10 results about "Channel equalisation" patented technology

The application provides a channel equalization method based on a frequency sampling method, which comprises the following steps: obtaining a frequency domain quotient by taking a frequency response of a channel to be equalized and a reference channel; obtaining a frequency of an ideal equalization filter through a band-pass filter and a frequency domain entropy; performing an IFFT transformation on the ideal equalization filter; and performing several iterations according to channel performance to obtain an ideal equalization filter coefficient. Based on the ideal equalization filter coefficient, an actual equalization filter is constructed through a FIR filter model to complete equalization of the channel to be equalized. The method avoids inverse operation of a matrix of a conventional frequency domain least square method, reduces complexity of the algorithm, and can be realized in real time in engineering. Meanwhile, the method further improves equalization accuracy by using a band-pass filter and an iteration mode.

The invention particularly relates to a multi-frame medium signal-to-noise ratio channel equalization method based on measured data, and the method comprises the steps: carrying out the MTD of single-frame CPI data based on a multi-frame pulse coherent accumulation principle, achieving the clutter separation of a moving target and a ground object, extracting a target echo, carrying out the pulse compression processing, carrying out the envelope alignment and peak value normalization of multi-frame pulse pressure data, and carrying out the equalization of a multi-frame medium signal-to-noise ratio channel. Inter-frame complex number accumulation is carried out after normalization, on the premise that extra hardware cost input is not needed, the main lobe and side lobe ratio gain of the target pulse pressure result is remarkably improved, meanwhile, accurate channel equalization is achieved with the ideal pulse pressure result as the reference, dependence of a traditional channel equalization algorithm on the high signal-to-noise ratio is avoided, and MTD moving target detection of each frame of data can be taken into consideration. The actual measurement result shows that the algorithm can effectively improve the echo signal-to-noise ratio through coherent accumulation of multi-frame target pulse pressure results under the signal-to-noise ratio in a single frame, then carries out channel equalization, is convenient to implement, and can greatly improve the experiment efficiency of an external field.

This invention provides a multi-point network system comprising N network devices, capable of improving transmission speed through equalizer coefficient training. The N network devices include a master control device and multiple slave devices. The N network devices perform time zone synchronization in a synchronization phase, followed by equalizer coefficient training in a training phase, and then sequentially obtain transmission opportunities in a data transmission phase. Each of the N network devices includes a channel equalizer, which is used to be trained in the training phase and process data in the data transmission phase. In the training phase, the master control device issues a training notification to inform the multiple slave devices to enter the training phase; after issuing the training notification, the master control device performs equalizer coefficient training; after receiving the training notification, the multiple slave devices perform equalizer coefficient training. After the equalizer coefficient training is completed, the master control device issues a beacon to begin the data transmission phase.

The invention relates to the technical field of batteries, and provides a battery pack SOC cooperative control method and system based on multi-channel parallel equalization and rolling optimization, and the method comprises the steps: collecting the SOC of each single body in each control period, calculating an average value, screening out high and low SOC candidate single bodies, carrying out the discharge-charge pairing according to the number of channels, and carrying out the control of the SOC of each single body. Establishing a mapping relation between the monomer and the bidirectional DC-DC channel; an SOC state equation in a future prediction time domain is constructed on the basis, and an objective function is constructed by taking the equalization current of each channel as a decision variable: a deviation quadratic sum cumulant of each single SOC relative to the average SOC in the prediction time domain and a weighted sum of equalization energy consumption of each channel are minimized, and a current amplitude constraint is applied at the same time; and solving the optimization problem to obtain the optimal equalizing current, driving the corresponding monomer to execute charging and discharging operation, and then rolling to enter the next period. According to the method, overbalance and oscillation are effectively suppressed, high-precision, low-loss and self-adaptive global collaborative equalization is realized, the SOC dispersion is remarkably reduced, and the service life of the battery is prolonged.

The invention relates to a visible light and infrared video pedestrian re-identification method based on modal feature decomposition, which effectively reduces modal difference through channel equalization and enhancement operation, and directly decomposes modal sharing and unique features by adopting'dimension division + cross correlation constraint '. Time sequence feature extraction is carried out on video features by adopting an intra-modal time sequence consistency constraint method, so that the model robustness is improved; and through the constraint of a plurality of loss functions, the network can learn identity features with higher discrimination capability. Practical application shows that the method can effectively obtain sharing and unique features of the infrared and visible light models, can query video pairs, infrared or visible light independent videos and infrared or visible light images, and has a good application effect.

A Token-level multi-modal model quantification method based on learnable weights comprises the steps that calibration samples meeting the modal consistency requirement are input into a Token-level sensitivity model, and a Token-level sensitivity score used for representing Token relative importance is generated; generating a Token error vector between the non-quantized floating point output and the pseudo-quantized output according to the non-quantized floating point output and the pseudo-quantized output obtained by inputting the calibration sample into the multi-modal model; according to the Token-level sensitivity score and the Token error vector, defining a learnable Token weight used for dynamically allocating Token precision resources in the quantization process; defining a channel equalization factor used for performing scaling processing on each output channel of the multi-modal model; constructing a weighted reconstruction loss function according to the channel equalization factor and the learnable Token weight; minimizing a weighted reconstruction loss function by adjusting a channel equalization factor and a learnable Token weight; and performing deployment and reasoning execution of the multi-modal model by using the channel equalization factor with the minimized weighted reconstruction loss function and the learnable Token weight.

The invention discloses a digital pre-distortion processing method for a large-scale phased array, which comprises the following steps of: adding a coupling channel at the output end of a power amplifier in each transmitting channel in each sub-array of the phased array, equally dividing radio-frequency signals output by all the coupling channels into a plurality of groups, and selecting one group of radio-frequency signals to perform down-conversion and analog-to-digital conversion, entering a channel calibration mode and a digital pre-distortion resolving mode in a time-sharing manner, and performing digital pre-distortion processing, beam forming / calibration and channel equalization on the signal according to a preset condition in combination with the calibration parameter and the pre-distortion parameter; in order to solve the problems of high system complexity and high hardware cost caused by configuring an independent digital pre-distortion processing loop for each channel in the traditional scheme, a packet gating control mechanism is introduced, and feedback signals of a large number of channels are routed by groups, so that a set of shared rear-end processing circuit can serve all channels in a time-sharing manner, and the time-sharing performance of the system is improved. Therefore, the number of hardware is greatly reduced, the system structure is simplified, and the overall cost and power consumption are reduced.

The application discloses a kind of full-automatic planar waveguide coupling equipment based on nanoscale sensing contact, including chip placement assembly, input component, output component;Input component and output component are respectively transmission connection with stepper motor;Stepper motor is electrically connected with driver;Coupling software controls driver to control stepper motor to drive input component and output component to carry out displacement in different directions respectively, realize that FA fiber core and chip waveguide carry out mode field matching, so as to realize optical waveguide preliminary alignment, coupling software automatically switches displacement step, again through input component, output component realizes accurate alignment;Coupling software controls driver to realize multi-channel equalization by controlling angle stepper motor step and movement;The application adopts high-precision stepper motor with high-precision sensing contact sensor, completely realizes that planar optical waveguide is fully automatic high progress coupling to light, improves the precision of planar waveguide coupling to light, and completely realizes high-precision automatic angle adjustment function.

The application particularly relates to a multi-frame medium signal-to-noise ratio channel equalization method based on measured data, which is based on a multi-frame pulse coherent accumulation principle, first performs MTD on single-frame CPI data to realize moving target and ground clutter separation, then extracts target echo for pulse compression processing, then takes multi-frame pulse compression data for envelope alignment and peak value normalization, and after normalization, inter-frame complex accumulation is carried out, on the premise of not needing additional hardware cost investment, the mainlobe-to-sidelobe ratio gain of target pulse compression result is significantly improved, meanwhile, accurate channel equalization is realized based on ideal pulse compression result, the dependence of a traditional channel equalization algorithm on high signal-to-noise ratio is avoided, and MTD moving target detection of each frame of data is considered. The measured result shows that the algorithm can effectively improve the echo signal-to-noise ratio through multi-frame target pulse compression result coherent accumulation under single-frame medium signal-to-noise ratio, then channel equalization is carried out, the algorithm is convenient to realize, and the experimental efficiency in an outfield can be greatly improved.

A physical layer transceiver for connecting a host device to a wired channel medium divided into a total number of link segments includes a host interface for coupling to the host device, a line interface for coupling to the wired channel medium, and a feed-forward equalization circuit (FFE) operably coupled to the line interface to add back time-scattered components into a signal. Each individual filter segment is selectably configurable to correspond to each individual link segment by adjusting a respective delay line. The FFE circuit also includes control circuitry configured to detect a signal energy peak of at least one specific link segment of the total number of link segments, and, upon detecting the signal energy peak of the specific link segment, configure each one of the respective individual filter segments to correspond to each specific link segment by adjusting a respective delay line.