10 results about "Amplitude scaling" patented technology

The present disclosure discloses a seismic inversion method based on joint constraint of a physical model and priori information. The method includes: extracting seismic wavelets based on seismic data, and determining an amplitude scaling factor of the wavelets; counting priori information of impedance parameters; establishing an initial impedance parameter model by using seismic structural interpretation information and logging data; obtaining a simplified approximate equation based on an interface weak elasticity difference hypothesis, forward modeling a seismic gather by using the simplified equation, and calculating an inversion residual; rewriting an objective function into a function related to the impedance parameters by using a generalized linear inversion idea, solving the impedance parameters by using an iterative reweighted least squares algorithm, and updating the impedance parameters; and repeating the above steps until the inversion residual reaches the requirements or reaches the maximum number of iterations, and outputting a final processing result.

A method and corresponding system for generating harmonics based on an input signal are provided. The method includes (S1) obtaining one or more frequency band signals from the input signal; generating (S2) a first amplitude scaling signal comprising one or more harmonic components based on at least one or each of the frequency band signals, the amplitude of the first amplitude scaling signal being proportional to the amplitude of the input signal; generating (S3) a normalized signal based on at least one or each of the frequency band signals, the amplitude of the normalized signal being independent of the amplitude of the input signal; and for at least one or each of the frequency band signals, multiplying the normalized signal by the corresponding first amplitude scaling signal to generate a second amplitude scaling signal comprising one or more harmonic components; and for at least one or each of the frequency band signals, generating (S4) an output signal based on the first amplitude scaling signal comprising one or more harmonic components and the second amplitude scaling signal comprising one or more harmonic components.

The invention discloses a multi-antenna amplitude comparison and direction finding method and system based on beam pattern intelligent interpolation, and the method comprises the steps: carrying out the sparse sampling of a complete and continuous one-dimensional pattern response signal of an array antenna, and obtaining an analog data set for training; performing normalization processing on the analog data set, reconstructing an obtained one-dimensional normalized signal into a three-dimensional tensor form, and obtaining a normalized sparse signal tensor; taking the normalized sparse signal tensor as input, introducing a position coding module in a sine and cosine form to construct a neural network model, and outputting and obtaining a complete directional diagram signal under a recovered original amplitude scale; and applying the recovered complete directional diagram signal under the original amplitude scale to a target angle estimation task so as to carry out amplitude comparison and direction finding. According to the method, the deep learning network specially designed for the one-dimensional amplitude signal is constructed, and the high-fidelity reconstruction of the sparse under-sampled signal is realized by using the powerful global attention mechanism of the deep learning network.

The invention provides a simulation joint coding modulation method, device and equipment based on amplitude probability shaping, and the method comprises the steps: carrying out the deep coupling of S-K mapping coding and multi-ring constellation modulation based on amplitude probability shaping, and constructing a geometric shaping and probability shaping joint modulation architecture; wherein the radius of each ring of the constellation diagram is jointly determined by an amplitude scaling factor and a cumulative probability, uniform phase offset is introduced between adjacent rings to optimize geometric layout, and then the signal distortion ratio is maximized through joint iterative optimization of S-K mapping coding parameters, constellation amplitude and phase parameters, so that the system complexity is not increased, and the system performance is improved. Cooperative adaptive optimization of coding and modulation can be realized by a single modulation architecture, and the end-to-end transmission performance of the system is remarkably improved.

An apparatus (e.g. a UE 410) receives assistance information comprising phase shift and / or amplitude scaling parameters for subcarriers from a network node (e.g. gNb 420). The UE creates OFDM symbols

The invention provides an EEG self-supervised representation learning method based on a potential diffusion model, and the method comprises the steps: carrying out the preprocessing of a multi-channel EEG original signal, and segmenting the multi-channel EEG original signal into a plurality of EEG samples; each EEG sample and a channel enhancement signal thereof are input into an EEG encoder to generate an EEG representation, and the channel enhancement comprises zero mask and amplitude scaling; performing principal component analysis (PCA) processing on the EEG sample, and mapping the EEG sample to a potential space with a high signal-to-noise ratio to obtain a potential representation; the potential representation is input into a conditional diffusion model, the conditional diffusion model conducts conditional guidance based on the noise time step and the EEG representation, reconstructed potential representation is generated through the denoising process, the reconstructed potential representation is mapped back to an original space through inverse PCA, and a reconstructed EEG signal is obtained. In the tasks of anomaly detection, event type classification and the like, the EEG self-supervised representation learning model based on the potential diffusion model provided by the invention obtains the performance equivalent to that of the most advanced method with less pre-training data, and shows the advantages of the model in the aspect of EEG signal reconstruction.

The invention discloses a feature construction and parameter joint optimization method for time delay-Doppler domain blind source separation. The method comprises the following steps: performing orthogonal time frequency space demodulation on a time domain mixed signal to obtain a time delay-Doppler domain received signal matrix; screening salient points according to a salient point detection threshold value; calculating a complex gain ratio of each receiving node relative to the reference node, and constructing a stable feature vector finally used for clustering based on complex gains; clustering is carried out on the feature vectors; constructing a global performance evaluation function integrating clustering quality, noise suppression and clustering number matching, and jointly optimizing a salient point detection threshold, a clustering radius and a minimum sample number; and reconstructing the desired signal by using the optimal parameter. According to the method, the problems of phase boundary discontinuity and amplitude scale difference are solved through refined feature construction, the overall performance of the system is optimal through parameter joint optimization, and the clustering precision and robustness of blind source separation are remarkably improved.

This invention provides an analog joint coding modulation method, apparatus, and device based on amplitude probability shaping. By deeply coupling S-K mapping coding with multi-ring constellation modulation based on amplitude probability shaping, a modulation architecture combining geometric shaping and probability shaping is constructed. The radius of each ring in the constellation diagram is jointly determined by the amplitude scaling factor and the cumulative probability. A uniform phase offset is introduced between adjacent rings to optimize the geometric layout. Then, through joint iterative optimization of S-K mapping coding parameters and constellation amplitude and phase parameters, the signal distortion ratio is maximized. Thus, without increasing system complexity, a single modulation architecture can achieve collaborative adaptive optimization of coding and modulation, significantly improving the end-to-end transmission performance of the system.

The invention provides a frame-level alignment boundary confrontation attack method, system and device for a sequence recognition model, equipment and a medium, and belongs to the field of computer vision, voice processing and confrontation machine learning. The method comprises the following steps: S1, inputting a sample into a target sequence identification model in a test stage to obtain an initial reference alignment tag sequence; s2, based on the initial reference alignment tag sequence, constructing an alignment boundary between the reference alignment tag and the competition tag; s3, on the basis of the aligned boundary margins, dynamic continuous gating weights are generated through smooth mapping, a gating weighted marginal optimization target is constructed and iteratively updated, and candidate adversarial samples are obtained; and S4, performing total variation (TV) smoothing and amplitude scaling search on the candidate adversarial samples under successful retention constraints to generate high-fidelity adversarial samples. The method can improve the attack resisting efficiency and success rate, and can be used for scenes of robustness evaluation, privacy protection, copyright protection and the like of a sequence recognition model.

The present disclosure relates to the technical field of communications, and specifically relates to a transport layer determination method and apparatus, a terminal, a network device, and a storage medium. The transport layer determination method comprises: on the basis of an amplitude scaling factor of a spatial domain basis vector, determining a transmit power of each transport layer corresponding to the spatial domain basis vector; determining receiving information on the basis of spatial domain basis vector and the amplitude scaling factor; on the basis of the receiving information, determining a strongest transport layer from among the transport layers corresponding to the spatial domain basis vector; and sending first indication information to a network device, wherein the first indication information is used for indicating the strongest transport layer. In the present disclosure, the problem that a strongest transport layer determined by a terminal is inconsistent with an actual strongest transmission layer caused by determining the strongest transport layer still according to the condition that transmit powers of transport layers are equal when the transmit powers need to be scaled on the basis of an amplitude scaling factor can be avoided.