We experimentally demonstrate a record net capacity per wavelength of 1.23~Tb/s over a single silicon-on-insulator (SOI) multimode waveguide for optical interconnects employing on-chip mode-division multiplexing and 11$\times$11 multiple-in-multiple-out (MIMO) digital signal processing.
Spectrum scarcity necessitates innovative, spectral-efficient strategies to meet the ever-growing demand for high data rates. Faster-than-Nyquist (FTN) signaling emerges as a compelling spectral-efficient transmission method that pushes transmit data symbols beyond the Nyquist limit, offering enhanced spectral efficiency (SE). While FTN signaling maintains SE with the same energy and bandwidth as the Nyquist signaling, it introduces increased complexity, particularly at higher modulation levels. This complexity predominantly arises from the detection process, which seeks to mitigate the intentional intersymbol interference generated by FTN signaling. Another challenge involves the generation of reliable log-likelihood ratios (LLRs) vital for soft channel decoders. In this study, we introduce a lookup table (LUT) aided soft output semidefinite relaxation (soSDR) based sub-optimal FTN detector, which can be extended to higher modulation levels. This detector possesses polynomial computational complexity, given the negligible complexity associated with soft value generation. Our study assesses the performance of this soft output detector against that of the optimal FTN detector, Bahl, Cocke, Jelinek and Raviv (BCJR) algorithm as the benchmark. The likelihood values produced by our LUT aided semidefinite relaxation (SDR) based FTN signaling detector show promising viability in coded scenario.
Automatic pronunciation error detection (APED) plays an important role in the domain of language learning. As for the previous ASR-based APED methods, the decoded results need to be aligned with the target text so that the errors can be found out. However, since the decoding process and the alignment process are independent, the prior knowledge about the target text is not fully utilized. In this paper, we propose to use the target text as an extra condition for the Transformer backbone to handle the APED task. The proposed method can output the error states with consideration of the relationship between the input speech and the target text in a fully end-to-end fashion.Meanwhile, as the prior target text is used as a condition for the decoder input, the Transformer works in a feed-forward manner instead of autoregressive in the inference stage, which can significantly boost the speed in the actual deployment. We set the ASR-based Transformer as the baseline APED model and conduct several experiments on the L2-Arctic dataset. The results demonstrate that our approach can obtain 8.4\% relative improvement on the $F_1$ score metric.
Multi-user spatial modulation (SM) assisted by sparse code multiple access (SCMA) has been recently proposed to provide uplink high spectral efficiency transmission. The message passing algorithm (MPA) is employed to detect the transmitted signals, which suffers from high complexity. This paper proposes three low-complexity algorithms for the first time to the SM-SCMA. The first algorithm is referred to as successive user detection (SUD), while the second algorithm is the modified version of SUD, namely modified SUD (MSUD). Then, for the first time, the tree-search of the SM-SCMA is constructed. Based on that tree-search, another variant of the sphere decoder (SD) is proposed for the SM-SCMA, referred to as fixed-complexity SD (FCSD). SUD provides a benchmark for decoding complexity at the expense of bit-error-rate (BER) performance. Further, MSUD slightly increases the complexity of SUD with a significant improvement in BER performance. Finally, FCSD provides a near-optimum BER with a considerable reduction of the complexity compared to the MPA decoder and also supports parallel hardware implementation. The proposed algorithms provide flexible design choices for practical implementation based on system design demands. The complexity analysis and Monte-Carlo simulations of the BER are provided for the proposed algorithms.
Traditional beamforming of medical ultrasound images relies on sampling rates significantly higher than the actual Nyquist rate of the received signals. This results in large amounts of data to store and process, imposing hardware and software challenges on the development of ultrasound machinery and algorithms, and impacting the resulting performance. In light of the capabilities demonstrated by deep learning methods over the past years across a variety of fields, including medical imaging, it is natural to consider their ability to recover high-quality ultrasound images from partial data. Here, we propose an approach for deep-learning based reconstruction of B-mode images from temporally and spatially sub-sampled channel data. We begin by considering sub-Nyquist sampled data, time-aligned in the frequency domain and transformed back to the time domain. The data is further sampled spatially, so that only a subset of the received signals is acquired. The partial data is used to train an encoder-decoder convolutional neural network, using as targets minimum-variance (MV) beamformed signals that were generated from the original, fully-sampled data. Our approach yields high-quality B-mode images, with higher resolution than previously proposed reconstruction approaches (NESTA) from compressed data as well as delay-and-sum beamforming (DAS) of the fully-sampled data. In terms of contrast to noise ratio, our results are comparable to MV beamforming of the fully-sampled data, thus enabling better and more efficient imaging than is mostly used in clinical practice today.
Recent advances in Automatic Speech Recognition (ASR) demonstrated how end-to-end systems are able to achieve state-of-the-art performance. There is a trend towards deeper neural networks, however those ASR models are also more complex and prone against specially crafted noisy data. Those Audio Adversarial Examples (AAE) were previously demonstrated on ASR systems that use Connectionist Temporal Classification (CTC), as well as attention-based encoder-decoder architectures. Following the idea of the hybrid CTC/attention ASR system, this work proposes algorithms to generate AAEs to combine both approaches into a joint CTC-attention gradient method. Evaluation is performed using a hybrid CTC/attention end-to-end ASR model on two reference sentences as case study, as well as the TEDlium v2 speech recognition task. We then demonstrate the application of this algorithm for adversarial training to obtain a more robust ASR model.
The reliability of using fully convolutional networks (FCNs) has been successfully demonstrated by recent studies in many speech applications. One of the most popular variants of these FCNs is the `U-Net', which is an encoder-decoder network with skip connections. In this study, we propose `SkipConvNet' where we replace each skip connection with multiple convolutional modules to provide decoder with intuitive feature maps rather than encoder's output to improve the learning capacity of the network. We also propose the use of optimal smoothing of power spectral density (PSD) as a pre-processing step, which helps to further enhance the efficiency of the network. To evaluate our proposed system, we use the REVERB challenge corpus to assess the performance of various enhancement approaches under the same conditions. We focus solely on monitoring improvements in speech quality and their contribution to improving the efficiency of back-end speech systems, such as speech recognition and speaker verification, trained on only clean speech. Experimental findings show that the proposed system consistently outperforms other approaches.
This paper presents a novel model-driven deep learning (DL) architecture, called TurboNet, for turbo decoding that integrates DL into the traditional max-log-maximum a posteriori (MAP) algorithm. The TurboNet inherits the superiority of the max-log-MAP algorithm and DL tools and thus presents excellent error-correction capability with low training cost. To design the TurboNet, the original iterative structure is unfolded as deep neural network (DNN) decoding units, where trainable weights are introduced to the max-log-MAP algorithm and optimized through supervised learning. To efficiently train the TurboNet, a loss function is carefully designed to prevent tricky gradient vanishing issue. To further reduce the computational complexity and training cost of the TurboNet, we can prune it into TurboNet+. Compared with the existing black-box DL approaches, the TurboNet+ has considerable advantage in computational complexity and is conducive to significantly reducing the decoding overhead. Furthermore, we also present a simple training strategy to address the overfitting issue, which enable efficient training of the proposed TurboNet+. Simulation results demonstrate TurboNet+'s superiority in error-correction ability, signal-to-noise ratio generalization, and computational overhead. In addition, an experimental system is established for an over-the-air (OTA) test with the help of a 5G rapid prototyping system and demonstrates TurboNet's strong learning ability and great robustness to various scenarios.
In this work we propose a multi-scale recurrent encoder-decoder architecture to predict the breathing induced organ deformation in future frames. The model was trained end-to-end from input images to predict a sequence of motion labels. Targets were created by quantizing the displacement fields obtained from deformable image registration. We report results using MRI free-breathing acquisitions from 12 volunteers. Experiments were aimed at investigating the proposed multi-scale design and the effect of increasing the number of predicted frames on the overall accuracy of the model. The proposed model was able to predict vessel positions in the next temporal image with a mean accuracy of 2.07 (2.95) mm showing increased performance in comparison with state-of-the-art approaches.
This study (The work was accomplished during the internship in Tencent AI lab) addresses semi-supervised acoustic modeling, i.e. attaining high-level representations from unsupervised audio data and fine-tuning the parameters of pre-trained model with supervised data. The proposed approach adopts a two-stage training framework, consisting of masked pre-trained encoder (MPE) and Joint CTC-Transformer (JCT). In the MPE framework, part of input frames are masked and reconstructed after the encoder with massive unsupervised data. In JCT framework, compared with original Transformer, acoustic features are applied as input instead of plain text. CTC loss performs as the prediction target on top of the encoder, and decoder blocks remain unchanged. This paper presents a comparison between two-stage training method and the fully supervised JCT. In addition, this paper investigates the our approach's robustness against different volumns of training data. Experiments on the two-stage training method deliver much better performance than fully supervised model. The word error rate (WER) with two-stage training which only exploits 30\% of WSJ labeled data achieves 17\% reduction than which trained by 50\% of WSJ in a fully supervised way. Moreover, increasing unlabeled data for MPE from WSJ (81h) to Librispeech (960h) attains about 22\% WER reduction.
Sparse signals, encountered in many wireless and signal acquisition applications, can be acquired via compressed sensing (CS) to reduce computations and transmissions, crucial for resource-limited devices, e.g., wireless sensors. Since the information signals are often continuous-valued, digital communication of compressive measurements requires quantization. In such a quantized compressed sensing (QCS) context, we address remote acquisition of a sparse source through vector quantized noisy compressive measurements. We propose a deep encoder-decoder architecture, consisting of an encoder deep neural network (DNN), a quantizer, and a decoder DNN, that realizes low-complexity vector quantization aiming at minimizing the mean-square error of the signal reconstruction for a given quantization rate. We devise a supervised learning method using stochastic gradient descent and backpropagation to train the system blocks. Strategies to overcome the vanishing gradient problem are proposed. Simulation results show that the proposed non-iterative DNN-based QCS method achieves higher rate-distortion performance with lower algorithm complexity as compared to standard QCS methods, conducive to delay-sensitive applications with large-scale signals.
In this paper, we propose a finite-precision decoding method that features the three steps of Reconstruction, Computation, and Quantization (RCQ). Unlike Mutual-Information-Maximization Quantized Belief Propagation (MIM-QBP), RCQ can approximate either belief propagation or Min-Sum decoding. One problem faced by MIM-QBP decoder is that it cannot work well when the fraction of degree-2 variable nodes is large. However, sometimes a large fraction of degree-2 variable nodes is necessary for a fast encoding structure, as seen in the IEEE 802.11 standard and the DVB-S2 standard. In contrast, the proposed RCQ decoder may be applied to any off-the-shelf LDPC code, including those with a large fraction of degree-2 variable nodes.Our simulations show that a 4-bit Min-Sum RCQ decoder delivers frame error rate (FER) performance around 0.1dB of full-precision belief propagation (BP) for the IEEE 802.11 standard LDPC code in the low SNR region.The RCQ decoder actually outperforms full-precision BP in the high SNR region because it overcomes elementary trapping sets that create an error floor under BP decoding. This paper also introduces Hierarchical Dynamic Quantization (HDQ) to design the non-uniform quantizers required by RCQ decoders. HDQ is a low-complexity design technique that is slightly sub-optimal. Simulation results comparing HDQ and an optimal quantizer on the symmetric binary-input memoryless additive white Gaussian noise channel show a loss in mutual information between these two quantizers of less than $10^{-6}$ bits, which is negligible for practical applications.
Spatially Coupled Low-Density Parity-Check (SC-LDPC) codes offer excellent decoding performance and can be elegantly decoded with a Windowed Decoder (WD). We determine an efficient WD configuration with low control overhead. For fair comparisons, we normalize all configurations to the same maximal computational complexity, which is an important measure of the decoding effort in packet-based data communication systems. We determine an optimized configuration from a joint evaluation of the window size, the window update strategy, and parity check-based Early Termination (ET). Firstly, we use a variable node-centered update strategy, which omits updates of messages in some parts of the decoding window. With the complexity normalization, the window size can be increased compared to a check node-centered update strategy, which uniformly updates all messages in the decoding window. Secondly, we only require the satisfaction of the top-most parity-check equations in each window to move to the next position more quickly. Using a surprisingly large window size, the resulting WD halves the average decoding complexity of the block decoder while maintaining a rather small gap in the decoding performance.
In recent years, deep neural networks have played a major role solving various challenges in two dimensional image processing.Fully Convolutional Networks (FCN) such as U-net have been shown to be highly successful at segmentation tasks for medical images analysis and denoising images taken in dark venues. This paper harnesses this well-known deep neural network for the channel decoding challenge recently proven to be suitable for deep neural networks.Previous work have successfully managed to decode convolutional codes using different architectures,such as Recurrent Neural Networks(RNN) and Fully Connected Neural Networks(FCNN) with promising results.However,these approaches are extremely costly in latency,computational resources and memory.This paper shows that taking the approach used in two dimensional image processing,by simple manipulation on the data in the preprocessing phase,achieves better results in a Bit Error Rate(BER) measurement with a large discount on the latency and the number of parameters required to maintain the neural decoder.
In this paper, the spectral efficiency of permutation modulation-based multiple input multiple output (MIMO) visible light communication is improved using systematically designed, multiweight codeword matrices. Soft-decision, low-complexity detection schemes are then designed for the receiver and compared with the maximum likelihood (ML) detection method. Bit error rate (BER) results show the soft-decision detection algorithm is able to decode the transmitted information without knowledge of the channel state information. This enables the mobile receiver decode information while within the field of view of the transmitter unit. The BER results also show a close match with the ML detection in some codebooks and the performance of the soft-decision decoder is evaluated for different positions of the receiver in an indoor environment.
In this paper, soft-decision (SD) decoders of permutation trellis code (PTC) with $M$-ary frequency shift keying are designed using three optimization algorithms and presented in four decoding schemes. In a concatenated code such as PTC, the Viterbi decoder for the outer convolutional code provides maximum likelihood decoding. Hence, the error correction performance is dependent on the decoding scheme used for the inner code. Due to the structure of the encoder with the modulation scheme, the channel output can be interpreted as an assignment problem. SD decoding can then be designed accordingly, using the presented, low-complexity optimization-based schemes. The bit error rate (BER) performance of the schemes are simulated in an additive white Gaussian noise (AWGN) and powerline communication (PLC) channel. The complexities of the schemes are also presented. The performance of the SD schemes are compared with the existing SD threshold detector, with BER results showing significant coding gain for certain codebooks. From the results, a reasonable trade-off between the complexity and coding gain is observed for a noisy channel such as the PLC channel.
Light field technology represents a viable path for providing a high-quality VR content. However, such an imaging system generates a high amount of data leading to an urgent need for LF image compression solution. In this paper, we propose an efficient LF image coding scheme based on view synthesis. Instead of transmitting all the LF views, only some of them are coded and transmitted, while the remaining views are dropped. The transmitted views are coded using Versatile Video Coding (VVC) and used as reference views to synthesize the missing views at decoder side. The dropped views are generated using the efficient dual discriminator GAN model. The selection of reference/dropped views is performed using a rate distortion optimization based on the VVC temporal scalability. Experimental results show that the proposed method provides high coding performance and overcomes the state-of-the-art LF image compression solutions.
In this paper, we present a downlink pilot design scheme for Deep Learning (DL) based channel estimation (ChannelNet) in orthogonal frequency-division multiplexing (OFDM) systems. Specifically, in the proposed scheme, a feature selection method named Concrete Autoencoder (ConcreteAE) is used to find the most informative locations for pilot transmission. This autoencoder consists of a concrete layer as the encoder and a multilayer perceptron (MLP) as the decoder. During the training, the concrete layer selects the most informative pilot locations, and the decoder reconstructs an approximate estimation of the channel. Eventually, the ChannelNet is trained on the output of the ConcreteAE aiming to reconstruct the ideal channel response. The estimation error results show that this approach outperforms the previously presented ChannelNet with a uniformly distributed pilot pattern, and its performance is comparable to the minimum mean square error (MMSE).
In this paper, we design a novel high-throughput random access scheme for an indoor optical wireless communication (OWC) massive Internet of Things (IoT) scenario. Due to the large density of both IoT devices and OWC access points (APs), we base the proposed scheme on multi-receiver coded slotted ALOHA. In this scenario, collisions at APs are resolved by a centralized interference cancellation decoder that may exploit both spatial and temporal diversity. By applying adaptive control of each OWC AP field of view (FOV), the proposed system is able to dynamically adapt to different IoT device activation rates, in order to maintain a high total throughput. Using illustrative simulation results, we demonstrate the design methodology and performance possibilities of the proposed method.
Versatile Video Coding (VVC) is the next generation video coding standard expected by the end of 2020. Compared to its predecessor, VVC introduces new coding tools to make compression more efficient at the expense of higher computational complexity. This rises a need to design an efficient and optimised implementation especially for embedded platforms with limited memory and logic resources. One of the newly introduced tools in VVC is the Multiple Transform Selection (MTS). This latter involves three Discrete Cosine Transform (DCT)/Discrete Sine Transform (DST) types with larger and rectangular transform blocks. In this paper, an efficient hardware implementation of all DCT/DST transform types and sizes is proposed. The proposed design uses 32 multipliers in a pipelined architecture which targets an ASIC platform. It consists in a multi-standard architecture that supports the transform block of recent MPEG standards including AVC, HEVC and VVC. The architecture is optimized and removes unnecessary complexities found in other proposed architec-tures by using regular multipliers instead of multiple constant multi-pliers. The synthesized results show that the proposed method which sustain a constant throughput of two pixels/cycle and constant la-tency for all block sizes can reach an operational frequency of 600 Mhz enabling to decode in real-time 4K videos at 48 fps.