67results about How to "Reduce computing delay" patented technology

The invention provides an on-ramp vehicle merging management method and a system. The method includes the following steps: information of vehicles to enter a predetermined merging information collection point on a main road and a ramp is collected; the time of a vehicle from the current position to the predetermined merging information collection point is estimated according to the collected vehicle information, and the traffic order of the vehicle is numbered according to a time sequence; the traffic sequence number of the vehicle is issued so that the vehicle passes through the predeterminedmerging information collection point in sequence. The on-ramp vehicle merging management method and the system of the invention provide a more simplified mergence processing method for lane merging management; it is conducive to shortening the calculation delay; the division of responsibilities between roadside equipment and vehicle ends can be clearly defined; and it is easier to identify the accident responsibility party.

The invention discloses a task migration method in an edge computing platform, belongs to the technical field of edge computing, and is used for reducing energy consumption expenditure, communicationexpenditure, migration expenditure and comprehensive expenditure of the energy consumption expenditure, the communication expenditure and the migration expenditure during task migration in the edge computing platform. The method comprises the steps of dividing each server node of a physical network topology into a hotspot node, a warmspot node and a coldspot node according to the CPU resource utilization rate; preprocessing the physical network topology; respectively carrying out short-term resource utilization rate prediction on the hotspot node and the coldspot node; and respectively carrying out task migration on the hotspot node and the coldspot node which need to be subjected to task migration, and obtaining the position of the server where each migrated task is located. The method isused for eliminating servers exceeding the upper limit of the resource utilization rate and lower than the lower limit of the resource utilization rate in the edge computing platform, effectively reducing the energy consumption overhead, the computing time delay and the data transmission pressure of the system, fully utilizing the storage and computing capabilities of the edge servers and relieving the pressure of a cloud computing center.

The invention belongs to the field of wireless communication, and discloses an LMS algorithm processing time-delay sensitive based delayed decision feedback equalization method and system. The systemapplies a variable step LMS algorithm based decision feedback equalizer to the receiving processing of a disturbed signal. As a wireless communication channel exists a time-varying property, the convergence speed of LMS needs to be further accelerated to make the equalizer track channel changes better and in real time. The variable step LMS algorithm adopts a larger iterative step to accelerate the convergence speed during an initial working phase, and adopts a smaller iterative step to reduce steady-state error during a convergence phase, so that an original signal can be effectively recovered. The method and system have advantages of fast in convergence speed, less in steady-state error, little in processing time-delay and easy for realization.

The invention discloses a large-scale access method for an edge intelligent network. A cell deploys a multi-antenna base station as an edge server, and a large number of mobile devices are accessed to a wireless network through the base station for federated learning. In each round of learning, a base station firstly selects part of equipment, and a transmitting beam broadcast global model is designed through a downlink. After each selected device recovers the global model through the receiver, a new local model is trained based on the local data set, and then the new local model is sent out after beam forming. The superposition characteristic of a wireless channel is utilized, through a receiver, a base station can directly obtain an aggregation model which is calculated in the air, and then the weighted average of the aggregation model is calculated to serve as an updated global model. The invention provides an effective large-scale access method for the edge intelligent network.

The invention discloses a reading voltage optimization method of a memory cell. The memory cell is selected from a plurality of memory cells corresponding to a selected physical page. The read voltage optimization method comprises the following steps: counting the number of memory cells with a first threshold voltage to obtain a first number; counting the number of the memory cells with the second threshold voltage to obtain a second number; setting an optimal reading voltage based on the first number, the second number and the adjustment parameter, wherein different adjusting parameters are set according to different data storage time, read interference, cross temperature and programming / erasing times. According to the reading voltage optimization method of the memory cell, the controller of the 3D memory and the operation method of the controller, when the optimal reading voltage is set, only the limited number of times of reading is needed, the time delay for determining the optimal reading voltage is effectively reduced, and the adjustment parameters are set based on parameters such as reading interference and programming / erasing times, and the accuracy and reliability of products are improved.

The invention discloses a large-scale access method integrating calculation and communication. A multi-antenna base station is arranged in the center of a cell, and a large number of mobile terminalsare accessed to a wireless network through the base station. The mobile terminal performs beam forming on the communication signal and the calculation signal to be transmitted respectively, and then transmits the communication signal and the calculation signal after superposition coding. On one hand, the base station can directly receive the summation function obtained through air calculation by utilizing the superposition characteristic of the wireless channel, and then recover the target function through calculating the receiver. And on the other hand, the base station decodes the communication signal of each mobile terminal through the communication receiver. The invention provides an effective calculation and communication fusion large-scale access method for the Internet of Things with a large-scale mobile terminal.

The embodiment of the invention provides a mobile web deep learning cooperation method and system. The method comprises the steps of loading a target task through a mobile web browser, preprocessing the target task, and sending a deep learning calculation request to an edge server; the edge server calculating a compressed deep neural network model demand according to the deep learning calculationrequest and the current context information, and returning a satisfied compressed deep neural network model to the mobile web browser; the mobile web browser receiving and executing a feedforward process of compressing the deep neural network to obtain a first task calculation result, and taking the first convolutional layer as a temporary storage result of the shared layer; the mobile web browsercalculating the standard cross entropy of the first task calculation result, and if it is judged that the standard cross entropy is larger than a preset threshold value, receiving the first task calculation result. According to the mobile Web deep learning cooperation provided by the embodiment of the invention, the transmission and feedforward calculation time delay of the model is effectively reduced.

The invention provides a click rate estimation system and method based on a field-programmable gate array. The click rate estimation system based on the field-programmable gate array comprises a datainterface module, a model parameter storage module, a calculation control module, a factorization machine calculation module and a neural network calculation module, wherein the data interface moduleis used for acquiring model parameters, acquiring to-be-estimated data and outputting a click rate estimation result; the model parameter storage module is used for storing model parameters; the calculation control module is used for obtaining to-be-estimated data from the data interface module and obtaining model parameters from the model parameter storage module; the factorization machine calculation module is used for calculating a factorization machine expression of the to-be-estimated data; the neural network calculation module is used for calculating a neural network forward propagationresult of the to-be-estimated data; and the calculation control module is also used for obtaining a click rate estimation result. According to the technical scheme of the embodiment of the invention,the calculation delay of click rate estimation is reduced, and the calculation speed is improved.