13 results about "Round-off error" patented technology

The present invention discloses a robot calibration method for LM and ARO algorithms, comprising step S1, establishing a robot kinematic error model using an MDH model; step S2, obtaining robot joint data and laser tracker measurement data of sampling points; step S3, calculating the distance values ​​between two points corresponding to the robot end points, and establishing an objective function through the distance difference between the two points; step S4: performing parameter iterative optimization on the objective function of step S3 using the LM‑ARO algorithm to obtain a robot kinematic parameter error value that satisfies the minimum objective function value. The present invention solves the rounding error of high-order terms of the single LM algorithm, and also overcomes the problem of slow search speed and low efficiency caused by the large search range of the single ARO algorithm. No constraints of standard parts are required. At the same time, the calibration cost is low, and the robot kinematic parameter error can be effectively calibrated, significantly improving the accuracy level of the robot.

This disclosure proposes a solution for improving rational number discontinuous receive (DRX) cycles. Specifically, a network device transmits a discontinuous receive configuration to a terminal device that indicates at least one of a discontinuous receive DRX cycle or the start time offset of said DRX cycle. The terminal device determines a target start time offset relative to the on-period of the DRX cycle without rounding error, based on a predetermined formula. When the DRX cycle is expressed as a rational number, the target start time offset is guaranteed to match the start time offset of the DRX cycle. The terminal device performs discontinuous receive based on the determined target start time offset. In this way, since there is no rounding error when the DRX cycle is expressed as a rational number, the transmission delay is reduced.

The application provides a radionuclide activity calculation method based on a Padé approximation considering multistage decay, which comprises the following steps: S1, constructing a complete decay matrix containing all nuclides; step S1 comprises the following steps: S 11 , receiving a decay chain data structure as input, constructing an n*n square matrix A; S 12 , initializing all elements of the entire matrix to zero values; S 13 , setting the element of the ith row and the ith column to the negative value of the decay constant of the ith nuclide, that is, A(i,i)=; S 14 , traversing all possible nuclide pairs (i,j); S2, calculating the matrix exponential by using a matrix norm driven adaptive scaling algorithm combined with a Padé approximation matrix; S3, performing corresponding radionuclide activity calculation according to the calculation mode. The application avoids truncation error by constructing a complete decay matrix and considering the contribution of all nuclides. The adaptive scaling algorithm combined with the Padé approximation matrix is used to solve the rigid system matrix calculation, which effectively avoids numerical overflow and rounding error, and guarantees the stability of the calculation process.

The present disclosure relates to a solution to improve rational discontinuous reception (DRX) cycles. Specifically, a network device transmits a discontinuous reception configuration to a terminal device, the discontinuous reception configuration indicating at least one of the following items: a discontinuous reception DRX cycle or a start time offset of the DRX cycle. The terminal device determines a target start time offset for an activation duration of a DRX cycle without a rounding error based on a predefined formula. When the DRX cycle is represented as a rational number, the target start time offset is ensured to match the start time offset of the DRX cycle. The terminal device performs discontinuous reception based on the determined target start time offset. In this manner, when the DRX cycle is represented as a rational number, there is no rounding error, thereby reducing the delay of the transmission.

The present invention describes systems and methods for video encoding. In some embodiments, inter-frame prediction of samples in a current block is performed by rounding an initial motion vector and determining a rounding error vector caused by the rounding. An unrefined prediction for the sample is generated using the rounded motion vector. Unrefined predictions are similarly generated for other samples in the current block. Based on the unrefined prediction, a spatial gradient is determined for each sample position in the block. A refined prediction is generated for each sample position by adding a scalar product between the spatial gradient at the sample position and the rounding error vector to the unrefined prediction. The exemplary method can reduce the number of reference pixels used to predict the current block and, therefore, can reduce memory access bandwidth.

The invention provides a radionuclide activity calculation method considering multistage decay based on Pade approximation. The radionuclide activity calculation method comprises the steps of S1, constructing a complete decay matrix containing all nuclides; the step S1 comprises the following steps: S11, receiving a decay chain data structure as input, and constructing an n * n square matrix A; s12, initializing all elements of the whole matrix into zero values; s13, setting elements in the ith row and the ith column as a negative value of the decay constant of the ith nuclide, namely A (i, i) =; s14, traversing all possible nuclide pairs (i, j); s2, calculating a matrix index by adopting a matrix norm driven adaptive scaling algorithm in combination with a Pade approximation matrix; and S3, executing corresponding radionuclide activity calculation according to the calculation mode. According to the method, the complete decay matrix is constructed, contribution of all nuclides is considered, and truncation errors are avoided. A self-adaptive scaling algorithm is combined with a Pade approximation matrix solution to realize rigid system matrix calculation, so that numerical value overflow and rounding errors are effectively avoided, and the stability of a calculation process is ensured.

The application belongs to the field of high-performance computing, and more particularly, a large eddy simulation large-scale parallelization and mixed precision optimization method for a domestic high-performance computing system is designed. Facing the domestic high-performance computing platform supporting multiple floating point precisions, a multi-level parallel optimization strategy is adopted, and the many-core accelerator architecture characteristics of the domestic high-performance computer are deeply integrated. The core scheme includes a multi-level parallel and optimization strategy for the domestic supercomputer general architecture, and an efficient collaborative computing scheme implemented on the CPU+DSP heterogeneous platform. The application faces the Tianhe new generation high-performance computing platform supporting multiple floating point precisions, and according to the numerical calculation rounding error accumulation characteristics in the application field, efficient parallel mixed precision calculation is realized through the programming environment, the memory access amount and the communication amount are effectively reduced, and the large eddy simulation calculation efficiency is improved, thereby providing strong technical support for the domestic transplantation and performance optimization of large-scale numerical simulation.

The invention relates to a quantized neural network hardware reasoning simulator and a method, hardware reasoning is simulated through software, a software deployment process is optimized, namely hardware calculation is completely simulated in the software, a hardware reasoning result with an accurate bit is achieved, reasoning results of different quantized schemes of a deep learning model are quickly verified, and hardware data types are simulated, so that the hardware reasoning accuracy is improved. The method comprises the following steps of: simulating low-precision data types on hardware in software to ensure that the data types are consistent, and finally simulating hardware calculation behaviors, namely simulating calculation details such as overflow processing, rounding error and saturation operation on the hardware to ensure that the calculation behaviors are consistent, and finally effectively reducing the reasoning difference between software and hardware in quantitative reasoning. The reliability of deep learning model deployment is improved.

Filter method for converting an analog input signal into a sampled digital output signal, wherein the z-transformed digital output signal Y(z) results from the z-transformed analog input signal X(z) and the z-transformed rounding error E(z) at order k as Y(z) = X(z) · z -1 + E(z) · (1- z -1 ) k , where Y r (z) = E(z) · (1-z -1 ) k the z-transform of the noise signal is characterized in that - the filter gain at the frequency limit approaches 0 to 1, - for an impulse response h i of the filter at time i for a filter length N: ∑ i = 0 N − 1 h i = 1, where the filter is set up such that the ratio of the variance of the noise signal is var(Y r ) to the variance of the rounding error var(E) yields: var ( Y r ) var ( E ) = h 0 2 + ( h 1 − h 0 ) 2 + ( h 2 − h 1 ) 2 + ⋯ + ( h N − 2 − h N − 1 ) 2 + h N − 1 2 and the variance of the noise signal var(Y r ) is minimized, where the filtering method is an infinite impulse response method.

While the symbol occurrence count is rounded to generate the symbol frequency, the lossless data compressor immediately prevents normalization overflow, allowing the code table generator to generate code table entries without waiting for the symbol frequency table to finish filling. As symbols are normalized, rounding errors accumulate, and are compensated for by reducing the symbol frequency when the symbol frequency is at least 2 and the accumulated error exceeds a threshold. The symbol frequency is also reduced when the number of remaining states in the code table is insufficient to accommodate the number of remaining unprocessed symbols and states for the current code table entry. Because error compensation occurs during symbol normalization, it does not force all symbols in the waiting block to be processed before generating the code table, thus reducing latency. Three pipeline stages operate on three input blocks: symbol counting, normalization / error compensation / code table generation, and data encoding.

The application provides a target location map element determination method and device and electronic equipment, and relates to the technical field of autonomous driving. The method comprises the following steps: firstly, a plurality of polygon map elements are acquired; secondly, the plurality of polygon map elements are split to obtain a plurality of triangular sub-elements; the vertices of the triangular sub-elements belong to the vertices of the polygon map elements; finally, based on the relative position relationship between the target and the plurality of triangular sub-elements, the polygon map element in which the target is located is determined. In the embodiment, the polygon map element is split into a plurality of triangular sub-elements, and the polygon map element in which the target is located is determined based on the relative position relationship between the target and the plurality of triangular sub-elements. Since the relative position relationship does not need to be determined based on specific numerical values when the relative position relationship between the target and the plurality of triangular sub-elements is calculated, the precision loss or rounding error caused by floating-point calculation can be avoided, and the accuracy of the calculation result is improved.

An adaptive quantization method considering a distance range for LiDAR-based point cloud compression is provided. According to an embodiment of the present invention, the adaptive quantization method classifies point cloud data on the basis of a distance from an origin, and quantizes the classified point cloud data using different quantization schemes. Accordingly, by performing adaptive quantization that takes the distance range of point cloud data into consideration, rounding errors and clipping errors during quantization are significantly reduced, thereby enabling data compression and integer conversion while minimizing data loss and accuracy degradation.

A method for generating an output binary word from an input binary word, including: executing a first floating-point multiplication wherein the input binary word interpreted as an integer number is multiplied with a first predetermined floating-point number, resulting in a first floating-point number argument; applying a predetermined invertible function to the first floating-point number argument delivering a first output floating-point number, the output binary word being based on the first output floating-point number. The predetermined invertible function is configured to have the first output floating-point number having a non-null output rounding error resulting in having an application, to the first output floating-point number, of an invert function of the predetermined invertible function that delivers an input floating-point number corresponding to an input operand with a non-null input error.