Nam Yul Yu

While semantic communications has shown great potential in single-user settings, interference over a shared multiple access channel (MAC) remains a key challenge in multi-user scenarios. In this letter, we propose a novel joint source-channel coding scheme with interference-mitigation learning (JSCC-IM) for task-oriented multi-user semantic communications. The proposed JSCC-IM employs a single-layer decoder to separate desired semantic features from multi-user interference in the aggregated signals over MAC. Then, we design a loss function that explicitly suppresses multi-user interference while preserving desired semantic features. Simulation results show that the JSCC-IM improves inference accuracy over conventional JSCC schemes by more than 5% and 3% in multi-user semantic inference and multi-view semantic fusion, respectively.

In this letter, we propose a communication efficient federated learning algorithm, coined random FLARE (R-FLARE), using a novel error compensation method within a framework of random sparsification. In the R-FLARE, all devices sparsify the local gradients using a common set of randomly selected indices to improve communication efficiency with over-the-air computation. To upload local gradients, only the selected gradient elements are compensated by the local errors accumulated due to sparsification, which prevents redundant error compensation. We conduct a theoretical analysis on the convergence of R-FLARE using the $l_{2}$ norm-based error compensation, which shows that it achieves the same convergence rate as the state-of-the-art algorithms. Numerical results show that the R-FLARE using $l_{1}$- and $l_{2}$-norm based error compensations outperform conventional algorithms in test accuracy and training speed.

Reconfigurable intelligent surface (RIS) is a key enabler for 6G wireless networks, reconfiguring wireless propagation to enhance the quality of signal transmission. In RIS-aided communications, channel estimation plays a crucial role because the performance improvement from RIS relies on the accuracy of channel state information. In this paper, we configure the phase shifts of RIS passive reflective elements, in an effort to improve the performance of cascaded channel estimation between a base station and users through the RIS in multiuser mmWave communications. We first apply the Golay complementary sequences to configure the bipolar phase shifts of RIS in training phase. Then, we study a theoretical recovery guarantee for compressed sensing based channel estimation with the Golay-based bipolar phase shifts of RIS. Simulation results demonstrate that the Golay-based phase shift configuration exhibits excellent performance for channel estimation, outperforming random and optimized phase shifts. The algebraic configuration of phase shifts contributes to improving the performance of channel estimation efficiently through a large-scale RIS with bi-phase resolution in RIS-aided multiuser mmWave communications.

Grant-free access is a key enabler for connecting wireless devices with low latency and low signaling overhead in massive machine-type communications (mMTC). For massive grant-free access, user-specific signatures are uniquely assigned to mMTC devices. In this paper, we first derive a sufficient condition for the successful identification of active devices through maximum likelihood (ML) estimation in massive grant-free access. The condition is represented by the coherence of a signature sequence matrix containing the signatures of all devices. Then, we present a design framework of non-orthogonal signature sequences in a deterministic fashion. The design principle relies on unimodular masking sequences with low correlation, which are applied as masking sequences to the columns of the discrete Fourier transform (DFT) matrix. For example constructions, we use four polyphase masking sequences represented by characters over finite fields. Leveraging algebraic techniques, we show that the signature sequence matrix of proposed non-orthogonal sequences has theoretically bounded low coherence. Simulation results demonstrate that the deterministic non-orthogonal signatures achieve the excellent performance of joint activity and data detection by ML- and approximate message passing (AMP)-based algorithms for massive grant-free access in mMTC.

Grant-free access is a key enabler for connecting wireless devices with low latency and low signaling overhead in massive machine-type communications (mMTC). For massive grant-free access, user-specific signatures are uniquely assigned to mMTC devices. In this paper, we first derive a sufficient condition for the successful identification of active devices through maximum likelihood (ML) estimation in massive grant-free access. The condition is represented by the coherence of a signature sequence matrix containing the signatures of all devices. Then, we present a design framework of non-orthogonal signature sequences in a deterministic fashion. The design principle relies on unimodular masking sequences with low correlation, which are applied as masking sequences to the columns of the discrete Fourier transform (DFT) matrix. For example constructions, we use four polyphase masking sequences represented by characters over finite fields. Leveraging algebraic techniques, we show that the signature sequence matrix of proposed non-orthogonal sequences has theoretically bounded low coherence. Simulation results demonstrate that the deterministic non-orthogonal signatures achieve the excellent performance of joint activity and data detection by ML- and approximate message passing (AMP)-based algorithms for massive grant-free access in mMTC.

In edge intelligence, deep learning~(DL) models are deployed at an edge device and an edge server for data processing with low latency in the Internet of Things~(IoT). In this letter, we propose a new end-to-end learning-based wireless image recognition scheme using the PyramidNet in edge intelligence. We split the PyramidNet carefully into two parts for an IoT device and the edge server, which is to pursue low on-device computation. Also, we apply a squeeze-and-excitation block to the PyramidNet for the improvement of image recognition. In addition, we embed compression encoder and decoder at the splitting point, which reduces communication overhead by compressing the intermediate feature map. Simulation results demonstrate that the proposed scheme is superior to other DL-based schemes in image recognition, while presenting less on-device computation and fewer parameters with low communication overhead.

Frames are recognized for their importance in many fields of communications, signal processing, quantum physics, and so on. In this paper, we design an incoherent tight frame by selecting some rows of a matrix that is the Kronecker product of Fourier and unitary matrices. The Kronecker-product-based frame allows its elements to have a small number of phases, regardless of the frame length, which is suitable for low-cost implementation. To obtain the Kronecker-product-based frame with low mutual coherence, we first derive an objective function by transforming the Gram matrix expression to compute the coherence. If the Hadamard matrix is employed as a unitary matrix, the objective function can be computed efficiently with low complexity. Then, we find a subsampling index set for the Kronecker-product-based frame by minimizing the objective function. In simulations, we show that the Kronecker-product-based frames can achieve similar mutual coherence to optimized harmonic frames of a large number of phases. We apply the frames to compressed sensing (CS) as the measurement matrices, where the Kronecker-product-based frames demonstrate reliable performance of sparse signal recovery.

Federated learning (FL) technology has emerged for efficient data collection, data privacy protection, and efficient utilization of computing resources. In FL-based systems, data owners only send intermediate model trained with their local data to a central FL server, and then the central server combines intermediate models to generate globally trained model instead of the central server's collecting raw data from the data owners. However, even if only the trained intermediate model is transmitted to the central server, there are still several instances of privacy infringements using intermediate model information. To prevent data privacy infringement from intermediate data, homomorphic encryption (HE) schemes that enable numerical operations with ciphertexts can be used. The HE schemes have a limitation that data can only be calculated when it is encrypted with the same encryption key. The proposed algorithm improves data security by developing a technique that enables homomorphic operations even if data is encrypted with different encryption keys.

In this paper, we propose a novel loss function for deep face recognition, called the additive orthant loss (Orthant loss), which can be combined for softmax-based loss functions to improve the feature-discriminative capability. The Orthant loss makes features away from the origin using the rescaled softplus function and an additive margin. Additionally, the Orthant loss compresses feature spaces by mapping features to an orthant of each class using element-wise operation and 1-bit quantization. As a consequence, the Orthant loss improves the inter-class separabilty and the intra-class compactness. We empirically show that the ArcFace combined with the Orthant loss further compresses and moves the feature spaces farther away from the origin compared to the original ArcFace. Experimental results show that the new combined loss has the most improved accuracy on CFP-FP, AgeDB-30, and MegaFace testing datasets, among some of the latest loss functions.

In massive machine-type communications (mMTC), grant-free access is a key enabler for a massive number of users to be connected to a base station with low signaling overhead and low latency. In this letter, we present a set of Kronecker-based non-orthogonal sequences and study the performance of uplink grant-free access using the sequences. The user-specific, non-orthogonal, unimodular sequences of small phases are designed by subsampling the Kronecker product of Hadamard and Fourier matrices, where we use the genetic algorithm (GA) to find the subsampling index set. Simulation results demonstrate that the Kronecker-based non-orthogonal sequences exhibit excellent performance for compressed sensing based joint activity detection and channel estimation in uplink grant-free access. Consequently, the Kronecker-based sequences can be a good option for mMTC, thanks to the superior performance and the practical benefits of arbitrary sequence length and small phases.