In the process of tracking and obstacle avoidance control of driverless vehicles, the controlled object has nonlinear characteristics and variable control parameters. The linear model and the fixed mathematical model of driverless vehicles are difficult to ensure the safety and stability of the vehicle in complex environments, and the driverless discrete control process increases the difficulty of control. To address such problems, in order to improve the accuracy of real-time control tracking trajectory of driverless vehicles, and at the same time reduce the difficulty of the whole control process, the paper proposed a Monte Carlo-depth deterministic policy gradient-based obstacle avoidance tracking control algorithm for driverless vehicles. The algorithm builds a control system model based on a deep reinforcement learning network, and adopts excellent training samples in the strategy learning sampling process. It optimizes the network training gradient with the Monte Carlo method, and makes a distinction between good and bad training samples for the algorithm. The excellent samples are used to find the optimal network parameters through a gradient algorithm, so as to enhance the learning ability of the network algorithm and realize a better and continuous control of the driverless vehicle. Simulation experiments of the control method were carried out in the computer simulation environment TORCS. The results show that the proposed improved DDPG algorithm can be applied to effectively achieve the obstacle avoidance tracking control of the driverless vehicle, and the tracking accuracy and obstacle avoidance effect of the unmanned car under its control is better than that of the deep Q network algorithm and the DDPG algorithm.

Aiming at the problem of multiple types of requirements and large volume, this paper introduced a highly integrated and reconfigurable Global Navigation Satellite System (GNSS) Radio Frequency (RF) receiver for high-precision positioning and orientation of satellite navigation. By adopting four reconfigurable reception channels and receiving full frequency band GNSS signals in parallel, it achieved a single chip supporting high-precision positioning or orientation applications, significantly reducing the volume and cost of navigation terminals. To improve broadband signal reception, the paper proposed a new type of inductorless high linear low noise transconductance amplifier (LNTA). It can eliminate the use of source and load inductors, reduce gain and noise fluctuations when operating in navigation signals at different frequency points, and it is conducive to the reconfiguration of multimode and multi frequency reception and reduces the power consumption of LNTA. A novel IQ phase compensation method was proposed to address the issue of IQ phase imbalance. A programmable switch array with variable impedance was directly designed on the clock controlled latch path of the binary frequency division circuit. By changing the delay time of the 25% duty cycle orthogonal LO, the corresponding branch output LO phase adjustment was realized, achieving calibration of IQ imbalance and improving the image rejection rate (IRR). Testing data shows that the RF receiver achieves full band signal coverage of GNSS from 1.15 to 1.65 GHz, a minimum noise figure of 2.7 dB, and an output third-order intermodulation point power of 34.7 dBm. Adopting a low intermediate frequency and zero intermediate frequency reconfigurable architecture, it can flexibly receive multimode GNSS signals with a bandwidth of 0.8~80 MHz. By compensating for IQ imbalance and improving channel layout, 58.1 dB IRR and 57 dB channel isolation can be achieved, effectively reducing the impact of image interference and inter channel interference. Under a 1.2 V power supply, the power consumption of the receiving channel is only 24.7 mW, which can meet the high integration and diversified application requirements of high-precision positioning and orientation GNSS RF receivers.

In the field of warehousing and logistics, automatic guided truck system (AGVS) has the merits of high reliability and flexibility, but with the increase of the complexity of its working environment, the difficulty of path planning also increases. Aiming at the problem of low efficiency and easy conflict in AGVS path planning in complex environment, this paper proposed an improved AGVS path planning algorithm based on hierarchical distributed framework. Firstly, in order to improve the search efficiency of the algorithm in the path planning process, the evaluation function of the traditional A ^* algorithm was improved and fused with the bidirectional Floyd algorithm to increase the path smoothness, and the global optimal AGVS path is finally obtained. Secondly, the AGVS kinematics modeling was established, and the key nodes in the global optimal path were taken as temporary target points. By adjusting the initial poses of the robot and optimizing the evaluation function, the AGVS local path planning was completed appying the DWA algorithm to the temporary target points. Finally, AGVS collaborative planning strategy was introduced to achieve unified scheduling of inter-AGVS motion by assigning task priorities to AGVS, reducing the probability of conflicts between mobile machines, improving the robustness of AGVS path planning algorithm. Matlab simulation results show that the proposed improved algorithm can generate collision-free paths in both simple and complex environments. In complex environments, AGVS path length planned by the improved algorithm is shortened by 2.26% compared with that planned by the traditional A ^* algorithm. In the process of AGVS motion, the angular velocity and the linear velocity of the mobile robot are always maintained within -0.4~0.4 rad/s, and 0.6~1.2 m/s, which conforms to the kinematic characteristics of the mobile robot.

With the characteristics of excellent electrothermal performance and high-density signal interconnection, ceramic column grid array (CCGA) packaging is the first choice in highly reliable applications such as aerospace.When the pin exceeds 1 000, due to the packaging form and the characteristics of the material itself, high-density CCGA is more likely to fail in the environment of temperature change.The paper carried out temperature cycling test for CCGA1144 structure, and studied the stress distribution, variation law, failure cause and mode of welding column through microstructure observation and finite element simulation under the temperature cycling condition of -55~125 ℃. The results show that the variation range of total deformation, equivalent stress, equivalent elastic strain and plastic strain of outer ring welding column is larger than that of inner ring welding column in the process of temperature cycle, and it is more prone to failure, especially the edge welding column of outer ring. The paper identified the weakness of welding column in the thermal fatigue test and pointed out that the two sections of 0.15~0.60 mm and 2.17~2.43 mm are the dangerous areas of crack and fracture failure, and the latter section is more prone to crack and fracture. It put forward three failure modes of thermal fatigue of welding column, and pointed out that the welding column was damaged under the joint action of fatigue mechanism and creep mechanism. Suggestions on the direction of reinforcement and optimization design were given.The research results have guiding significance and reference value for the quality improvement, development and application of high-density CCGA.

Intelligent reflecting surface (IRS) and cognitive simultaneous wireless information and power transfer (SWIPT) are regarded as potential key technologies to improve energy efficiency and spectrum utilization of wireless communication systems. This paper studied the IRS-aided cognitive SWIPT network based on a nonlinear energy harvesting model. In the network, a secondary transmitter simultaneously transmits information and energy to multiple secondary receivers, and each secondary receiver adopts the power splitting scheme to realize information decoding and energy harvesting. The aim is to minimize the transmit power of the secondary user transmitter by jointly optimizing the beamforming vector of the secondary transmitter, the power splitting coefficients of the secondary receivers, and the phase shifts of the IRS. In order to guarantee the information and energy transmission efficiency of the secondary users and limit the co-channel interference from the secondary users to the primary users, it is considered that the secondary receivers have the constraints of the minimum received signal-to-interference noise ratio, the minimum energy harvesting amount, and the values of power splitting coefficients, the secondary transmitter has the constraints of the maximum interference power values to the primary users, and the IRS has the constraints on its reflection phase shifts. The considered optimization problem is a non-convex quadratically constrained quadratic program problem with highly coupled optimization variables, which is difficult to solve. An alternating optimization algorithm based on the semidefinite relaxation and sequential rank one constraint relaxation techniques was proposed to solve the problem efficiently. In order to reduce the computation complexity, a low-complexity optimization algorithm based on IRS element grouping was further proposed. Simulation results show that compared to several benchmark algorithms, the proposed algorithms can effectively reduce the transmit power of the secondary transmitter.

Currently, the hardware design of spiking neural networks based on digital circuits has a low synaptic parallel nature in terms of learning function, leading to a large overall hardware delay, which limits the speed of online learning of spiking neural network models to some extent. To address the above problems, this paper proposed an efficient spiking neural network online learning hardware architecture based on FPGA parallel acceleration, which accelerates the training and inference process of the model through the dual parallel design of neurons and synapses. Firstly, a synaptic structure with parallel spike delivery function and parallel spike time-dependent plasticity learning function was designed; then, the learning layers of input encoding layer and winner-take-all structure were built, and the implementation of lateral inhibition of the winner-take-all network was optimized, forming an impulsive neural network model with a scale of 784~400. The experiments show, the hardware has a training speed of 1.61 ms/image and an energy consumption of about 3.18 mJ/image for the SNN model and an inference speed of 1.19 ms/image and an energy consumption of about 2.37 mJ/image on the MNIST dataset, with an accuracy rate of 87.51%. Based on the hardware framework designed in this paper, the synaptic parallel structure can improve the training speed by more than 38%, and reduce the hardware energy consumption by about 24.1%, which can help to promote the development of edge intelligent computing devices and technologies.

In recent years, cars have been becoming more and more intelligently connected, and modern vehicles have been transformed into mobile computer platforms with information systems such as navigation systems, entertainment systems, climate control systems, and vehicle performance systems. The amount of information that drivers can access and interact with in the vehicle has increased significantly. However, operating a wide range of automotive information systems while driving can distract the driver’s visual and auditory attention, causing distracted driving and increasing the likelihood of traffic accidents. The use of haptic modalities while driving can complement audiovisual modalities, reduce the driver’s audiovisual burden, and improve driving safety, interaction reliability and comfort. On the basis of summarizing the development history of haptic feedback technology in automotive applications, this paper classified the specific types of in-vehicle interaction activities and haptic modalities that can be displayed in the vehicle, and analyzed the research progress of haptic feedback technology in automotive information systems from three directions, namely, in-vehicle haptic warning system, in-vehicle haptic navigation system, and in-vehicle haptic operation assistance system. It also summarized the main sources of haptic excitation in in-vehicle warning systems, navigation systems and operation assistance systems, respectively. And by analyzing the relevant literature, it expatiated the experimental environment, evaluation indexes, evaluation methods and results of the research on the influence of automotive haptic information systems on driving performance. Based on the shortcomings of existing technologies and research methods, the paper discussed and prospected the future research of automotive haptic information system from the perspectives of the location of haptic feedback excitation signals, the selection of modes and parameters, the study of behavioral adaptation problems, the improvement of the experimental environment, the consideration of different factors affecting haptic perception, and the integration of multiple automotive haptic information systems.

Wastewater treatment system is a dynamic system with complex nonlinearity and large time delay. Due to the complexity of the process, the incompleteness of the testing equipment and the constraint of economic cost, some important effluent indicators cannot be detected accurately. To solve this problem, this paper proposes a soft-sensor method based on an ensemble Kalman filter-Elman neural network. The traditional dynamic neural network has the dynamic memory ability to process time-delay data, so it can be used in data-driven soft sensing modeling. However, the conventional training method is easy to trap in a local minimum, resulting in poor prediction performance. This paper introduces the ensemble Kalman filter and the dual finite-size ensemble Kalman filter, and, together with the Elman neural network for gradient-free training, to construct two soft sensor models, which not only improve the prediction performance of Elman neural network but also provide a simple and gradient-free training method for neural network. The two models are then applied to a dataset of the University of California, Irvine (UCI data). The results show that the proposed method based on ensemble Kalman filter-Elman neural network possesses good prediction performance, and that the ensemble Kalman filter can be used as an alternative gradient-free method to train neural networks.

For industrial information physical systems (ICPS) with gap denial of service (DoS) attacks, an observer-based adaptive event-triggered stability control scheme was proposed. Firstly, an observer was constructed to estimate the unmeasured state of the system, and a new adaptive event-triggered communication scheme was adopted to save network resources. Secondly, an attack model was established based on the influence of intermittent DoS attacks on ICPS limited by frequency and duration. ICPS subjected to DoS attacks was divided into active DoS attack system and dormant DoS attack system, and the current attacked states were reconstructed by some successful trigger states. An attack compensation mechanism based on dynamic estimator was proposed to ensure the asymptotic stability of the system during active DoS attacks. Finally, the piecewise Lyapunov functional method, Jensen’s inequality and Schur’s complement lemma were used to obtain the stability conditions of the system. On this basis, a cooperative design scheme of adaptive event triggering parameters, observer gain and controller gain was proposed. The intermittent reactor system was used to simulate the stability of ICPS based on adaptive event triggering under intermittent DoS attack. The simulation results show that the adaptive event triggering strategy and the compensation mechanism based on the dynamic estimator can not only improve the stability of the system, but also effectively reduce the number of data transmission. The times of the system triggered by the static event triggering policy is about 4.5 times that of the system triggered by the adaptive event triggering policy.

As a new generation technology of Ethernet, time sensitive networking (TSN) plays an increasingly important role in industrial control, vehicle network and other fields, as it guarantees low-delay and low-jitter transmission of time sensitive traffic. As one of the key shaping technologies of TSN, credit-based shaping (CBS) guarantees the deterministic transmission of audio video bridging (AVB) traffic by reserving bandwidth. The existing bandwidth allocation methods based on network calculus mostly do not consider the impact of routing on the schedulability of AVB traffic, and the bandwidth allocation results are poor and the solution time is long when the network is in a large scale. Therefore, this paper proposed a bandwidth allocation method based on link load balancing. Firstly, the link load balancing routing algorithm was used to calculate the optimal path for each AVB traffic. Then, based on the flow path and network calculus, the arrival and service curves of each switch’s outbound port traffic were analyzed to obtain the worst-case forwarding delay. Finally, the optimization objectives and constraints for bandwidth allocation were established, a heuristic algorithm was used to solve bandwidth allocation, and the bandwidth parameter configuration was optimized during the solving process. The experimental results show that the AVB traffic bandwidth allocation method based on link load balancing can improve the schedulability of AVB traffic by 15 to 45 percentage points, compared with existing bandwidth allocation methods. Optimizing parameter configuration can increase the bandwidth solving speed by more than twice and obtain better bandwidth allocation results, which can effectively cope with large-scale dynamic changes in TSN networks.

Body Area Network (BAN) is an important component for remote healthcare and health monitoring, and existing research often utilizes electromagnetic wave transmission for data transfer within the network. In contrast, ultrasound wave propagation in human soft tissue experiences less attenuation and generates less heat. Compared to electromagnetic wave transmission, ultrasound-based communication and networking within the human body have certain advantages, such as longer transmission distances and lower risks of inducing pathogenic effects. The existing ultrasonic wideband communication technology often uses ultra-short pulse to reduce path overlapping and the resulting interference. Besides, the multiple access through channel sharing can be realized by combining direct sequence spread spectrum. However, the interference caused by the increase of communication number reduces the received signal-to-noise ratio. Communication reliability can not be guaranteed with the fixed frame length and code length. In this paper, the adaptive rate adjustment of multiple access was modeled mathematically and the implementation methods based on competition and cooperation were studied. Then the close form solution was derived with the Lagrange multiplier method in convex optimization theory, with which the frame length and code length were dynamically changed to realize rate adjustment. Simulation results show that those methods can adapt to the changes of communication links and the resulting interference; the cooperative method based on bargaining game theory can obtain better network performance by balancing the communication rate of different access nodes; the competition scheme based on the maximum rate is helpful to improve the effective communication rate of the target communication link. In practical application, scheme should be selected according to the application scenario.

Automatic speech recognition (ASR) technology has been developed relatively mature, and general ASR engines have been widely used in transportation, medical, communication and other industries. However, due to non-independent homology of industry-specific vocabulary in the large-scale training corpus, there comes to low recognition accuracy of industry-specific vocabulary when the general ASR engines are applied to various subdivisions of industries. As compared with 16 kHz audio sampling rate in Internet environment, narrowband low sampling (8 kHz) of call center may result in more significant decrease of recognition accuracy of ASR. In order to improve the accuracy of speech recognition of industry-specific words, this paper proposes a translation optimization technology of ASR based on industry-specific vocabulary. Specifically, first, convolutional neural network model and deep neural network BERT model are used to predict word for corpus text data, and an industry-specific error correction vocabulary is generated. Next, in the production environment, a general ASR engine is used to perform initial transcription of telephone call voice data. Then, the transcribed text is corrected by using the Soft-Masked BERT model combined with the industry-specific error correction vocabulary, thus improving the accuracy of speech recognition. Finally, by using 12345 hotline customer service call voice data for modeling and testing, the proposed translation optimization technology is proved efficient in improving the accuracy of general ASR recognition by 10 percentage points with high error correction speed and good applicability.

The traditional consistency judgment methods of lip motion and voice mainly focus on processing the frontal lip motion video，without considering the impact of angle changes on the result during the video acquisition process. In addition, they are prone to ignoring the spatio-temporal characteristics of the lip movement process．Aiming at these problems, this paper focused on the influence of lip angle changes on consistency judgment，combined the advantages of three dimensional convolutional neural networks for non-linear representation and spatio-temporal dimensional feature extraction, and proposed a multi-view lip motion and voice consistency judgment method based on frontal lip reconstruction and three dimensional（3D） coupled convolutional neural network．Firstly，the self-mapping loss was introduced into the generator to improve the effect of frontal reconstruction, and then the lip reconstruction method based on self-mapping supervised cycle-consistent generative adversarial network (SMS-CycleGAN) was used for angle classification and frontal reconstruction of multi-view lip image．Secondly，two heterogeneous three dimensional convolution neural networks were designed to describe the audio and video signals respectively, and then the 3D convolution features containing long-term spatio-temporal correlation information were extracted．Finally, the contrastive loss function was introduced as the correlation discrimination measure of audio and video signal matching, and the output of the audio-video network was coupled into the same representation space for consistency judgment. The experimental results show that the method proposed in this paper can reconstruct frontal lip images of higher quality， and it is better than a variety of comparison methods on the performance of consistency judgment．

High Efficiency Video Coding (HEVC/H.265) is a widely used video coding standard in the international market. As the core encoding method of HEVC video encoding, Context Adaptive Binary Arithmetic Coding (CABAC) can improve the compression efficiency of arithmetic coding by establishing a more accurate probability model. Moreover, HEVC defines a larger variety of syntax elements and establishes more complex coding structures, further reducing information redundancy and thus reducing the bit rate. However, as the input data to CABAC, syntax elements’ high complexity of preprocessing process increases the difficulty of hardware parallel processing. As a result, the throughput rate of entropy coding hardware is difficult to improve, which becomes one of the bottlenecks for HEVC encoder to achieve higher resolution real-time coding. To further speed up the entropy encoding modules, this study designed a high-throughput CABAC entropy encoding architecture based on FPGA. Within the architecture, the pre-header information coding, pre-initialization and coding unit (CU) are able to accelerate the generation of syntax elements, which is dedicated to CABAC. Due to the scheme of efficient residual coding and partial context index pipeline computing, the reduction of path latency and the improvement of operating frequency can be achieved as well as high throughput. In this study, the proposed design, which is synthesized by using a 90 nm standard cell library, occupies a total of 2.099×10⁴ logic gates and operates in the frequency of 200 MHz. This paper also simulated the video sequence provided by HEVC official, and counted the time required for encoding a coding tree unit (CTU) under different quantitative parameters (QP). The experimental statistics show that the time of encoding a CTU was saved by 38.2% on average.

For low-dimensional multi-objective optimization problems, the existing multi-objective optimization algorithms have been able to ensure the proximity to the optimal front of the problem, and balance the convergence and the diversity of solution sets. However, the uniformity of the solution is ignored in most algorithms. In the multi-objective optimization problem with irregular Pareto front, the more uniform the distribution of solution is, the more the solution can reflect the true distribution of the optimal front of the problem, and the more reasonable the choices provided to decision makers. To improve the uniform distribution of solutions, a new multi-objective optimization algorithm CM-SPEA2 is proposed based on SPEA2 algorithm and the improved individual fitness calculation. In this algorithm, firstly, the initial population is divided into different clusters by means of hierarchical clustering. Next, the original calculation method of messy degree is improved to measure the messy degree of individuals in their clusters, and the individuals with the lowest messy degree are selected as reference points. Then, based on the Manhattan distance between other individuals and the reference point, the operator representing distribution is calculated and the fitness function is improved. Finally, the fitness threshold is set to screen non-dominated individuals near the reference point, so as to indirectly adjust the environmental selection strategy, make the distribution of retained individuals more uniform, thus improving the convergence and diversity. As compared with some similar multi-objective optimization algorithms, the proposed CM-SPEA2 algorithm has certain advantages in solving IMOP, ZDT and VNT test problems.

In scientific experiments and industrial production, the dynamic characteristics of the force sensor will directly affect the accuracy, so it is of great significance to research the dynamic characteristics of the force sensor. Aiming at the practical problem that the dynamic characteristics of strain gauge force sensor used in surgical robots are difficult to meet the accuracy requirements, this paper studied the application of least square parameter identification method in the vibration structure of force sensor. Because recursive least squares (RLS) is difficult to ensure the rapidity and anti-interference of the second order vibration system model identification, therefore, this paper proposed a recursive least squares parameter identification method based on variable forgetting factor. Firstly, the parameters of the forgetting factor function were determined by establishing the random vibration system model, simulating and analyzing the input/output characteristics of the system. The simulation results show that the proposed method in the paper can significantly reduce the parameter identification error and convergence prediction error compared with RLS while maintaining a faster convergence speed, and has better time variability compared with the least squares. Furthermore, the dynamic parameters of the force sensor used in minimally invasive surgical robot were identified based on the step test calibration method to obtain the structural dynamic characteristics (i.e. natural frequency and damping ratio) of the sensor system. The experimental results show that the proposed method in the paper has good convergence and stability, and can effectively improve the identification accuracy.

In order to improve the utilization of computing or communication resources and reduce energy consumption, the study proposed the event-triggered impulsive observer-based output feedback control method for a class of nonlinear systems with aperiodic sampling and stochastic measurement noises. Firstly, by introducing an event-triggering mechanism that only depends on the discrete-time measurement output and an auxiliary variable, it designed a novel event-triggered impulsive observer. Then by constructing an augmented system composed of the original system and the observer error system and developing the quasi-periodic discretized Lyapunov function method, it established an ultimate bounded stability criterion in the mean square sense of the augmented closed-loop systems. The criterion reveals the influence mechanism of the sampling period, noise intensity, and event trigger parameters on system performance. Next, combined with the joint design approach, the output feedback controller synthesis problem was transformed into solving a set of LMIs based on the augmented system, thus solving the difficult problem that the state feedback gain and the observer gain cannot be separated in the presents of the stochastic measurement noise. Finally, on the Matlab platform, the performance of the proposed control method was analyzed via a connecting rod robotic arm. The experimental results demonstrate that the proposed method is effective in reducing the number of transmissions and conserving communication/computing resources. Furthermore, it successfully addresses the stabilization problems of nonlinear systems with stochastic measurement noise, thus confirming the effectiveness of the proposed approach.

Auxiliary resonant transform pole inverter (ARCP) has the advantages of high efficiency and low loss in high frequency and high power field. However, in practical application, the problem of neutral point voltage deviation will lead to the increase of failure rate of ARCP soft switch. In this paper, a neutral point voltage active disturbance rejection control strategy was proposed to solve the problem of neutral point voltage deviation of ARCP bus divider capacitor and the failure of zero voltage opening of main switch tube caused by neutral point voltage deviation. The influence of bus capacitor equivalent series resistance (ESR) on neutral point voltage and AC side voltage was analyzed, and a neutral point voltage model including bus capacitor ESR in variable sequential control mode was established. On this basis, the neutral point voltage active disturbance rejection controller was designed and the auxiliary tube was used to control the overcharge regulation time, which realize the current balance of the three-phase auxiliary branch resonant inductor, achieving the goal of neutral point voltage stability, zero voltage opening of the main switching tube and improving the efficiency of the inverter. Compared with 97.0% inverter efficiency in PI control mode, active disturbance rejection control mode can achieve 97.5% inverter efficiency at maximum power. The control strategy is easy to implement as it doesn’t need to change the circuit structure. The effectiveness of the proposed control strategy was verified by simulation and experimental results.

Dual-mode ultrasound is widely used in medical clinical diagnosis. The B-mode pulse is used for imaging and Doppler pulse is used for blood flow velocity estimation. The data collection time is shared between the two modes. To improve the update frequency of B-mode image, it is necessary to reduce the number of Doppler pulses, that is, to estimate the blood flow velocity by sparse Doppler emissions. However, the existing algorithms for sparse pulse sampling, such as iterative adaptive algorithm, sparse Bayesian algorithm and subspace method based on array virtual expansion, are huge in expense and can not meet the requirements of real-time imaging. What’s more, they will lead to obvious artifacts in the case of large sparsity. Therefore, this paper proposed a low complexity blood flow velocity estimation algorithm via sparse pulse sampling. Based on the fact that ultrasonic Doppler echo signal is generated by the scattering of red blood cells, so echoes are strong coherence signals with time-variation sources number, this paper firstly explained the cause of artifacts from the perspective of subspace, and verified that the sparse emission pulse arrangement with uniform pulse can effectively suppress artifacts. Then the covariance matrix was constructed with uniform pulse echo, and the eigenvalues were obtained after spatial smoothing. The frequency distribution characteristics of blood flow at different segments were derived by the number of larger eigenvalues and the ratio of each other. Finally, based on the frequency distribution characteristics, the B-MUSIC algorithm or TBVAM algorithm was adaptively used for blood flow velocity estimation to reduce the complexity of the algorithm. The experimental results with Matlab simulation and human body measurement data show that the algorithm can obtain continuous, clear blood flow velocity estimation results with well artifact suppression while reducing the computational complexity significantly.

The available bandwidth of 5G networks is a pivotal factor influencing real-time video services. However, how to achieve accurate predictions in the context of low-latency real-time video services remains a difficult problem. Conventional algorithms for predicting available bandwidth typically rely on data metrics at the application layer and complete the forecast according to the packet transmission strategies. Such algorithms can lead to prediction lag in complex scenarios, thereby significantly impairing the received video quality for users. To address this concern, this study proposed a novel available bandwidth prediction algorithm based on cross-layer multi-dimensional parameters. This algorithm comprehensively integrates pertinent data metrics from the application, physical, and network layers, utilizes multiple dimensions of parameters to enhance the precision of wireless network bandwidth detection. In this paper, deep reinforcement learning was adopted as the model framework to integrate offline prediction and online prediction through cross-layer and multi-dimensional data model learning for different motion scenarios. Furthermore, network packet loss rates, image quality assessments, end-to-end delays, and other link-related factors were introduced as constraints, to realize the real-time adjustment and optimization of the prediction model during the transmission process. Experimental results conducted on a semi-physical platform show that: the prediction performance of the proposed algorithm is better than the traditional prediction method, and the fitting degree of the prediction curve and the actual curve is more than 95.8%; compared with the single-layer prediction algorithm, the packet loss rate of the proposed algorithm in walking and driving scenarios decreases by 47.3% and 30.9%, respectively, and the quality of received video is improved by 12%.

In recent years, the resolution and frame rate of video have been continuously improved to meet people’s increasing demand for video data. However, the compression encoding speed of real-time video sequence is often restricted by frame rate and resolution. The higher the frame rate and resolution are, the longer the encoding time will be. In order to achieve real-time compression encode for video sequences with higher resolution and frame rate, this paper designed a new parallel pipeline hardware architecture of intra rate-distortion optimization prediction mode, which supports intra prediction coding of up to 64×64 coding tree unit. Firstly, a parallel scheme with 9-way prediction mode was designed. Secondly, a pipeline hardware architecture was implemented based on a 4×4 block as the basic processing unit in a Z-shaped scanning order, and the prediction data of 32×32 prediction units were reused to replace the prediction data of 64×64 prediction units so as to reduce the amount of calculation. Lastly, a new Hadamard transform circuit was proposed based on this pipelined architecture for efficient pipelined processing. The experimental results show that: on the Altera Arria 10 series field programmable gate array, the 9-way mode parallel architecture only occupies 75 kb look up table and 55 kb register resources, the main frequency can reach 207 MHz, and it only takes 4 096 clocks cycles to complete a 64×64 coding tree unit prediction and can support real-time encoding of 1 080 P resolution 99 f/s full I-frame at most. Compared with the existing design scheme, the scheme designed in this paper can realize higher frame rate 1 080 P real time video encoding with smaller circuit area.