For processors with vectorial computing units like DSP, it is very important to ensure vector load/store operations alignment of memory blocks, and minimize space wastage when making memory allocations. In this paper, we design and implement a memory management method, vector memory pool, suitable for embedded vector processors. By partitioning an entire block of memory space into many aligned vector objects and making efficiently use of vector processing units, the processing of memory manipulation library functions such as memset/memcpy is accelerated. The implementation and comparative verification of vector memory pool on RT-Thread Nano based on SWIFT DSP was completed, and the running efficiency reached a tens of times improvement compared to the original method.
The sixth generation (6G) mobile networks need to meet various performance requirements such as the number of connections, latency, reliability, and energy efficiency. In particular, for the Internet of Things (IoT) scenarios with short packet transmission, it is necessary to analyze and optimize various performances while achieving massive connections. In addition, practical constraints (such as imperfect channel state information, limitations of classical Shannon’s capacity, inter-cell interference, massive user interference, etc.) further aggravate the difficulties of theoretical analysis and performance improvement. We propose a performance analysis and optimization strategy for short packet transmission systems based on cell-free massive multiple-input multiple-output (CF mMIMO), which points out the idea of improving system performance with large-scale connections under practical constraints. Furthermore, with the combination of simultaneous wireless information and power transfer (SWIPT) technology and finite blocklength (FBL) information theory, we derive the closed-form expressions of downlink signal-to-interference-plus-noise ratio (SINR), achievable data rate, and energy collected based on CF mMIMO. Simulation results verify the effectiveness of the proposed strategy, which is also expected to support massive ultra-reliable and low latency communications (mURLLC) with ultra-high energy efficiency or spectral efficiency in the future.
With the rapid development of Internet of Things (IoT) technology, the number of large numbers of Machine Type Communication (MTC) devices involved in M2M has increased dramatically. When large scale MTC devices access the base station at the same time in a short period of time, this can cause traffic overload and lead to a sharp drop in the success rate of access of MTC devices. 3GPP has proposed the access class barring (ACB) scheme to defer access requests from certain activated MTC devices to avoid congestion at the base station (BS). In this paper, we propose a dynamic ACB scheme for grouping MTC devices and a resource allocation scheme for preamble. First, MTC devices are classified into two categories according to their characteristics: delay-sensitive and energy-constrained. The two categories use separate preamble resources, and a temporary ACB factor is calculated for each time slot based on the current preamble resources and the number of devices. The preamble resources are reallocated based on this temporary ACB factor using reinforcement learning methods, and then the ACB factor is dynamically adjusted according to the new preamble resources. Simulation results show that the solution improves the access success rate of M2M devices, reducing the total service time of delay-sensitive devices by 40 $$\%$$ compared to the traditional solution, while reducing the access collision rate of energy-constrained devices by 30 $$\%$$ .
Orthogonal time frequency space (OTFS) modulation can provide significant error performance than orthogonal frequency division multiplexing (OFDM) modulation in the high-speed scenario. However, the fractional Doppler effects cause Doppler diffusion. In this paper, we analyze the Doppler diffusion from both the formula and geometric levels. In order to alleviate Doppler diffusion, we propose an adaptive orthogonal basis scheme by using the Doppler shifts feedback of the receiver. Our scheme alter the matrix of the inverse symplectic finite Fourier transform (ISFFT) by the feedback. This scheme makes it possible to estimate the channel more accurately. In the simulation results, we show that the bit error rate (BER) and block error rate (BLER) performance of our proposed scheme is not affected compared with OTFS. In addition, our scheme can work even if the Doppler domain dimension is limited and the Doppler shifts feedback of the receiver is inaccurate.
With the development of fifth generation (5G) technology, mobile edge computing (MEC) is becoming an essential architecture which is envisioned as a cloud extension version. MEC system can push the resources from cloud side to edge side, aiming to solve many computation intensive problems. The task offloading policy is vital and has an important influence on MEC system. Meanwhile, privacy leakage may occur during the task offloading period which may degrade MEC system performance. The attention on these issues is lack according to existing works. Inspired by this, we present a privacy-preserving aware Multi-Armed Bandits based task allocation algorithm, Privacy Upper Confidence Bound (pUCB), to find a balance between the privacy preserving and the efficiency of task processing. In addition, we take regret analysis of the proposed algorithm. The extensive simulation results show that pUCB scheme can achieve a higher optimal rate, a lower lock rate and less total time cost comparing with traditional Multi-arm bandits (MAB) based algorithm.
Applications of digital signal processor (DSP) involve large amounts of data processing. In order to be able to improve the speed of DSP application development, it is necessary to be able to implement debugging functions that can support File I/O at a small cost of modification. This paper proposes a complete set of implementation methods of the semihosting debugging function on a DSP without an operating system, including hardware support and software algorithms. On hardware, this article adds only one self-trapping instruction to support semihosting debugging. In software, this article is based on the GDB File I/O extension protocol to design and implement the debugging agent software and C language library underlying file operations. And after optimizing library files and application source code, the I/O speed of the architecture can meet users’ debugging needs. This solution can realize the semihosting debugging function for DSP chip at a low cost and has good performance. Therefore, provides more powerful debugging functions for DSP application development. In assembly level debugging, observing the execution of each assembly instruction can also verify the instruction execution of the chip itself. The addition of file operation can greatly validate the chip with larger data.
Speech enhancement algorithms based on deep learning have greatly improved speech’s perceptual quality and intelligibility. Complex-valued neural networks, such as deep complex convolution recurrent network (DCCRN), make full use of audio signal phase information and achieve superior performance, but complex-valued operations increase the computational complexity. Inspired by the deep cosine transform convolutional recurrent network (DCTCRN) model, in this paper real-valued discrete cosine transform is used instead of complex-valued Fourier transform. Besides, the ideal cosine mask is employed as the training target, and the real-valued convolutional recurrent network (CRNN) is used to enhance the speech while reducing algorithm complexity. Meanwhile, the frequency-time-LSTM (F-T-LSTM) module is used for better temporal modeling and the convolutional skip connections module is introduced between the encoders and the decoders to integrate the information between features. Moreover, the improved scale-invariant source-to-noise ratio (SI-SNR) is taken as the loss function which enables the model to focus more on the part of signal variation and thus obtain better noise suppression performance. With only 1.31M parameters, the proposed method can achieve noise suppression performance that exceeds DCCRN and DCTCRN.
In order to solve the problem of resources waste caused by insufficient computing resources and the lack of an effective admission mechanism in the service environment of wireless virtual network (WVN), this paper proposes an effective robust admission control mechanism. This mechanism mainly controls the access of user groups of WVN in a dynamic resource environment, which is in the assistance environment of vehicular fog computing (VFC). Firstly, considering the resource uncertainty caused by the characteristics of VFC, a robust optimization access model is established. It predicts the change of resources during the association of user time to determine whether to allow user groups access. Secondly, task offloading and the allocation of computing resources are processed. Since the coupling of task offloading and resource allocation leads to the non-convexity of the problem, we convert it into a convex optimization problem to resolve. Simulation results show that the admission control mechanism proposed in this paper can admit larger user groups in the assistance computing environment of VFC while ensuring the quality of user experience.
Novel view synthesis is regarded as one of the efficient ways to realize stereoscopic vision, which paves the way to virtual reality. Image-based rendering (IBR) is one of the view synthesis strategies, which warps pixels from source views to target views in order to protect low-level details. However, IBR methods predict the pixels correspondence in an unsupervised way and have limits in getting accurate pixels. In this paper, we propose Depth and Pose Net (DPNet) for novel view synthesis via depth map estimation. We introduce two nearby views as implicit supervision to improve the pixels correspondence accuracy. Besides, the depth net firstly predicts the source depth map and then the pose net transforms the source depth map to the target depth map which is used to calculate pixels correspondence. Experimental results show that DPNet generates accurate depth maps and thus synthesizes novel views with higher quality than state-of-the-art methods on the synthetic object and real scene datasets.
Multi-beam satellite communication systems have received considerable attentions in recent years. By generating multiple beams at the transmitting satellites, the coverage areas can be enlarged and the transmission capacity can be improved. In this paper, we study beam illumination and resource allocation problem in multi-beam satellite communication systems. Jointly considering the revenue received from successful packet transmission and the energy consumption required for packet transmission, we define a system utility function and formulate joint user grouping, beam illumination, and time-frequency resource allocation problem as a constrained utility function maximization problem. Since beam illumination and time-frequency resource allocation problems are of different time and regional scales, this paper proposes a two-stage resource allocation algorithm. On relatively large temporal scale, we propose an offline user grouping algorithm and a group-oriented beam illumination scheme. Then based on the obtained user grouping and beam assignment strategy, an online resource scheduling algorithm is proposed, which schedules time and frequency resources for individual users. Numerical results verify the effectiveness of the proposed algorithm.
In network intrusion detection, the frequency of some rare network attacks is low, and such samples collected are relatively few. It results in an imbalanced proportion of each category in the dataset. Training the classifier with imbalanced datasets will bias the classifier to majority class samples and affect the classification performance on minority class samples. In response to this problem, researchers usually increase minority class samples and reduce majority class samples to get a balanced dataset. Therefore, we propose a data balancing technique based on AutoEncoder-Flow (AE-Flow) Model. Firstly, we use AutoEncoder (AE) to improve the deep generative model-Flow, obtaining AE-Flow. Then we use it to learn the distribution of minority class samples and generate new samples. Secondly, we use K-means and OneSidedSelection (OSS) algorithms to finish the undersampling of majority class samples. Finally we get a balanced dataset and use machine learning (ML) classifier to finish intrusion detection. We conducted comparative experiments on NSL-KDD dataset. The experimental results show that the balanced dataset obtained by our proposed method can effectively improve the Recall rate on minority class samples and the classification performance on overall samples.
With the development of lightweight convolutional neural networks (CNNs), these newly proposed networks are more powerful than previous conventional models [4, 5] and can be well applied in Internet-of-Things (IoT) and edge computing. However, they perform inefficiently on conventional hardware accelerators because of the irregular connectivity in the structure. Though there are some accelerators based on unified engine (UE) architecture or separated engine (SE) architecture which can perform well for both standard convolution and depthwise convolution, these versatile structures are still not efficient for lightweight CNNs such as EfficientNet-lite. In this paper, we propose a reconfigurable engine (RE) architecture to improve the efficiency, which is used in communications such as IoT and edge computing. In addition, we adopt integer quantization method to reduce computational complexity and memory access. Also, the block-based calculation scheme is used to further reduce the off-chip memory access and the unique computational mode is used to improve the utilization of the processing elements. The proposed architecture can be implemented on Xilinx ZC706 with a 100 MHz system clock for EfficientNet-lite0. Our accelerator achieved 196 FPS and 72.9
Information-Centric Networking (ICN) replaces identifying endpoints with identifying content. One of the most common and important features of ICN architectures is in-network caching, which can significantly reduce content request latency and improve user quality of service(QoS) and network performance. Therefore, how to efficiently utilize cache resources and optimize cache performance has become one of the research hotspots in ICN. This paper proposes a local popularity caching strategy based on content relevance caching (CRC). In this strategy, the local popularity of the content is calculated based on the relevance of the content requested. Routing nodes make caching decisions based on local popularity. In the forwarding process of the interest packet, the cache node ID table in the interest packet is updated according to the local popularity, and the cache decision is made. In the backhaul process of the data packet, content caching and replacement strategy need to be performed according to the caching parameters and local popularity parameters. In cache replacement, if the remaining cache capacity of the current routing node cannot satisfy the required capacity of the cached content, the local popularity of the content needs to be used to replace the cached content. In this way, repeatedly requested content ends up being cached in routing nodes closer to the user. The simulation results show that CRC has better network performance than several other classical caching strategy.
To address the problems of low network throughput and delivery rate and high communication overhead between nodes due to frequent topology changes of flight self-assembled networks, this paper proposes a SE-AODV routing protocol optimized based on the traditional on-demand distance vector routing protocol. The new protocol takes into account the actual influencing factors such as energy and received signal strength, and combines the pheromone calculation rules of the ant colony algorithm to update the routing table, while optimizing the optimal path selection basis. Simulation results show that compared with traditional on-demand distance vector routing and EE-AODV, the new routing protocol not only shows more obvious advantages in throughput and packet delivery rate, but also reduces routing overhead, and is more suitable for scenarios where nodes are in high-speed motion.
As an important part of modern communication, the satellite communication is attracting more and more attention. In recent years, the low earth orbit (LEO) satellite constellations are being developed vigorously. To achieve better performance and flexibility, many researchers have proposed to combine the LEO and geostationary earth orbit (GEO) satellites to construct a double-layer network. However, this double-layer structure brings significant challenges to the design of routing strategy. In this paper, we propose a routing strategy for GEO/LEO satellite network based on dynamic delay prediction and link control. In particular, we craft a link cost function that comprehensively considers the link load, queuing delay, processing delay and transmission delay. The cost function can provide more accurate estimation of the link delay compared with the existing methods roughly using hop counts. Based on it, the proposed double-layer strategy can not only optimize the end-to-end delay but also achieve better traffic balance. Simulations on STK and OPNET verify that the proposed approach can greatly improve the performance of packet loss rate, end-to-end delay and throughput.
Automatic detection of anomalous sounds is very important for industrial equipment maintenances. However, anomalous sounds are difficult to collect in practice, and self-supervised methods have received extensive attentions. It is well-known that the self-supervised methods show poor performances on certain machine types. To improve the detection performance, in this work, we introduce other types of data as targets to train a general classifier. After that, the model has certain prior knowledge, and then we fine tune the parameters of the model for a specific machine type. We also studied the impact of input features on performance, and it is shown that for machine types, filtering out low-frequency noise interference can significantly improve model performance. Experiments conducted using the DCASE 2021 Challenge Task2 dataset showed that the proposed method improves the detection performance on each machine type and outperforms the DCASE 2021 Challenge first-placed ensemble model by 8.73% on average according to the official scoring method.
In recent years, the attention transformer model has been widely used in the field of speech enhancement. With the introduction of a convolutionally enhanced transformer (Conformer), it models both the local and the global information of the speech sequence to achieve a better performance. In this paper, we propose a speech enhancement structure using conformer with time-frequency (TF) domain in DCCRN. To that aim, the second layer LSTM in DCCRN is replaced with TF-Conformer. By doing this, information between and within frames can be better utilized. An attention convolution path between the convolutional encoder and decoder is also developed to better convey nonlinear information. The results show that the model’s PESQ surpasses DCCRN and DCCRN+ on the testset of Interspeech 2020 Deep Noise Suppression (DNS) Challenge, with the best model size of 2.3 M. At the same time, the excellent results have been obtained on the blind test set of ICASSP 2021 DNS Challenge, and the overall MOS score exceeds the winner team by 0.06.
In this paper, a full-duplex orthogonal frequency division multiplexing (OFDM) millimeter wave systems with correlated estimation errors and partially-connected structures is investigated to improve the spectral efficiency and energy efficiency. To attenuate the self-interference caused by full-duplex mode, a zero-space projection based method is proposed. Then, a two-stage hybrid beamforming scheme is developed. In the fist stage, the correlated channel estimation errors are considered to design the robust fully digital beamformers. In the second stage, an alternating algorithm with closed-form solutions is proposed to solve the hybrid processors. The numerical results show that the proposed scheme has superior performance over existing designs.
In this paper, we investigate a millimeter wave multi-user system to further improve the spectral efficiency. Millimeter wave channels with correlated estimation errors and sub-connected structures are considered to develop a two-stage hybrid beamforming scheme. In the first stage, the analog parts of beamformers are designed to maximize RF-to-RF channel gains. In the second stage, the digital parts of the beamformers are optimized by utilizing the equivalence between the maximization of mutual information and the minimization of weighted minimum mean square error. The numerical results show that the proposed scheme has superior performance over other existing designs.
In this paper we construct a hybrid architecture that combines terrestrial cell-free massive MIMO system and multi-beam satellite communication system to maximize the throughput within the coverage. We propose an interference-level based user selection algorithm to divert terrestrial users with stronger interference to satellite service, thereby reducing the total system interference. For satellite users, to further increase the system throughput, we use the co-frequency across beams, which include two steps, inter-beam and intra-beam resource allocation. The former uses soft frequency reuse and the latter is allocated based on water-filling algorithm. The simulation results show that the proposed satellite-aided cell-free system has a significant improvement in throughput compared with the traditional system. At the same time, the throughput of satellite users is also improved compared with the traditional frequency reuse scheme.
