373results about How to "Shorten convergence time" patented technology

Generalized learning rules may be implemented. A framework may be used to enable adaptive signal processing system to flexibly combine different learning rules (supervised, unsupervised, reinforcement learning) with different methods (online or batch learning). The generalized learning framework may employ non-associative transform of time-averaged performance function as the learning measure, thereby enabling modular architecture where learning tasks are separated from control tasks, so that changes in one of the modules do not necessitate changes within the other. The use of non-associative transformations, when employed in conjunction with gradient optimization methods, does not bias the performance function gradient, on a long-term averaging scale and may advantageously enable stochastic drift thereby facilitating exploration leading to faster convergence of learning process. When applied to spiking learning networks, transforming the performance function using a constant term, may lead to non-associative increase of synaptic connection efficacy thereby providing additional exploration mechanisms.

A method of constructing a backup path in an autonomous system (AS) for failure of an inter-AS link is described. The method comprises identifying an alternate inter-AS path and constructing a tunnel to an end point on the alternate path.

A system and method for decreasing the route convergence time in a wireless communication network, such as a wireless ad-hoc peer-to-peer network, by finding an alternate route if the nodes anticipate weakening or breakage of a route currently in use. The system and method which enables reactive routing protocols to find optimal routes between nodes in these types of networks when those routes cannot otherwise be found in certain conditions. The system and method thus decrease the route convergence time, provide an effective and efficient way to find optimal routes, and improve overall performance of the network with regard to throughput, delay, packet completion rate and other factors.

The present invention is intended to attain simplified circuit configuration and low current consumption in a discrete time amplifier circuit and an AD converter, to improve the convergence from the transient response state to the steady state of the amplifier circuit and to reduce noise and distortion owing to the variation in the output common-mode voltage. The discrete time amplifier circuit and the AD converter are provided with a switched-capacitor common-mode feedback (CMFB) circuit capable of detecting and feeding back the output common-mode voltage at every sampling timing in the case that the circuit operates at double sampling timing (every ½ cycle).

Some embodiments provide a controller for managing several managed switching elements that forward data in a network. The controller includes an interface for receiving input logical control plane data in terms of input events data. The controller includes an input scheduler for (1) categorizing the input events data into different groups based on certain criteria and (2) scheduling supplying of the input event data into a converter based on the groups so that the converter processes a group of input events data together. The controller includes the converter for converting the input logical control plane data to output logical forwarding plane data. The controller includes a network information base (NIB) data structure module for storing the output logical forwarding plane data. The logical forwarding plane data is for subsequent translation into physical control plane data.

The invention discloses a route exchange method suitable for a GEO/LEO double layered constellation network. According to the method, a system period is divided into equal-duration time sections based on a dynamic boundary value; a GEO layer satellite high in on-board processing capacity is used for calculating the best route and the second best route for each LEO satellite; in the processes of information transmission and exchange, when loads of the LED satellites are large, the GEO satellite shares part of low priority services in time for the LED satellites, and it is guaranteed that important information is reliably transmitted in real time; when link congestion, node ineffectiveness and other emergency conditions happen in the satellite network, in order to avoid rerouting of the whole network, and the GEO satellite only calculates rerouting for affected routes; after link congestion is eliminated, the LED satellites recover route information in time before congestion in order to avoid link resource waste in the network. In the network that topology time varying happens, links are prone to congestion, on-board resources are limited, continuous high-load flows are prone to being generated, and nodes are prone to being ineffective at the special period, the method can reduce constellation system cost, shorten convergence time, save the on-board resources, increase the utilization rate of the link resources, guarantee that important information is reliably transmitted in real time and improve invulnerability and robustness of the satellite network.

A system for providing a tree topology for a network having an interior gateway protocol. A first router receives a hello message from all connected routers in the network. The hello messages include tree topology information. The first router then uses the tree topology information to determine a parent of the router. The first router then establishes connections with directly connected routers at the same level in the tree topology. The first router also generates link messages that include all of the prefixes for children of the first router and broadcasts the link messages.

The present invention discloses a feature extraction and state recognition method for one-dimensional physiological signal based on depth learning. The method comprises: establishing a feature extraction and state recognition analysis model DBN of a on-dimensional physiological signal based on depth learning, wherein the DBN model adopts a "pre-training+fine-tuning" training process, and in a pre-training stage, a first RBM is trained firstly and then a well-trained node is used as an input of a second RBM, and then the second RBM is trained, and so forth; and after training of all RBMs is finished, using a BP algorithm to fin-tune a network, and finally inputting an eigenvector output by the DBN into a Softmax classifier, and determining a state of an individual that is incorporated into the one-dimensional physiological signal. The method provided by the present invention effectively solves the problem that in the conventional one-dimensional physiological signal classification process, feature inputs need to be selected manually so that classification precision is low; and through non-linear mapping of the deep confidence network, highly-separable features/feature combinations are automatically obtained for classification, and a better classification effect can be obtained by keeping optimizing the structure of the network.

The invention relates to a global navigation satellite system (GNSS) triple-frequency motion-to-motion positioning method. In the original epoch, the triple-frequency precision single-point positioning technology is adopted to obtain the coordinates of two movable carriers, one of the two movable carries is selected as the reference station for the triple-frequency double-difference position, the triple-frequency double-difference positioning technology is adopted to calculate the baseline component of the two movable carriers and the integer ambiguity resolution of the double-difference carrier phase, and the baseline component of the two movable carriers and the integer ambiguity resolution of the double-difference carrier phase are used as the constraint conditions for the triple-frequency precision single-point positioning in the subsequent epochs to improve the single-point position precision and the convergence speed thereof. The geometry-Base TCAR ambiguity resolution fixation method is adopted to calculate the triple-frequency integer ambiguity resolution. The three irrelevant combination observation values of the triple-frequency non-ionizing layer and the long wave-length and low noise carrier are adopted to detect and repair the cycle slip of the original observation data.

A management and orchestration server which manages a plurality of types of virtualized nodes included in server group includes a scaling executing unit configured to command the server group to scale virtual machines included in each of the plurality of types of virtualized nodes, a traffic load measuring unit configured to manage a traffic load of each of the virtual machines, and a scaling amount determining unit configured to determine a number of the virtual machines to be scaled. In this case, the traffic load measuring unit acquires a traffic load for each of the traffic types to be processed by each of the plurality of types of virtualized nodes, and the scaling amount determining unit determines the type of virtualized node for which virtual machines are to be scaled and a number of machines to be scaled based on the traffic loads of the traffic types.

An embedded pattern-based adaptive equalizer coefficient updating mechanism updates weighting coefficients of an adaptive equalizer in accordance with an error or differential signal obtained by processing the distorted version of a known signal pattern, that is repetitively distributed among information signals contained within the electrical communication signal output from the equalizer, with a copy of the known signal pattern. No dedicated ‘training sequence’ communication with the transmitter is required, so that the coefficient update mechanism does not impact the operation or bandwidth availability of the channel.

The invention discloses a time domain bidirectional iteration-based turbine vane flutter stress forecasting method. The method is characterized in that: a set of bidirectional iteration method for a time domain is designed by taking a vane and a surrounding flow field thereof as a three-dimensional fluid solid coupling system, and the flutter stress of the vane is obtained by alternately solving vane deformation and unsteady flow field. The method comprises the following steps of: setting a structural calculation module, a fluid calculation module, a data conversion module, a flutter stress output module, an initial value calculation module and a bidirectional iteration module in a computer; acquiring static vane deformation and steady flow field serving as an initial value by nonlinear iteration; alternately calling the structural calculation module and the fluid calculation module to propel the whole system on time; transmitting fluid solid boundary information through the data conversion module; and outputting the flutter stress on time history. The method realizes integrated calculation of the vane and the flow field, takes the nonlinearity of the coupling system into consideration, and can allow the observation of the whole flutter development process and forecast the flutter stress of the vane.

The embodiment of the invention provides an Internet of Things service low-delay load distribution method and device based on edge computing. The Internet of Things service low-delay load distributionmethod comprises the steps: acquiring a task request of each application in each terminal and the computing capacity of each edge node; and inputting the task request of each application in each terminal and the computing capacity of each edge node into a preset optimization problem model, and outputting a resource allocation matrix and a task allocation matrix, wherein the preset optimization problem model comprises a particle swarm algorithm model improved through an ant colony algorithm and a semi-definite relaxation algorithm model improved through a Gaussian random algorithm. The Internet of Things service low-delay load distribution method and device based on edge computing improve particle swarm optimization by applying the ant colony algorithm, reduce the convergence time of the algorithm, improve the performance of the resource allocation result, solve the rank 1 constraint in the semi-definite relaxation problem by applying the Gaussian random variable, improve the performance of the task allocation result, and finally reduce the service delay.

A method and apparatus of multicast traffic redundancy protection is provided. After a fault occurs, the VRRP selects a new active router and notifies the PIM routing protocol, and the PIM uses the new active router as its DR.

The invention discloses a method for controlling steering stability of a distributed-drive electric vehicle. The method comprises the following steps of: step 1: according to signals detected by a vehicle speed sensor and a steering wheel angle sensor, obtaining a longitudinal vehicle speed u and a forewheel steering angle delta of a running vehicle through calculation of a state observation module; step 2: according to the longitudinal vehicle speed u and the forewheel steering angle delta obtained in the step 1, and obtaining a reference yaw rate omega rd of the running vehicle by a linear two-degree-of-freedom vehicle dynamic model; step 3: detecting an actual yaw rate omega r of the vehicle by using a gyroscope; step 4: according to the reference yaw rate omega rd and the actual yaw rate omega r, designing a steering stability controller and obtaining a yaw moment required by the steering stability control of the vehicle; and step 5: using the minimum tire utilization rate as an optimization objective, designing a distribution function and obtaining a driving torque Ti of four wheel hub motors. The phenomenon of ''chattering'' of a system is effectively suppressed, robustness is improved, and the steering stability of the distributed-drive electric vehicle is well controlled.

The invention discloses an automatic gain control method and an automatic gain control circuit which are suitable for an orthogonal frequency division multiplexing (OFM) system. The method comprises the following steps of: measuring a baseband signal; and adjusting gain. Slide window processing is performed during signal measuring so as to average signal energy in a window. A gain adjustment algorithm determines reasonable gain of the next adjustment period through conditions such as the gain of an upper frame, a measured value of the upper frame, the number of overflow sampling points, the receiving signal intensity indication on the current antenna and the like. Simultaneously, the gain of a radio frequency receiving chip is changed by circulating prefix time so as to track an OFDM symbol level.

The invention relates to a sensorless brushless DC motor commutation deviation fast correction control system and a method. An analytical equation between the bus current difference before and after commutation and commutation deviation is derived according to the bus current fluctuation value before and after commutation. In view of the influence of commutation ripple and motor three-phase winding asymmetry on the accuracy of sampled current during actual bus current sampling, a method for current offset sampling in the same conduction zone is designed. The bus current difference is obtained according to the designed current sampling method, the commutation deviation is calculated based on the analytical equation, and the commutation signal is compensated and corrected in real time. By designing the sensorless motor commutation deviation fast correction control system, the commutation deviation convergence time is reduced greatly, and the commutation precision is improved.

An apparatus and method for recovering a carrier frequency of a received signal using a predefined symbol in an OFDM system is provided. A maximum value detector detects at least three values and their corresponding offsets in descending order among auto-correlation values calculated within a specific frequency offset range between a received signal and a predefined symbol. A frequency offset corrector compares a ratio of a first auto-correlation value to a second auto-correlation value with a predetermined first threshold, determines whether first and second frequency offsets corresponding to the first and second auto-correlation values are adjacent to each other if the ratio of the auto-correlation values is less than the first threshold, and determines an initial carrier frequency offset using a difference between the first frequency offset and a frequency offset correction value determined according to system, if the first and second frequency offsets are adjacent to each other and a ratio of the second auto-correlation value and a third auto-correlation value is greater than a predetermined second threshold.

The invention provides a parameter sound source modeling method based on an improved BP (Back Propagation) neural network, in order to solve the problem of difficulty in modeling a parameter sound source system at present. The method comprises the following steps: firstly, collecting sufficient training and testing sample data and preprocessing the data; establishing a neural network model; adopting a genetic algorithm for performing optimizing process on the structure and parameter of the neural network model; searching for an initial weight value and a threshold value between the number of better neural network hidden layers and neural cells; and lastly, training and testing an established parameter sound source model based on the improved BP neural network by using the sample data. The model has the advantages of reliability and higher assessing precision.

Distributed data processing of problems representing resource assignment tasks. The problems are modeled as programs, and the programs are partitioned into sub-instances. Those sub-instances are executed in a distributed computing environment. The partitioning reduces communication costs between sub-instances and convergence time for the optimization program.

A method is set forth for reducing the total acoustic echo cancellation convergence time for all look directions in a microphone array based full-duplex system. The method is based on capturing the loudspeaker signal due to the first far-end speech bursts when the conferencing system is first used, as well as the corresponding loudspeaker feedback signals in the individual microphones. The captured signals are then used for consecutive adaptation of the acoustic echo canceller on all echo paths corresponding to all look directions of the beamformer, thereby training the AEC. This training process can be executed concurrently with normal phone operation, for example, as a background process that utilizes available processing cycles.

The invention provides a method for route convergence and routing device. The described method includes: receive the route information of multiple adjacent routing devices, select the route from the received route message, and save the route selection results into the first route information library; identify that all route messages of multiple adjacent routing devices are received and the route selection is completed; update the second route information library according to the information of the first route information library and the second route information library is sued to guide the forwarding of the data packets. The implementation of the invention reduces the convergence time of the route and improves the convergence speed of the route. Especially when the route volume is higher, it will spend much time in selecting the route, so the implementation of the invention will reduce the convergence time effect of this route more effectively.