This paper introduces a novel cost-aware structure learning framework for optimizing distributed algorithms, balancing computational performance and aggregation costs. We demonstrate its effectiveness through application to multiple measurement vector compressed sensing (MMV-CS) problems. Our proposed Learned Distributed Multiple Measurement Vector Iterative Shrinkage Thresholding (LDM-IST) algorithm extends a single measurement vector distributed recovery algorithm to the MMV model, incorporating deep unfolding to optimize hyperparameters and enhance convergence. Moreover, cost-aware structure learning is applied to improve the aggregation efficiency of the LDM-IST. Numerical experiments show that LDM-IST achieves normalized mean square error performance comparable to the centralized recovery algorithm, and takes a good tradeoff between recovery performance and the in-network computing overhead.

This paper introduces an extension of Gradient Flow (GF) decoding for LDPC codes. GF decoding, a continuous-time methodology based on gradient flow, employs a potential energy function associated with bipolar codewords of LDPC codes. The original GF decoding was designed for AWGN channels but in this paper, we introduce the negative log-likelihood function of the channel for generalizing the original method. The proposed method is shown to be tensor-computable, which means that the gradient of the objective function can be evaluated with the combination of basic tensor computations. This characteristic is well-suited to emerging AI accelerators, potentially applicable in wireless signal processing. The paper assesses the decoding performance of the generalized GF decoding in LDPC-coded MIMO channels. Our numerical experiments reveal that this method's decoding performance rivals that of established techniques such as MMSE + BP. A further benefit of the proposed method is its suitability for deep unfolding. This advantage stems from the fact that each component of the method is differentiable.

This paper introduces a deep unfolding-assisted proximal decoding for low-density parity-check (LDPC) codes. Proximal decoding method is a decoding algorithm based on the proximal gradient method. Our proposal, deep unfolding-assisted proximal (DU-Proximal) decoding, is obtained by applying deep unfolding to the proximal decoding. We especially focus on the decoding performance in multiple-input multiple-output (MIMO) channels. In numerical experiments, we compare the error correcting capability of the proposed algorithm with the minimum mean squared error (MMSE) signal detection method, the tanh signal detection method, and the MMSE jointly with belief propagation (BP) algorithm for LDPC decoding. The experimental results demonstrate that parameter optimization via DU yields significant performance gains, especially at high signal-to-noise ratio levels.

The delay of rendering on AR devices requires prediction of head motion using sensor data acquired tens of even one hundred milliseconds ago to avoid misalignment between the virtual content and the physical world, where the misalignment will lead to a sense of time latency and dizziness for users. To solve the problem, we propose a method for the 6DoF motion prediction to compensate for the time latency. Compared with traditional hand-crafted methods, our method is based on deep learning, which has better motion prediction ability to deal with complex human motion. In particular, we propose a MOtion UNcerTainty encode decode network (MOUNT) that estimates the uncertainty of input data and predicts the uncertainty of output motion to improve the prediction accuracy and smoothness. Experiments on the EuRoC and our collected dataset demonstrate that our method significantly outperforms the traditional method and greatly improves AR visual effects.

User identification (UI) and channel estimation (CE) are essential in wireless networks with numerous users. Signature-code-based UI and CE schemes are widely used owing to their high spectral efficiency. Traditional signature code uses a discrete sensing matrix as a dictionary to generate codewords. Subsequently, the sparse vector recovery algorithm is used to recover the user state and channel state information in the received signal to complete the UI and CE. In this study, we propose an end-to-end machine-learning-aided signature code scheme under a multiple-access Rayleigh fading channel called machine-learning signature code (ML-SC). The ML-SC consists of a binarized-neural-networks-based (BNNs-based) trainable encoder and a trainable-iterative-soft-threshold-algorithm-based (TISTA-based) trainable decoder. To improve the accuracy, the dictionary is optimized by minimizing the mean squared error between the original and recovered information. Our proposed scheme achieved better performance and efficiency than the conventional schemes in the simulation. Moreover, it was confirmed that the dictionary generated by the ML-SC is suitable for various conventional decoders. Finally, by analyzing the results of the simulations, we found that ML-SC improves the restricted isometric constants and coherence of the dictionary.

The user identification scheme for a multiple-access fading channel based on the binary signature code is considered. In previous works, the signature code was used over a noisy multiple-access adder channel, and only the status of uses was decoded by the signature decoder. In this study, by considering the communication model as a compressed sensing process, it is possible to estimate the channel coefficients while identifying users. To improve the efficiency of the decoding process, we proposed an iterative deep neural network (DNN)-based decoder. Our simulation results show that for the binary signature code, our proposed DNN-based decoder requires less computing time to achieve higher active user detection accuracy and channel estimation accuracy than the classical signal recovery algorithm used in compressed sensing.