Abstract Inter‐cell interference and smart jammer attacks significantly impair the performance of non‐orthogonal multiple access (NOMA) networks. This issue is particularly critical when considering strategic interactions with malicious actors. To address this challenge, the power allocation problem is framed in a two‐cell NOMA network as a sequential game. In this game, each base station acts as a leader, choosing a power allocation strategy, while the smart jammer acts as a follower, reacting optimally to the base stations' choices. To address this multi‐agent scenario, four multi‐agent reinforcement learning algorithms are proposed: Q‐learning based unselfish (QLU), deep QLU, hot booting deep QLU, and decreased state deep QLU. A game‐theoretic analysis that demonstrates the algorithms' convergence to the optimal network‐wide strategy with high probability is provided. Simulation results further confirm the superiority of our proposed algorithms compared to the Q‐learning‐based selfish NOMA power allocation method.

ABSTRACT In this study, an innovative architecture is proposed to enhance the performance of semantic communication networks by leveraging deep learning and joint source‐channel coding. A fundamental challenge in this field is the strong dependence of conventional networks on a fixed signal‐to‐noise ratio (SNR) during training, which leads to performance degradation under varying channel conditions. To address this limitation, we introduce a novel attention‐based approach that enables dynamic adaptation to different SNR levels, ensuring more stable and optimized communication performance. The proposed model learns more generalized features that exhibit greater resilience to channel variations. To evaluate its effectiveness, extensive simulations were conducted, comparing the performance of the proposed architecture with DeepSC, a state‐of‐the‐art benchmark model in the field. While the baseline model, trained at a single SNR, experiences performance drops under mismatched conditions, the proposed model, trained across a range of SNRs, achieves improvement of 16.2%, 30.8%, 42.8%, and 53.8% for 1, 2, 3, and 4‐gram precisions, respectively, in bilingual evaluation understudy score and an 11.4% increase in sentence similarity across challenging low‐SNR conditions. Furthermore, the model maintains robust performance with 48% less training data, highlighting its efficiency and data efficiency under practical constraints. These gains confirm the model's superior adaptability and high‐quality data reconstruction under diverse conditions. The results of this study underscore the significant benefits of attention‐based architectures in semantic communication, particularly in environments with unpredictable channel variations, and highlight their potential for reliable deployment in real‐world applications.

Abstract Isomap is a well‐known nonlinear dimensionality reduction method that highly suffers from computational complexity. Its computational complexity mainly arises from two stages; a) embedding a full graph on the data in the ambient space, and b) a complete eigenvalue decomposition. Although the reduction of the computational complexity of the graphing stage has been investigated by graph processing methods, the eigenvalue decomposition stage remains a bottleneck in the problem. In this paper, we propose the Low‐Rank Isomap (LRI) algorithm by introducing a projection operator on the embedded graph from the ambient space to a low‐rank latent space to facilitate applying the partial eigenvalue decomposition. This approach leads to reducing the complexity of Isomap to a linear order while preserving the structural information during the dimensionality reduction process as long as the number of observations remains extensively larger than the dimensionality of the ambient space. The superiority of the LRI algorithm compared to some state‐of‐art algorithms is experimentally verified on facial image clustering in terms of speed and accuracy.

Low-Rank Representation (LRR) suffers from discarding the locality information of data points in subspace clustering. We propose a hypergraph-based model by incorporation of a variable number of adjacent nodes, locality information, and sparsity of subspaces. An optimization problem is defined and solved by developing a tensor Laplacian-based algorithm.The outperformance of the proposed method is substantial when the inherent structure of the data exploits severe nonlinearity, geometrical overlapping, and outliers.

BACKGROUND: Haplotype information is essential for many genetic and genomic analyses, including genotype-phenotype associations in human, animals and plants. Haplotype assembly is a method for reconstructing haplotypes from DNA sequencing reads. By the advent of new sequencing technologies, new algorithms are needed to ensure long and accurate haplotypes. While a few linked-read haplotype assembly algorithms are available for diploid genomes, to the best of our knowledge, no algorithms have yet been proposed for polyploids specifically exploiting linked reads. RESULTS: The first haplotyping algorithm designed for linked reads generated from a polyploid genome is presented, built on a typical short-read haplotyping method, SDhaP. Using the input aligned reads and called variants, the haplotype-relevant information is extracted. Next, reads with the same barcodes are combined to produce molecule-specific fragments. Then, these fragments are clustered into strongly connected components which are then used as input of a haplotype assembly core in order to estimate accurate and long haplotypes. CONCLUSIONS: Hap10 is a novel algorithm for haplotype assembly of polyploid genomes using linked reads. The performance of the algorithms is evaluated in a number of simulation scenarios and its applicability is demonstrated on a real dataset of sweet potato.

A new estimator for sparse time dispersive channels in pilot aided orthogonal frequency division multiplexing (OFDM) systems is developed by considering prior knowledge on channel time dispersions. The authors propose a weighted atomic norm minimisation (WANM) in order to incorporate the prior information into the estimator. The dual of the WANM is then converted to a tractable semidefinite programming using positive trigonometric polynomial theory. After solving the dual problem, the channel response is identified by solving a least squares approach. In this work, they assume that time dispersions associated delays can take any value with a mild minimum separation condition on the normalised interval . The performance of the new estimator is compared with conventional approaches. With respect to the pilot number and signal to noise ratio (SNR), simulation results reveal that the proposed estimator performs superior to that of traditional methods. It is shown that both a lower SNR and number of pilots are required to achieve the same mean square error reported in previous works.

The authors solve the matrix completion (MC) problem based on manifold optimisation by incorporating the side information under which the columns of the intended matrix are drawn from a union of low‐dimensional subspaces. It is proved that this side information leads us to construct new manifolds, as an embedded sub‐manifold of the manifold of constant rank matrices, using which the MC problem is solved more accurately. The required geometrical properties of the aforementioned manifold are then presented for MC. Simulation results over both synthetic and real‐world data show that the proposed method outperforms some recent techniques either based on side information or not.

The problem of blind super-resolution of sparse signals using arbitrary sampling and the atomic lift is studied. Using the Prolate Spheroidal Wave Functions (PSWFs) and matrix atomic norm, a new Semi-Definite Program (SDP) is proposed. Unlike the previous results, the proposed new version of SDP can localize spikes with higher precision and no need for the magnitude recovery and integer sampling schemes. Numerical simulations show that our method outperforms the recent techniques with generating a lower Normalized Mean Square Error (NMSE).

In this study, deterministic construction of measurement matrices in compressive sensing is considered. First, by employing the column replacement concept, a theorem for construction of large minimum distance linear codes containing all‐one codewords is proposed. Then, by applying an existing theorem over these linear codes, deterministic sensing matrices are constructed. To evaluate this procedure, two examples of constructed sensing matrices are presented. The first example contains a matrix of size and coherence , and the second one comprises a matrix with the size and coherence , where p is a prime integer. Based on the Welch bound, both examples asymptotically achieve optimal results. Moreover, by presenting a new theorem, the column replacement is used for resizing any sensing matrix to a greater‐size sensing matrix whose coherence is calculated. Then, using an example, the outperformance of the proposed method is compared to a well‐known method. Simulation results show the satisfying performance of the column replacement method either in created or resized sensing matrices.

Recent matrix completion based methods have not been able to properly model the haplotype assembly problem for noisy observations. To deal with such cases, we propose a new minimum error correction (MEC) based matrix completion problem over the manifold of rank-one matrices. We then prove the convergence of a specific iterative algorithm to solve this problem. From the simulation results, the proposed method not only outperforms some well-known matrix completion based methods, but also shows a more accurate result compared to a most recent MEC-based algorithm for haplotype estimation.

The tree‐structured complex wavelet Bayesian compressing sensing (TSCW‐BCS) is introduced. The Bessel K form (BKF) probability density function; which has heavy tails out of the origin, is used as the prior. The inter‐scale statistical relation between complex wavelet coefficients is modelled by the hidden Markov tree. The Markov chain Monte Carlo inference is obtained based on the BKF and then the posterior parameters of wavelet coefficients are derived. Simulation results show that the proposed TSCW‐BCS outperforms many well‐known CS methods.

The authors study robust beamforming and power allocation in cognitive relay networks using several relays in the secondary link. The channel state information is imperfect modelled by Gaussian random variables. First, minimisation of the total transmit power of relays and the secondary transmitter is considered with a constraint on the interference in the primary receiver. Also, a constraint is assumed on the received signal‐to‐interference plus noise ratio (SINR) in the secondary receiver. In the other scenario, maximisation of the SINR in the secondary receiver is investigated with a limit on the maximum transmit power of relays and the secondary transmitter. For each non‐convex optimisation problem, an iterative and robust algorithm is derived. It is proved that the first proposed algorithm converges to the global optimum. Simulation results similarly show the convergence of both algorithms with a few number of iterations. The outperformance of the robust algorithms is shown compared with a non‐robust design.

In this study, cross-talk cancellation in convolutive blind source separation (BSS) is investigated using a new post-processing algorithm. In this algorithm, exclusive activity periods (EAPs) are used to model cross-talks in which only one BSS output signal is assumed to be active. Inactive intervals of each EAP are used to estimate the cross-talk leaked from each active signal. Then, the estimated cross-talk is subtracted from BSS outputs. Simulation results show that the proposed algorithm successfully suppresses the crosstalk by improving the signal-to-interference ratio (SIR) and signal-to-distortion ratio (SDR) of BSS outputs about 12 and 5 dB, respectively.

In this paper, a new joint structure for noise reduction in a reverberant environment will be proposed. The proposed structure consists of an acoustic echo canceller (AEC) followed by a noise reduction stage like generalized side-lobe canceller (GSC). This configuration is called AEC-GSC. It improves noise cancellation of the GSC beamformer in the presence of acoustic echoes in highly reverberant environments where GSC-alone fails to work properly. The AEC section is accomplished by our recently proposed segment variable step-size proportionate normalized least mean squares (SVS-PNIMS) algorithm. The proposed AEC-GSC algorithm is evaluated through both computer simulations and experimental results. The results demonstrate that the proposed AEC-GSC algorithm performs significantly better than GSC-alone in terms of speech distortion parameters and resulting ERLE.

Higher-order statistics (HOS) has been proposed as a solution to blind separation of an instantaneous mixture of sources with the same PSDs. However, in this paper, we'll introduce a new method based on the first and second order of conditional statistics and modifying it for noisy sensor conditions. Comparing the performance of our newly proposed algorithm with that of the previous ones, it is demonstrated that our newly proposed algorithm results in a better result.

The dual-mode CMA-assisted decision adjusted modulus algorithm (CADAMA) has been proved to be a very effective way of equalizing non-minimum phase channels blindly [Axford RA, et al., 1996]. However, it still suffers from a slow convergence when applies to the higher order QAMS. In this paper, we propose a new dual mode method for blind equalization of QAM signals, which achieves a faster convergence compared to the CADAMA. The simulation results show the better convergence rate of the proposed algorithm.