‐voltage threshold optimization strategy with a well‐designed Low Density Parity Check (LDPC) code can not only improve the program‐and‐erase endurance but also reduce the read latency.

In this article, under a general cost function C, we present a dynamic programming (DP) method to obtain an optimal sequential deterministic quantizer (SDQ) for q-ary input discrete memoryless channel (DMC). The DP method has complexity O(q (N-M)2M), where N and M are the alphabet sizes of the DMC output and quantizer output, respectively. Then, starting from the quadrangle inequality, two techniques are applied to reduce the DP method's complexity. One technique makes use of the Shor-Moran-Aggarwal-Wilber-Klawe (SMAWK) algorithm and achieves complexity O(q (N-M) M). The other technique is much easier to be implemented and achieves complexity O(q (N2- M2)). We further derive a sufficient condition under which the optimal SDQ is optimal among all quantizers and the two techniques are applicable. This generalizes the results in the literature for binary-input DMC. Next, we show that the cost function of α-mutual information ( α-MI)-maximizing quantizer belongs to the category of C. We further prove that under a weaker condition than the sufficient condition we derived, the aforementioned two techniques are applicable to the design of α-MI-maximizing quantizer. Finally, we illustrate the particular application of our design method to practical pulse-amplitude modulation systems.

The protograph low‐density parity‐check (LDPC) codes possess many attractive properties, such as the low encoding/decoding complexity and better error floor performance, and hence have been successfully applied to different types of communication and data storage channels. In this study, the authors design protograph LDPC codes for communication systems corrupted by the impulsive noise, which are modelled as additive white symmetric alpha‐stable noise (AWS SN) channels. They start by presenting a novel simulation‐based protograph extrinsic information transfer analysis to derive the iterative decoding threshold of the protograph codes. By further applying the asymptotic weight distribution analysis, the authors design new protograph codes for the AWS SN channel. Both theoretical analysis and simulation results demonstrate that the proposed protograph codes can provide better error rate performance than the prior art AR4JA code, the irregular codes optimised for the AWGN channel, as well as the irregular codes optimised for the AWS SN channel.

The practical NAND flash memory suffers from various non-stationary noises that are difficult to be predicted. For example, the data retention noise induced channel offset is unknown during the readback process, and hence severely affects the reliability of data recovery from the memory cell. In this paper, we first propose a novel recurrent neural network (RNN)-based detector to effectively detect the data stored in the multi-level-cell (MLC) flash memory without the prior knowledge of the channel. However, compared with the conventional threshold detector, the proposed RNN detector introduces much longer read latency and more power consumption. To tackle this problem, we further propose an RNN-aided (RNNA) dynamic threshold detector, whose detection thresholds can be derived based on the outputs of the RNN detector. We thus only need to activate the RNN detector periodically when the system is idle. Moreover, to enable soft-decision decoding of error-correction codes, we first show how to obtain more read thresholds based on the hard-decision read thresholds derived from the RNN detector. We then propose integer-based reliability mappings based on the designed read thresholds, which can generate the soft information of the channel. Finally, we propose to apply density evolution (DE) combined with the differential evolution algorithm to optimize the read thresholds for low-density parity-check (LDPC) coded flash memory channels. Computer simulation results demonstrate the effectiveness of our proposed RNNA dynamic read thresholds design, for both the uncoded and LDPC-coded flash memory channels, without any prior knowledge of the channel.

A severe problem for mutual information-maximizing lookup table (MIM-LUT) decoding of low-density parity-check (LDPC) code is the high memory cost for using large tables, while decomposing large tables to small tables deteriorates decoding error performance. In this paper, we propose a systematic method, called mutual information- maximizing quantized belief propagation (MIM-QBP) decoding, to remove the lookup tables used for MIM-LUT decoding. Our method leads to a very practical decoder, namely the MIM-QBP decoder, which can be implemented based only on simple mappings and additions. Simulation results show that the proposed MIM-QBP decoder can outperform the state-of-the-art MIM-LUT decoder. Moreover, the MIM-QBP decoder with only 3 bits per message can outperform the floating-point belief propagation (BP) decoder at high signal-to-noise ratio (SNR) regions with a maximum of 10 iterations.

Mobile edge computing (MEC) enables feasible and scalable computation services for delay-sensitive and delay-tolerant tasks from mobile users. This paper considers an MEC network that consists of multiple users, multiple edge nodes (ENs), and a remote cloud computing server, where the ENs and the cloud server execute the delay-sensitive and delay-tolerant tasks respectively. First, we propose a unified linear network coded (NC) task offloading policy at the ENs to either execute the delay-sensitive tasks or assist the cloud server in the execution of delay-tolerant tasks. For the delay-sensitive task, the users jointly precode their task input messages according to a signal space alignment pattern, such that each EN can compute the linear NC messages for its intended user. For the delay-tolerant task, we put forth a compute-and-upload strategy for the ENs to upload their computed NC messages to the cloud server, such that the cloud server can first recover the input messages from all users and then execute the computation. Second, we develop the EN computation rules for different types of tasks. Finally, we characterize the computation load and normalized uploading time for the proposed task offloading. Our analytical results show that the proposed task offloading scheme is applicable to different NC computation with flexible requirements on computation load and uploading time.

The memory physics induced unknown offset of the channel is a critical and difficult issue to be tackled for many non-volatile memories (NVMs). In this paper, we first propose novel neural network (NN) detectors by using the multilayer perceptron (MLP) network and the recurrent neural network (RNN), which can effectively tackle the unknown offset of the channel. However, compared with the conventional threshold detector, the NN detectors will incur a significant delay of the read latency and more power consumption. Therefore, we further propose a novel dynamic threshold detector (DTD), whose detection threshold can be derived based on the outputs of the proposed NN detectors. In this way, the NN-based detection only needs to be invoked when the error correction code (ECC) decoder fails, or periodically when the system is in the idle state. Thereafter, the threshold detector will still be adopted by using the adjusted detection threshold derived base on the outputs of the NN detector, until a further adjustment of the detection threshold is needed. Simulation results demonstrate that the proposed DTD based on the RNN detection can achieve the error performance of the optimum detector, without the prior knowledge of the channel.

In this paper, we present a general framework of applying dynamic programming (DP) to the sequential deterministic quantization for discrete memoryless channels (DMCs) with pre-labelled outputs. The DP has complexity O(q(N -M)2M), where q, N, and M are alphabet sizes of the DMC input, DMC output, and the quantizer output, respectively. Then, starting from the quadrangle inequality (QI), we apply two techniques to reduce the DP's complexity. One technique makes use of the SMAWK algorithm with complexity O(q(N - M)M), while the other technique is much easier to be implemented and has complexity O(q(N2- M2)). Moreover, we give a sufficient condition on the channel transition probability, under which the two low-complexity techniques can be applied for designing quantizers that maximize the α-mutual information, which is a generalized objective function for channel quantization. This condition works for the general q-ary input case, including the previous work for q = 2 as a subcase.

Impulsive noises severely degrade the performance of a communication system. This study deals with the performance of the soft decision decoder for low‐density parity‐check codes over impulsive noise channels. To simplify the calculation of log likelihood ratio (LLR) and cooperate with the soft decision decoder, a new non‐linear approximation named inverse demapper of LLR over the impulsive noise is proposed. Without carrying the noise statistics, the inverse demapper performs close to the optimal demapper. In addition, the density evolution is employed to obtain optimal parameters of the inverse demapper. Then, the extrinsic information transfer chart analysis and simulation results are presented to verify the effectiveness of the authors' proposed demapper.

As emerging memories such as spin-torque transfer magnetic random access memory (STT-MRAM) suffer from reliability issues caused by process variations and thermal fluctuations, the design of channel quantizer with the minimum number of quantization bits is critical to support effective error correction coding for ensuring high-density and high-speed memory data storage. In this paper, we first propose a quantized channel model for STT-MRAM. Based on the quantized channel model, we derive various information theoretic bounds, including the mutual information, cutoff rate, and the Polyanskiy-Poor-Verdú (PPV) finite-length performance bound. By using these bounds as design criteria, we optimize the quantizer design for the polar-coded STT-MRAM channel. Moreover, we also propose a polar-code-specific quantization design with the successive cancellation decoding algorithm, by using the block error probability bound obtained from density evolution (DE). Simulation results show that all our proposed quantizers generally outperform the prior art greedy merging quantizer. In addition, both the cutoff rate and PPV bound based quantizers outperform the most widely applied mutual information based quantizer for short-length polar codes with 2-bit quantization. Furthermore, the DE quantizer designed specifically for polar codes achieves the best performance among all the proposed quantizers.

Channel output quantization plays a vital role in high-speed emerging memories such as the spin-torque transfer magnetic random access memory (STT-MRAM), where high-precision analog-to-digital converters (ADCs) are not applicable. In this paper, we investigate the design of the 1-bit quantizer which is highly suitable for practical applications. We first propose a quantized channel model for STT-MRAM. We then analyze various information theoretic bounds for the quantized channel, including the channel capacity, cutoff rate, and the Polyanskiy-Poor-Verdu ́(PPV) finite-length performance bound. By using these channel measurements as criteria, we design and optimize the 1-bit quantizer numerically for the STTMRAM channel. Simulation results show that the proposed quantizers significantly outperform the conventional minimum mean-squared error (MMSE) based Lloyd-Max quantizer, and can approach the performance of the 1-bit quantizer optimized by error rate simulations.

Spatial diversity is an effective method to mitigate the effects of fading, and when used in conjunction with low-density parity-check (LDPC) codes, it can achieve excellent error-correcting performance. Noise added at each branch of the diversity combiner is generally assumed to be additive white Gaussian noise, but there are many applications where the received signal is impaired by noise with a non-Gaussian distribution. In this paper, we derive the exact bit-error probability of different linear combining techniques on Rayleigh fading channels with impulsive noise, which is modeled using symmetric alpha-stable distributions. The relationship for the signal-to-noise ratios of these linear combiners is derived and then different non-linear detectors are presented. A detector based on the bi-parameter Cauchy-Gaussian mixture model is used and shows near-optimal performance with a significant reduction in complexity when compared with the optimal detector. Furthermore, the threshold signal-to-noise ratio of LDPC codes for different combining techniques on these channels is derived using density evolution and an estimation of the waterfall performance of LDPC codes is derived that reduces the gap between simulated and asymptotic performance.