Neural Network-based Decoder Research Articles

Performance Evaluation of Conventional and Neural Network-Based Decoder for an Audio of Low-Girth LDPC Code

Noise in a communication system degrades the signal level at the receiver, and as a result, the signal is not properly recovered or eliminated at the receiver side. To avoid this, it is necessary to modify the signal before transmission, which is achieved using channel coding. Channel coding provides an opportunity to recover the noisy signal at the receiver side. The low-density parity-check (LDPC) code is an example of a forward error correcting code. It offers near Shannon capacity approaching performance; however, there is a constraint regarding high-girth code design. When the low-girth LDPC code is decoded using conventional methods, an error floor can occur during iterative decoding. To address this issue, a neural network (NN)-based decoder is utilized to overcome the decoding problem associated with low-girth codes. In this work, a neural network-based decoder is developed to decode audio samples of both low- and high-girth LDPC codes. The neural network-based decoder demonstrates superior performance for low-girth codes in terms of bit error rate (BER), peak signal-to-noise-ratio (PSNR), and mean squared error (MSE) with just a single iteration. Audio samples sourced from the NOIZEUS corpus are employed to evaluate the designed neural network. Notably, when compared to a similar decoder, the decoder developed in this study exhibits an improved bit error rate for the same signal-to-noise ratio.

Deep Learning Algorithms for BCH Decoding in Satellite Communication

Deep learning is widely used in various fields due to the advancement of algorithms, the enrichment of high-efficiency databases, and the increase in computing power. Especially in the satellite communication, the learning and parallel computing capabilities of neural networks make them ideal for decoding. Many researchers have recently applied deep learning neural networks to decode high-density parity check (HDPC) codes (such as BCH and RS code), improving the decoder’s performance. This review aims to provide general insights on applying neural network decoders to satellite communications. Due to the neural network’s learning ability, the neural network-based decoder can be trained to change the weights, thereby reducing the influence of non-white noise in satellite communications, such as the influence between the satellite and the terrestrial network and the mutual interference within the satellites. To compensate for non-white noise, shortest circles in Tanner graph and unreliable information, a decoder system model for satellite communication constructed by three neural networks is presented.

NanoNET: machine learning platform for predicting nanoparticles distribution in a polymer matrix.

Polymer nanocomposites (PNCs) offer a broad range of thermophysical properties that are linked to their compositions. However, it is challenging to establish a universal composition-property relationship in PNCs due to their wide-ranging composition and chemical space. Here, we address this problem and develop a new method to model the composition-microstructure relation of a PNC through an intelligent machine-learning pipeline named nanoNET. The nanoNET is a nanoparticles (NPs) distribution predictor, built upon computer vision and image recognition concepts. It integrates unsupervised deep learning and regression in a fully automated pipeline. We conduct coarse-grained molecular dynamics simulations of PNCs and utilize the data to establish and validate the nanoNET. Within this framework, a random forest regression model predicts the distribution of NPs in a PNC in a latent space. Subsequently, a convolutional neural network-based decoder converts the latent space representation to the actual radial distribution function (RDF) of NPs in the given PNC. The nanoNET predicts NPs distribution in many unknown PNCs very accurately. This method is very generic and can accelerate the design, discovery, and fundamental understanding of composition-microstructure relationships in PNCs and other molecular systems.

Decoding of stimuli time series by neural activity patterns of recurrent neural network

The study is concerned with question whether it is possible to identify the specific sequence of input stimuli received by artificial neural network using its neural activity pattern. We used neural activity of simple recurrent neural network in course of “Even-Odd” game simulation. For identification of input sequences we applied the method of neural network-based decoding. Multilayer decoding neural network is required for this task. The accuracy of decoding appears up to 80%. Based on the results: 1) residual excitation levels of recurrent network’s neurons are important for stimuli time series processing, 2) trajectories of neural activity of recurrent networks while receiving a specific input stimuli sequence are complex cycles, we claim the presence of neural activity attractors even in extremely simple neural networks. This result suggests the fundamental role of attractor dynamics in reflexive processes.

A Parallel Turbo Decoder Based on Recurrent Neural Network

A neural network-based decoder, based on a long short-term memory (LSTM) network, is proposed to solve the problem that large decoding delay and performance degradation under non-Gaussian noise due to poor parallelism of existing turbo decoding algorithms. The proposed decoder refers to a unique component coding concept of turbo codes. First, each component decoder is designed based on an LSTM network. Next, each layer of the component decoder is trained, and the trained weights are loaded into the turbo code decoding neural network as initialization parameters. Then, the turbo code decoding network is trained end-to-end. Finally, a complete turbo decoder is realized. The structural advantage of turbo code component coding is fully considered in the design process, and the problem of decoding delay caused by the existence of interleaver is cleverly avoided. The introduction of deep learning technology provides a new idea to solve the traditional communication problems. Simulation results show that the performance of the proposed decoder is improved by 0.5–1.5 dB compared with the traditional serial decoding algorithm in Gaussian white noise and t-distribution noise. When BER performance is close, the LSTM decoder requires half or even less than that of BCJR. Moreover, the results demonstrate that the proposed decoder is adaptive and can be applied to communication systems with various turbo codes. The LSTM decoder shows lower bit error rate, computational complexity and higher decoding efficiency under the same conditions. Therefore, it is necessary to study the turbo code decoding technology based on deep learning combined with the actual channel environment.

Decoding of Raman spectroscopy-encoded suspension arrays based on the detail constraint cycle domain adaptive model

Previous studies have already shown that Raman spectroscopy can be used in the encoding of suspension array technology. However, almost all existing convolutional neural network-based decoding approaches rely on supervision with ground truth, and may not be well generalized to unseen datasets, which were collected under different experimental conditions, applying with the same coded material. In this study, we propose an improved model based on CyCADA, named as Detail constraint Cycle Domain Adaptive Model (DCDA). DCDA implements the classification of unseen datasets through domain adaptation, adapts representations at the encode level with decoder-share, and enforces coding features while leveraging a feat loss. To improve detailed structural constraints, DCDA takes downsample connection and skips connection. Our model improves the poor generalization of existing models and saves the cost of the labeling process for unseen target datasets. Compared with other models, extensive experiments and ablation studies show the superiority of DCDA in terms of classification stability and generalization. The model proposed by the research achieves a classification with an accuracy of 100% when applied in datasets, in which the spectrum in the source domain is far less than the target domain.

Neural Network-Based Decoding Input Stimulus Data Based on Recurrent Neural Network Neural Activity Pattern

The paper reports the assessment of the possibility to recover information obtained using an artificial neural network via inspecting neural activity patterns. A simple recurrent neural network forms dynamic excitation patterns for storing data on input stimulus in the course of the advanced delayed match to sample test with varying duration of pause between the received stimuli. Information stored in these patterns can be used by the neural network at any moment within the specified interval (three to six clock cycles), whereby it appears possible to detect invariant representation of received stimulus. To identify these representations, the neural network-based decoding method that shows 100% efficiency of received stimuli recognition has been suggested. This method allows for identification the minimum subset of neurons, the excitation pattern of which contains comprehensive information about the stimulus received by the neural network.

GP-Aligner: Unsupervised Groupwise Nonrigid Point Set Registration Based on Optimizable Group Latent Descriptor

In this paper, we propose a novel unsupervised method named GP-Aligner to address the problem of groupwise non-rigid point set registration. Compared to previous non-learning-based approaches, the proposed method gains competitive advantages by leveraging deep neural networks to effectively and efficiently align a large number of highly deformed 3D shapes with superior performance. Unlike most learning-based methods that use an explicit feature encoding network to extract per-shape features and their correlations, our model leverages a model-free learnable latent descriptor to characterize shape correlations among groups. More specifically, for a given group we first define an optimizable Group Latent Descriptor (GLD) to characterize the relationship among a group of point sets. Each GLD is randomly initialized from a Gaussian distribution and then concatenated with the coordinates of each point of the associated point sets in the group. A neural network-based decoder network is further constructed to predict the coherent flow fields to optimally deform the input groups of shapes to the aligned ones. During the optimization process, GP-Aligner jointly updates all GLDs and weight parameters of the decoder network towards the minimization of an unsupervised groupwise alignment loss. After optimization, for each group, our model coherently drives each point set towards a mean position (shape) without specifying one as the target. GP-Aligner does not require large-scale training data for network training, and it can directly align groups of point sets in a one-stage optimization process. GP-Aligner shows both accuracy and computational efficiency improvement in comparison with the state-of-the-art methods for groupwise point set registration. Moreover, GP-Aligner exhibits high efficiency in aligning a large number of groups of real-world 3D shapes.

Recognition of stimulus received by recurrent neural network

The study is concerned with the comparison of two methods for identification of stimulus received by artificial neural network using neural activity pattern that corresponds to the period of storing information about this stimulus in the working memory. We used simple recurrent neural networks learned to pass the delayed matching-to-sample test. Neural activity was detected at the period of pause between receiving stimuli. The analysis of neural excitation patterns showed that neural networks encoded variables that were relevant for the task during the delayed matching-to-sample test, and their activity patterns were dynamic. The method of centroids allowed identifying the type of the received stimuli with efficiency up to 75% while the method of neural network-based decoder showed 100% efficiency. In addition, this method was applied to determine the minimal set of neurons whose activity was the most significant for stimulus recognition.

Deep Gated Recurrent Unit for Smartphone-Based Image Captioning

Expressing the visual content of an image in natural language form has gained relevance due to technological and algorithmic advances together with improved computational processing capacity. Many smartphone applications for image captioning have been developed recently as built-in cameras provide advantages of easy-operation and portability, resulting in capturing an image whenever or wherever needed. Here, an encoder-decoder framework based new image captioning approach with a multi-layer gated recurrent unit is proposed. The Inception-v3 convolutional neural network is employed in the encoder due to its capability of more feature extraction from small regions. The proposed recurrent neural network-based decoder utilizes these features in the multi-layer gated recurrent unit to produce a natural language expression word-by-word. Experimental evaluations on the MSCOCO dataset demonstrate that our proposed approach has the advantage over existing approaches consistently across different evaluation metrics. With the integration of the proposed approach to our custom-designed Android application, named “VirtualEye+”, it has great potential to implement image captioning in daily routine.

Improved NSC decoding algorithm for polar codes based on multi-in-one neural network

Neural network-based decoding algorithms have potential value to be researched in the field of polar codes due to their low decoding latency. The neural successive cancellation (NSC) algorithm, combining deep learning and successive cancellation (SC) algorithm of polar codes, was proposed to reduce the latency of decoding. In terms of overall latency, the NSC algorithm does not fully consider the parallel decoding of special nodes in SC decoding tree, which limits the reduction of system delay to a certain extent. In this paper, we propose a multi-in-one neural simplified successive cancellation (MIO-NSSC) decoding algorithm for polar codes based on deep learning. The proposed MIO-NSSC algorithm, which is suitable for general nodes, mainly improves the existing fast simplified successive cancellation (FSSC) and the NSC algorithms to obtain a multi-in-one neural network instead of multiple neural networks in the NSC algorithm by using a new training strategy. Through applying the FSSC algorithm to a special node, the decoding delay of the proposed algorithm is further reduced. The experimental results demonstrate that the proposed MIO-NSSC algorithm can achieve significant latency reduction and resource consumption efficiency improvement compared with the NSC algorithm. The latency of the proposed MIO-NSSC decoding algorithm is about 21% lower than that of the NSC algorithm, and approximately seven neural networks are saved compared with the NSC algorithm. Furthermore, the MIO-NSSC algorithm can reduce the computational complexity without loss of performance.

A passive and low-complexity Compressed Sensing architecture based on a charge-redistribution SAR ADC

An innovative analog-to-digital converter (ADC) architecture is proposed, with the aim of acquiring an input signal according to the Compressed Sensing (CS) paradigm and without the need for dedicated active analog blocks. Its core is the capacitive array employed in traditional successive-approximation-register (SAR) ADCs. Introducing only a few additional switches, the array can compute the linear combination of consecutive signal samples, as required by the CS encoding.To manage the presence of leakage currents, which may impair signal reconstruction, a compensation circuit is considered, allowing close-to-ideal performance of the system when properly designed. A neural network-based decoding strategy is also analyzed, with up to 20 dB of additional reconstruction quality with respect to standard algorithms. Synthetic electrocardiogram signals are used to validate optimizations both at the hardware level in the encoding block and at the software level in the decoder.