CN116840759A - Rapid calibration system and method suitable for hybrid integrated circuit test system - Google Patents
Rapid calibration system and method suitable for hybrid integrated circuit test system Download PDFInfo
- Publication number
- CN116840759A CN116840759A CN202310811393.3A CN202310811393A CN116840759A CN 116840759 A CN116840759 A CN 116840759A CN 202310811393 A CN202310811393 A CN 202310811393A CN 116840759 A CN116840759 A CN 116840759A
- Authority
- CN
- China
- Prior art keywords
- module
- signal
- channel
- calibration
- output
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000012360 testing method Methods 0.000 title claims abstract description 38
- 238000000034 method Methods 0.000 title claims abstract description 23
- 238000013528 artificial neural network Methods 0.000 claims abstract description 30
- 238000012549 training Methods 0.000 claims description 26
- 238000004422 calculation algorithm Methods 0.000 claims description 24
- 230000006870 function Effects 0.000 claims description 21
- 230000004913 activation Effects 0.000 claims description 6
- 238000012886 linear function Methods 0.000 claims description 2
- 210000002569 neuron Anatomy 0.000 claims description 2
- 239000010410 layer Substances 0.000 description 65
- 230000008569 process Effects 0.000 description 10
- 238000005259 measurement Methods 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 230000005284 excitation Effects 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000010276 construction Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 230000001143 conditioned effect Effects 0.000 description 2
- 238000013499 data model Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 239000002356 single layer Substances 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R35/00—Testing or calibrating of apparatus covered by the other groups of this subclass
- G01R35/005—Calibrating; Standards or reference devices, e.g. voltage or resistance standards, "golden" references
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R35/00—Testing or calibrating of apparatus covered by the other groups of this subclass
- G01R35/005—Calibrating; Standards or reference devices, e.g. voltage or resistance standards, "golden" references
- G01R35/007—Standards or reference devices, e.g. voltage or resistance standards, "golden references"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Tests Of Electronic Circuits (AREA)
Abstract
The application discloses a rapid calibration system and a rapid calibration method suitable for a hybrid integrated circuit test system, wherein the system comprises a control module, a first signal source module, a second signal source module, a first acquisition module, a second acquisition module, a signal channel switching module, a change-over switch module, a first single-ended-to-differential module, a second single-ended-to-differential module and a channel to be calibrated of the hybrid integrated circuit test system; the channel to be calibrated of the hybrid integrated circuit test system comprises an input channel module and an output channel module, wherein the input channel module comprises a plurality of input channels, and the output channel module comprises a plurality of output channels. According to the application, the measured data is processed through neural network modeling, the data characteristics are searched, the prediction and automatic generation of the calibration data are realized, and the method has the characteristics of high speed, strong universality and high accuracy.
Description
Technical Field
The present application relates to a hybrid integrated circuit test system, and more particularly, to a system and method for fast calibration of a hybrid integrated circuit test system.
Background
Modules requiring calibration of hybrid integrated circuit test systems generally include: the device comprises a digital test module, an analog test module and a power supply test module. The calibration items include: PPMU output/acquisition unit calibration, digital output channel voltage amplitude calibration, digital input channel comparison threshold calibration, analog output channel calibration, and DPS voltage/current output calibration. Taking the digital output channel voltage amplitude calibration as an example, the single channel output voltage range is set to be-1.5V to +6.5V, if the output voltage is measured by taking 1mV as a step, at least 8000 data points are collected in a single channel. Meanwhile, considering the number of channels of other modules, the voltage needs to be collected again and calibration data is generated after the output voltage is set every time, so that a great deal of time is spent in calibrating all channels and projects by adopting a traditional calibration method.
Measurement of the traditional instrument is limited to single or a plurality of variables, the correlation among parameters is weak, and the measurement data can be processed by adopting a simple least square method or regression model. The mixed integrated circuit tester has the advantages that the measured parameter variable is increased, the data is influenced by the outside more and more, the functional relationship is complex, if the traditional regression model is used for modeling, the prediction accuracy is low, the data model and the actual situation can have non-negligible deviation, and the result of the data model is seriously influenced.
Disclosure of Invention
The application aims to overcome the defects of the prior art, provides a rapid calibration system and a rapid calibration method suitable for a hybrid integrated circuit test system, processes measurement data through neural network modeling, searches data characteristics, realizes the prediction and automatic generation of calibration data, and has the characteristics of high speed, strong universality and high accuracy.
The aim of the application is realized by the following technical scheme: a rapid calibration system suitable for a hybrid integrated circuit test system comprises a control module, a first signal source module, a second signal source module, a first acquisition module, a second acquisition module, a signal channel switching module, a change-over switch module, a first single-end-to-differential module, a second single-end-to-differential module and a channel to be calibrated of the hybrid integrated circuit test system;
the channel to be calibrated of the hybrid integrated circuit test system comprises an input channel module and an output channel module, wherein the input channel module comprises a plurality of input channels, and the output channel module comprises a plurality of output channels;
the control input end of the first signal source module is connected with the control module, and the output end of the first signal source module is connected with the signal channel switching module; the signal channel switching module is also connected with each input channel respectively, the output end of each input channel is connected with the first acquisition module, the output end of the first acquisition module is connected with the control module, the output end of the first acquisition module is also connected with the first single-ended differential module, and the output end of the first single-ended differential module is connected with the control module;
the control input end of the second signal source module is connected with the control module, the output end of the second signal source module is connected with the change-over switch module, the change-over switch module is respectively connected with each output channel, the output end of each output channel is connected with the signal channel change-over module, the signal channel change-over module is also connected with the second acquisition module, the second acquisition module is connected with the control module, the output end of the second acquisition module is also connected with the second single-ended to differential module, and the output end of the second single-ended to differential module is connected with the control module;
the control module is also communicated with an external computer through a LAN bus, and receives a calibration command sent by the computer to calibrate each input channel and each output channel:
when any one of the input channels is calibrated, the control module controls the first signal source module to generate a standard signal, meanwhile, the signal channel switching module is controlled to be communicated with the input channel, an output signal output by the input channel is acquired by the first acquisition module to obtain an acquisition signal, the acquisition signal is transmitted back to the control module, the acquisition signal is converted into a differential signal through the first single-ended differential module and transmitted back to the control module, the control module builds a model through a BP neural network algorithm, and model training is carried out through received data to obtain a calibration model of the input channel;
when any one of the output channels is calibrated, the control module controls the second signal source module to generate a standard signal, meanwhile, controls the change-over switch module and the signal channel change-over module to switch to the output channel, after the signal generated by the standard signal is transmitted to the output channel through the change-over switch module, the output signal of the output channel is transmitted to the second acquisition module through the signal channel change-over module, the acquired signal is transmitted to the control module by the second acquisition module, and meanwhile, the signal acquired by the second acquisition module is converted into a differential signal through the second single-ended to differential module and is transmitted back to the control module; the control module builds a model through a BP neural network algorithm, and performs model training through the received data to obtain a calibration model of the output channel.
A rapid calibration method suitable for a hybrid integrated circuit test system comprises an input channel calibration step and an output channel calibration step:
the input channel calibration step includes:
step A1: the computer inputs a calibration command to the control module through the LAN bus;
step A2: the control module selects an input channel as a calibration channel and switches to the calibration channel by sending a control signal to the signal channel switching module;
step A3: the control module controls the first signal source module to generate a reference signal through an internal bus;
step A4: the reference signal enters a calibration channel through a signal channel switching module, signals output by the calibration channel are acquired through a first acquisition module, and the acquired signals are transmitted to a control module;
meanwhile, the first acquisition module transmits the acquired acquisition signals to a first single-ended differential module and converts the acquired acquisition signals into differential signals;
the acquisition signal, the differential signal, the reference signal generated by the first reference source module and the increment of the reference signal to the acquisition signal are processed; wherein the differential signal comprises a positive side signal and a negative side signal;
step A5: the control module controls the first signal source to adjust the output reference signal in a set voltage step by step, and the steps A3 to A4 are repeatedly executed to obtain a plurality of signal samples;
step A6: constructing a calibration model of the calibration channel through a BP neural network algorithm, training the calibration model by using the obtained signal sample, and obtaining a model of the current calibration channel after training is finished;
step A7: when each input channel is sequentially selected as a calibration channel, repeatedly executing the steps A2 to A6 under each selected input channel to obtain a calibration model of each input channel;
the output channel calibration step includes:
step B1: the external computer outputs a calibration command to the control module through the LAN bus;
step B2: the control module selects one output channel as a calibration channel, and switches the calibration channel by sending a control signal to the signal channel switching module and the switcher module;
step B3: the control module controls the second signal source module to generate a reference signal;
step B4: the reference signal generated by the second signal source module enters the calibration channel through the change-over switch module, the signal output by the calibration channel is transmitted to the second acquisition module through the signal channel change-over module, the acquired signal is transmitted to the control module by the second acquisition module, and meanwhile, the signal acquired by the second acquisition module is converted into a differential signal through the second single-ended-to-differential module and is transmitted back to the control module;
the acquired signal, the differential signal, the reference signal generated by the second reference source module and the increment of the reference signal to the acquired signal are processed; wherein the differential signal comprises a positive side signal and a negative side signal;
step B5: the control module controls the second signal source module to repeatedly execute the steps B3-B4 according to the set voltage step-by-step adjustment output reference signal, and a plurality of signal samples are obtained;
step B6: constructing a calibration model of the calibration channel through a BP neural network algorithm, training the calibration model by utilizing each signal sample, and obtaining the calibration model of the current calibration channel after training is finished;
step B7: and when each output channel is sequentially selected as a calibration channel, repeatedly executing the steps B2 to B6 under each selected output channel to obtain a calibration model of each output channel.
The beneficial effects of the application are as follows: (1) The BP neural network with stronger fitting capability is adopted as a calibration algorithm, and the weight is initialized by adopting normal distribution, so that the accuracy and convergence speed of the BP neural network are improved;
(2) The neural network algorithm is realized on ARM and FPGA, and the hardware construction of the calibration system is completed;
(3) The application combines the calibration of the hybrid integrated circuit testing system and the neural network algorithm, is not limited to the traditional method for measuring and calibrating by collecting voltage/current values one by one, and has the advantages of high speed and high precision.
Drawings
FIG. 1 is a schematic diagram of the system principle of the present application;
FIG. 2 is a schematic diagram of a BP neural network;
FIG. 3 is a training set performance diagram for simulation in an embodiment;
fig. 4 is a graph of the relative error of the test set.
Detailed Description
The technical solution of the present application will be described in further detail with reference to the accompanying drawings, but the scope of the present application is not limited to the following description.
As shown in fig. 1, a fast calibration system suitable for a hybrid integrated circuit test system includes a control module, a first signal source module, a second signal source module, a first acquisition module, a second acquisition module, a signal channel switching module, a switch module, a first single-end-to-differential module, a second single-end-to-differential module, and a channel to be calibrated of the hybrid integrated circuit test system;
the channel to be calibrated of the hybrid integrated circuit test system comprises an input channel module and an output channel module, wherein the input channel module comprises a plurality of input channels, and the output channel module comprises a plurality of output channels;
the control input end of the first signal source module is connected with the control module, and the output end of the first signal source module is connected with the signal channel switching module; the signal channel switching module is also connected with each input channel respectively, the output end of each input channel is connected with the first acquisition module, the output end of the first acquisition module is connected with the control module, the output end of the first acquisition module is also connected with the first single-ended differential module, and the output end of the first single-ended differential module is connected with the control module;
the control input end of the second signal source module is connected with the control module, the output end of the second signal source module is connected with the change-over switch module, the change-over switch module is respectively connected with each output channel, the output end of each output channel is connected with the signal channel change-over module, the signal channel change-over module is also connected with the second acquisition module, the second acquisition module is connected with the control module, the output end of the second acquisition module is also connected with the second single-ended to differential module, and the output end of the second single-ended to differential module is connected with the control module;
the control module is also communicated with an external computer through a LAN bus, and receives a calibration command sent by the computer to calibrate each input channel and each output channel:
when any one of the input channels is calibrated, the control module controls the first signal source module to generate a standard signal, meanwhile, the signal channel switching module is controlled to be communicated with the input channel, an output signal output by the input channel is acquired by the first acquisition module to obtain an acquisition signal, the acquisition signal is transmitted back to the control module, the acquisition signal is converted into a differential signal through the first single-ended differential module and transmitted back to the control module, the control module builds a model through a BP neural network algorithm, and model training is carried out through received data to obtain a calibration model of the input channel;
when any one of the output channels is calibrated, the control module controls the second signal source module to generate a standard signal, meanwhile, controls the change-over switch module and the signal channel change-over module to switch to the output channel, after the signal generated by the standard signal is transmitted to the output channel through the change-over switch module, the output signal of the output channel is transmitted to the second acquisition module through the signal channel change-over module, the acquired signal is transmitted to the control module by the second acquisition module, and meanwhile, the signal acquired by the second acquisition module is converted into a differential signal through the second single-ended to differential module and is transmitted back to the control module; the control module builds a model through a BP neural network algorithm, and performs model training through the received data to obtain a calibration model of the output channel.
In an embodiment of the application, the rapid calibration system further comprises a power module for powering the entire rapid calibration system. The rapid calibration system further comprises a memory connected with the control module; the first acquisition module and the second acquisition module are ADC acquisition modules. The control module consists of ARM and FPGA, and the ARM and the FPGA are communicated through a local bus. The ARM receives a calibration command sent by a computer through a LAN bus, and then sends the calibration command to the FPGA through a local bus, the FPGA controls the on-off of a switch according to the command, switches the connection of a signal channel and a test channel with calibration, and simultaneously controls a reference source to generate a reference signal.
In the embodiment of the application, the control module comprises a calibration unit, wherein a BP neural network is adopted by a calibration algorithm of the calibration unit, the calibration algorithm is realized on ARM and FPGA, and the structure is a single hidden layer; the neuron numbers of the input layer, the hidden layer and the output layer are 2:3:1, wherein the input layer and the hidden layer use Sigmoid functions as activation functions, and the activation functions of the output layer adopt linear functions. The Sigmoid function is fitted by a polynomial piecewise fitting mode.
A rapid calibration method suitable for a hybrid integrated circuit test system comprises an input channel calibration step and an output channel calibration step:
the input channel calibration step includes:
step A1: the computer inputs a calibration command to the control module through the LAN bus;
step A2: the control module selects an input channel as a calibration channel and switches to the calibration channel by sending a control signal to the signal channel switching module;
step A3: the control module controls the first signal source module to generate a reference signal through an internal bus;
step A4: the reference signal enters a calibration channel through a signal channel switching module, signals output by the calibration channel are acquired through a first acquisition module, and the acquired signals are transmitted to a control module;
meanwhile, the first acquisition module transmits the acquired acquisition signals to a first single-ended differential module and converts the acquired acquisition signals into differential signals;
the acquisition signal, the differential signal, the reference signal generated by the first reference source module and the increment of the reference signal to the acquisition signal are processed; wherein the differential signal comprises a positive side signal and a negative side signal;
step A5: the control module controls the first signal source to adjust the output reference signal in a set voltage step by step, and the steps A3 to A4 are repeatedly executed to obtain a plurality of signal samples;
step A6: constructing a calibration model of the calibration channel through a BP neural network algorithm, training the calibration model by using the obtained signal sample, and obtaining a model of the current calibration channel after training is finished;
step A7: when each input channel is sequentially selected as a calibration channel, repeatedly executing the steps A2 to A6 under each selected input channel to obtain a calibration model of each input channel;
the output channel calibration step includes:
step B1: the external computer outputs a calibration command to the control module through the LAN bus;
step B2: the control module selects one output channel as a calibration channel, and switches the calibration channel by sending a control signal to the signal channel switching module and the switcher module;
step B3: the control module controls the second signal source module to generate a reference signal;
step B4: the reference signal generated by the second signal source module enters the calibration channel through the change-over switch module, the signal output by the calibration channel is transmitted to the second acquisition module through the signal channel change-over module, the acquired signal is transmitted to the control module by the second acquisition module, and meanwhile, the signal acquired by the second acquisition module is converted into a differential signal through the second single-ended-to-differential module and is transmitted back to the control module;
the acquired signal, the differential signal, the reference signal generated by the second reference source module and the increment of the reference signal to the acquired signal are processed; wherein the differential signal comprises a positive side signal and a negative side signal;
step B5: the control module controls the second signal source module to repeatedly execute the steps B3-B4 according to the set voltage step-by-step adjustment output reference signal, and a plurality of signal samples are obtained;
step B6: constructing a calibration model of the calibration channel through a BP neural network algorithm, training the calibration model by utilizing each signal sample, and obtaining the calibration model of the current calibration channel after training is finished;
step B7: and when each output channel is sequentially selected as a calibration channel, repeatedly executing the steps B2 to B6 under each selected output channel to obtain a calibration model of each output channel.
In an embodiment of the present application, a BP neural network model is shown in FIG. 3. The algorithm selects BP, the hidden layer number is determined to be 1, the activation function is selected to be a tan sig function, and the hidden layer node number is 10; setting training times to 2000 and target errors 1e-8, selecting 6000 groups of data as a training set, and using 2000 groups of data as a testing set to perform network training;
the BP neural network construction steps specifically include:
step 1, taking 1 channel as a measuring object and preparing data before constructing a network
(1) Setting input variables of the BP neural network: sampling voltage, feedback voltage at positive output end, feedback voltage at negative output end, actually measured voltage and voltage increment
(2) Setting a target error: 1e-8
(3) Setting the iterative training times of a network: 2000 times
(4) Setting a training set: 6000 groups
(5) Setting a test set: group 2000
(6) Selecting an activation function: sigmoid;
step 2, constructing BP neural network
(1) Input layer: five voltages generated in the 1 channel are used as input variables of the three-layer network architecture, and a vector matrix [ sampling voltage (x 1), positive output end feedback voltage (x 2), negative output end feedback voltage (x 3), actual measurement voltage (x 4) and voltage increment (x 5) ] is constructed.
(2) Hidden layer: the hidden layer in the network is of a single-layer structure, the number of the hidden layer nodes of the single-layer structure is defined as 10, each weight value from the input layer to the hidden layer is defined as a weight matrix weight, the calculation process of the final value of each hidden layer node is Sigmoid (weight. Transmit x1, x2, x3, x4, x 5), and the processing result is defined as weight_output.
(3) Output layer: the matrix formed by the nodes of the hidden layer is processed again by using an excitation function Sigmoid, and the processing process is Sigmoid (weight_output);
step 3, taking the steps (1) to (3) in the step 2 as one forward propagation of the BP neural network, wherein the process is defined as forward processing of voltage, the final output result of the processing is used as a calibrated voltage value, and the voltage compensation value is a parameter when gradient descent is carried out in the backward propagation of network error and can be defined as an error value;
step 4, error back propagation: obtaining a voltage error by calculating a difference value between the voltage value after the forward propagation and the acquired sample voltage data, reversely propagating the error as a parameter to an input layer in a gradient descending mode, and readjusting each weight value to serve as a basis for the second forward propagation;
step 5: and (3) combining error back propagation and the steps (1) to (3) in the step (2), repeating iteration for 2000 times based on 6000 groups of voltage data to perform network training, so as to obtain voltage compensation values of each iteration number, and obtaining more accurate voltage values when the final error is less than or equal to 1 e-8.
In the embodiment of the application, the calibration of the input signal of the hybrid integrated circuit takes the calibration of the PPMU acquisition unit as an example, firstly, the upper computer sends a calibration command to the control module through the LAN bus to transmit the calibration command to the first signal source module through the internal bus, the first signal source module outputs a signal (standard signal A1) after receiving the command, the acquired signal (acquired signal A1) is returned to the control module, the acquired signal A1 can realize the single-ended differential function through the ADC driver after the acquired signal A1 is conditioned, and a group of differential signals, namely a positive output end feedback signal P1 and a negative output end feedback signal P1, are obtained at the moment. The control module processes the received acquisition signal A1, the standard signal A1 output by the previous signal source, the differential signal and the voltage increment E through a three-layer neural network of a calibration algorithm (BP).
The output signal calibration of the hybrid integrated circuit takes the voltage amplitude calibration of a digital output channel as an example, an upper computer sends a calibration command to a control module through a LAN bus to enable the control module to control an output signal (B1) of the control module to a signal channel switching module through transmission to a second signal source module, finally, a signal (a collected signal B1) collected by a collection module is returned to the control module, the collected signal B1 can be conditioned to realize a single-ended differential transfer function through an ADC driver, and a group of differential signals, namely a positive output end feedback signal P '1 and a negative output end feedback signal P'1, are obtained at the moment. The control module processes the received output signal B1, the acquired signal B1 acquired by the acquisition module, the differential signal and the voltage increment E through a three-layer neural network of a calibration algorithm (BP).
In the algorithm of the present implementation, we describe the whole hybrid integrated circuit and neural network calibration algorithm in conjunction with PPMU output/acquisition unit calibration. Firstly, a standard signal A1, an acquisition signal A1, a positive output end feedback voltage P1, a negative output end feedback voltage P1 and a voltage increment E are all data and parameters which need to be calibrated, so that the five data are taken as node values in an input layer in the whole BP network, then each data of the input layer is processed by a three-layer neural network to obtain output data, the data are calibrated data, the current error is calculated according to the learning and the iteration of the network, the back propagation of the network is carried out, the weight of each layer of the network is adjusted according to the error value, and the forward propagation of the network is carried out again, so that the operation is repeated until the error is within a specified error range.
In the construction of the network model, w is defined as a weight value among all layers, all node values of a first hidden layer are obtained through node weight sums of the hidden layers, the next step is carried out, the output value of the first hidden layer needs to be subjected to nonlinear fitting by using an excitation function, a fitting curve based on a localized value is used as all the output weights of the next hidden layer, when the current hidden layer is transmitted, all node weights of the hidden layers are used again to calculate all the node numbers in the next hidden layer, then fitting of the excitation function is carried out again, and the like, when all node values of the first input layer are transmitted to the last hidden layer, and the output value is calculated, wherein the output value at the moment is used as the result of the first network iteration, and the above processes can be roughly summarized as follows: wherein, the variable Y is defined as the processed data, and x is the value of each node of the initial input layer:
input layer: x is x i (i＝1,2,3,…,n)
First layer hidden layer:
the first hidden layer outputs weights:
second layer hidden layer:
the second hidden layer outputs weights:
……
n-th layer hidden layer output:
by the above process, we can obtain that the fitted curve of each layer is the learning curve of the network passing through the next layer, so that the fitted curve of the network approximately passes through Y 1 、Y 2 、Y 3 、…、Y n Each fitted curve is closer to the division and trend of the actual voltage value than the previous fitted curve.
The above is the output result value of the first network iteration, and the result value is compared with the standard signal A1 collected by us to obtain an offset value (error), the offset value is voltage compensation, the voltage compensation is counter-propagated through the network to adjust each weight value in the network, the output of each layer of the network can be directly affected by the magnitude of the weight w known from the network structure, and the error is caused by the actually processed output voltage and theoretical voltage value due to the fact that w is not proper.
In the propagation of the reverse error, we use the bias of the error to w to construct the relation between the error and the weight, and take out a node of one hidden layer in the network.
The Error and the weight w have a close relation, and according to the chain rule, the deviation of w can obtain a new function relation according to the corresponding relation between the Error and the weight w, and the function represents the integral description of the gradient descent rate in the counter propagation in the network, and the excitation function selected before is usedDerivative Sigmoid of excitation function is obtained by conducting derivative calculation ‘ (x)＝1-tanh 2 (x) The gradient descent method in the network is represented by the function, when the error is reversely transferred, the output layer of the hidden layer in each layer uses the derivative to calculate the error value received by each node in each layer, the error of each node is calculated, the weight w of each layer is updated, the error is propagated through the hidden layer of each layer according to the forward propagation principle, the final processing result is obtained, the error is judged to be compared with the set error range, when the range is exceeded, the network iteration is continued, and when the range is exceeded, the network iteration is stopped.
The calculation of the reverse error follows the following calculation principle in each layer of the network:
n-th layer output layer error signal: er (Er) n_output ＝(x Standard value -x Actual measurement value )*Sigmoid‘(x n_active_output )
N-th layer hidden layer error signal: er (Er) n ＝(Er n_output ·x n_weight_output )*Sigmoid‘(x n_active_output )
The current ownership weight value is updated immediately each time more than one error propagation has been performed:
and (3) adjusting the weight value of the n-th hidden layer: w (W) output ＝W output (weight before update) +learning Rate (Er) n_output ·x n_active_output ) T +momentum factor output
And (3) adjusting the weight value of an n-th input layer:
W input ＝W input (weight before update) +learning Rate (Er) n ·x n_active_input ) T +momentum factor input
The input and output momentum factor calculation principle in the hidden layer is as follows:
momentum factor output =learning rate (Er n_output ·x n_active_output ) T
Momentum factor input =learning rate (Er n ·x n_active_input ) T
Repeatedly updating the w weight value in the process is the process of continuously re-fitting the actual measurement value by the network, and adjusting the actual measurement value to a proper weight value to obtain the final effective calibration voltage
In the embodiment of the application, matlab simulation is carried out by taking voltage amplitude calibration of a digital module output channel as an example, and a neural network model structure and some parameter settings are set:
the simulation results are shown in a training set performance diagram shown in fig. 3, wherein the X axis of the diagram represents the iteration times, the Y axis represents the mean square error, and the Y axis is used for measuring the performance of the whole neural network. It can be seen that the best performance error 1.7872e-08 is reached after 513 iterations.
FIG. 4 shows the relative error of the test set, the Y-axis shows the error value of the measured voltage and the output voltage in the 2000 data sets of the test set, and the X-axis shows the data points of the test set. It can be seen from the graph that the transformation of the relative error of the second half group of data is smoothed.
While the foregoing description illustrates and describes a preferred embodiment of the present application, it is to be understood that the application is not limited to the form disclosed herein, but is not to be construed as limited to other embodiments, but is capable of use in various other combinations, modifications and environments and is capable of changes or modifications within the spirit of the application described herein, either as a result of the foregoing teachings or as a result of the knowledge or skill of the relevant art. And that modifications and variations which do not depart from the spirit and scope of the application are intended to be within the scope of the appended claims.
Claims (7)
1. A rapid calibration system for a hybrid integrated circuit test system, comprising: the system comprises a control module, a first signal source module, a second signal source module, a first acquisition module, a second acquisition module, a signal channel switching module, a change-over switch module, a first single-ended-to-differential module, a second single-ended-to-differential module and a channel to be calibrated of a hybrid integrated circuit test system;
the channel to be calibrated of the hybrid integrated circuit test system comprises an input channel module and an output channel module, wherein the input channel module comprises a plurality of input channels, and the output channel module comprises a plurality of output channels;
the control input end of the first signal source module is connected with the control module, and the output end of the first signal source module is connected with the signal channel switching module; the signal channel switching module is also connected with each input channel respectively, the output end of each input channel is connected with the first acquisition module, the output end of the first acquisition module is connected with the control module, the output end of the first acquisition module is also connected with the first single-ended differential module, and the output end of the first single-ended differential module is connected with the control module;
the control input end of the second signal source module is connected with the control module, the output end of the second signal source module is connected with the change-over switch module, the change-over switch module is respectively connected with each output channel, the output end of each output channel is connected with the signal channel change-over module, the signal channel change-over module is also connected with the second acquisition module, the second acquisition module is connected with the control module, the output end of the second acquisition module is also connected with the second single-ended to differential module, and the output end of the second single-ended to differential module is connected with the control module;
the control module is also communicated with an external computer through a LAN bus, and receives a calibration command sent by the computer to calibrate each input channel and each output channel:
when any one of the input channels is calibrated, the control module controls the first signal source module to generate a standard signal, meanwhile, the signal channel switching module is controlled to be communicated with the input channel, an output signal output by the input channel is acquired by the first acquisition module to obtain an acquisition signal, the acquisition signal is transmitted back to the control module, the acquisition signal is converted into a differential signal through the first single-ended differential module and transmitted back to the control module, the control module builds a model through a BP neural network algorithm, and model training is carried out through received data to obtain a calibration model of the input channel;
when any one of the output channels is calibrated, the control module controls the second signal source module to generate a standard signal, meanwhile, controls the change-over switch module and the signal channel change-over module to switch to the output channel, after the signal generated by the standard signal is transmitted to the output channel through the change-over switch module, the output signal of the output channel is transmitted to the second acquisition module through the signal channel change-over module, the acquired signal is transmitted to the control module by the second acquisition module, and meanwhile, the signal acquired by the second acquisition module is converted into a differential signal through the second single-ended to differential module and is transmitted back to the control module; the control module builds a model through a BP neural network algorithm, and performs model training through the received data to obtain a calibration model of the output channel.
2. A rapid calibration system for a hybrid integrated circuit test system as recited in claim 1, wherein: the rapid calibration system also comprises a power module for supplying power to the whole rapid calibration system.
3. A rapid calibration system for a hybrid integrated circuit test system as recited in claim 1, wherein: the rapid calibration system further comprises a memory connected with the control module; the first acquisition module and the second acquisition module are ADC acquisition modules.
4. A rapid calibration system for a hybrid integrated circuit test system as recited in claim 1, wherein: the control module consists of ARM and FPGA, and the ARM and the FPGA are communicated through a local bus;
the ARM receives a calibration command sent by a computer through a LAN bus, and then sends the calibration command to the FPGA through a local bus, the FPGA controls the on-off of a switch according to the command, switches the connection of a signal channel and a test channel with calibration, and simultaneously controls a reference source to generate a reference signal.
5. A rapid calibration system for a hybrid integrated circuit test system as recited in claim 1, wherein: the control module comprises a calibration unit, wherein a BP neural network is adopted for a calibration algorithm of the calibration unit, the calibration algorithm is realized on ARM and FPGA, and the structure is a single hidden layer; the neuron numbers of the input layer, the hidden layer and the output layer are 2:3:1, wherein the input layer and the hidden layer use Sigmoid functions as activation functions, and the activation functions of the output layer adopt linear functions.
6. A rapid calibration system for a hybrid integrated circuit testing system according to claim 5, wherein: the Sigmoid function is fitted by a polynomial piecewise fitting mode.
7. A method of rapid calibration for a hybrid integrated circuit test system, based on the system of any one of claims 1-6, characterized by: the method comprises an input channel calibration step and an output channel calibration step:
the input channel calibration step includes:
step A1: the computer inputs a calibration command to the control module through the LAN bus;
step A2: the control module selects an input channel as a calibration channel and switches to the calibration channel by sending a control signal to the signal channel switching module;
step A3: the control module controls the first signal source module to generate a reference signal through an internal bus;
step A4: the reference signal enters a calibration channel through a signal channel switching module, signals output by the calibration channel are acquired through a first acquisition module, and the acquired signals are transmitted to a control module;
meanwhile, the first acquisition module transmits the acquired acquisition signals to a first single-ended differential module and converts the acquired acquisition signals into differential signals;
taking the acquired signal, the differential signal, the reference signal generated by the first reference source module and the increment from the reference signal to the acquired signal as signal samples; wherein the differential signal comprises a positive side signal and a negative side signal;
step A5: the control module controls the first signal source module to adjust the output reference signal in a set voltage step by step, and the steps A3-A4 are repeatedly executed to obtain a plurality of signal samples;
step A6: constructing a calibration model of the calibration channel through a BP neural network algorithm, training the calibration model by using the obtained signal sample, and obtaining a model of the current calibration channel after training is finished;
step A7: when each input channel is sequentially selected as a calibration channel, repeatedly executing the steps A2-A6 under each selected input channel to obtain a calibration model of each input channel;
the output channel calibration step includes:
step B1: the external computer outputs a calibration command to the control module through the LAN bus;
step B2: the control module selects one output channel as a calibration channel, and switches the calibration channel by sending a control signal to the signal channel switching module and the switcher module;
step B3: the control module controls the second signal source module to generate a reference signal;
step B4: the reference signal generated by the second signal source module enters the calibration channel through the change-over switch module, the signal output by the calibration channel is transmitted to the second acquisition module through the signal channel change-over module, the acquired signal is transmitted to the control module by the second acquisition module, and meanwhile, the signal acquired by the second acquisition module is converted into a differential signal through the second single-ended-to-differential module and is transmitted back to the control module;
taking the acquired signal, the differential signal, the reference signal generated by the second reference source module and the increment from the reference signal to the acquired signal as signal samples; wherein the differential signal comprises a positive side signal and a negative side signal;
step B5: the control module controls the second signal source module to repeatedly execute the steps B3-B4 according to the set voltage step-by-step adjustment output reference signal, and a plurality of signal samples are obtained;
step B6: constructing a calibration model of the calibration channel through a BP neural network algorithm, training the calibration model by utilizing each signal sample, and obtaining the calibration model of the current calibration channel after training is finished;
step B7: and when each output channel is sequentially selected as a calibration channel, repeatedly executing the steps B2-B6 under each selected output channel to obtain a calibration model of each output channel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310811393.3A CN116840759A (en) | 2023-07-04 | 2023-07-04 | Rapid calibration system and method suitable for hybrid integrated circuit test system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310811393.3A CN116840759A (en) | 2023-07-04 | 2023-07-04 | Rapid calibration system and method suitable for hybrid integrated circuit test system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116840759A true CN116840759A (en) | 2023-10-03 |
Family
ID=88172158
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310811393.3A Pending CN116840759A (en) | 2023-07-04 | 2023-07-04 | Rapid calibration system and method suitable for hybrid integrated circuit test system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116840759A (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103048550A (en) * | 2012-12-28 | 2013-04-17 | 成都泰格微电子研究所有限责任公司 | Test system and test data calibration method of S-parameter of surface mount microwave device |
| CN108333557A (en) * | 2018-01-31 | 2018-07-27 | 成都泰格微波技术股份有限公司 | A kind of phase alignment system and method for multichannel direction-finding receiver |
| US20190324434A1 (en) * | 2017-08-02 | 2019-10-24 | Strong Force Iot Portfolio 2016, Llc | Systems and methods for data collection and data sharing in an industrial environment |
| CN110726925A (en) * | 2019-09-29 | 2020-01-24 | 中国船舶重工集团公司第七0九研究所 | Multi-channel jitter injection calibration device and method |
| CN113726352A (en) * | 2021-09-23 | 2021-11-30 | 重庆两江卫星移动通信有限公司 | Method for correcting quadrature error of self-adaptive zero intermediate frequency receiver |
| CN114095022A (en) * | 2021-11-24 | 2022-02-25 | 复旦大学 | Method for calibrating successive approximation analog-to-digital converter of split assembly line based on machine learning |
| CN114330553A (en) * | 2021-12-29 | 2022-04-12 | 电子科技大学 | Digital acquisition system calibration method based on supervised learning |
| US20220302921A1 (en) * | 2021-03-19 | 2022-09-22 | Microsoft Technology Licensing, Llc | Cloud assisted calibration of analog-to-digital converters |
| CN115425976A (en) * | 2022-09-20 | 2022-12-02 | 重庆邮电大学 | ADC sampling data calibration method and system based on FPGA |
-
2023
- 2023-07-04 CN CN202310811393.3A patent/CN116840759A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103048550A (en) * | 2012-12-28 | 2013-04-17 | 成都泰格微电子研究所有限责任公司 | Test system and test data calibration method of S-parameter of surface mount microwave device |
| US20190324434A1 (en) * | 2017-08-02 | 2019-10-24 | Strong Force Iot Portfolio 2016, Llc | Systems and methods for data collection and data sharing in an industrial environment |
| CN108333557A (en) * | 2018-01-31 | 2018-07-27 | 成都泰格微波技术股份有限公司 | A kind of phase alignment system and method for multichannel direction-finding receiver |
| CN110726925A (en) * | 2019-09-29 | 2020-01-24 | 中国船舶重工集团公司第七0九研究所 | Multi-channel jitter injection calibration device and method |
| US20220302921A1 (en) * | 2021-03-19 | 2022-09-22 | Microsoft Technology Licensing, Llc | Cloud assisted calibration of analog-to-digital converters |
| CN113726352A (en) * | 2021-09-23 | 2021-11-30 | 重庆两江卫星移动通信有限公司 | Method for correcting quadrature error of self-adaptive zero intermediate frequency receiver |
| CN114095022A (en) * | 2021-11-24 | 2022-02-25 | 复旦大学 | Method for calibrating successive approximation analog-to-digital converter of split assembly line based on machine learning |
| CN114330553A (en) * | 2021-12-29 | 2022-04-12 | 电子科技大学 | Digital acquisition system calibration method based on supervised learning |
| CN115425976A (en) * | 2022-09-20 | 2022-12-02 | 重庆邮电大学 | ADC sampling data calibration method and system based on FPGA |
Non-Patent Citations (1)
| Title |
|---|
| 王娟 等: "《基于GA优化LS-SVM的霍尔位移传感器的温度补偿》", 《东华大学学报》, vol. 43, no. 5 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | A lithium-ion battery-in-the-loop approach to test and validate multiscale dual H infinity filters for state-of-charge and capacity estimation | |
| Xiong et al. | Co-estimation of state of charge and capacity for lithium-ion batteries with multi-stage model fusion method | |
| CN108120451A (en) | Based on silicon micro accerometer temperature-compensation method, the system for improving PSO optimization neural networks | |
| CN109541510B (en) | Channel amplitude and phase calibration method and device suitable for array radiometer | |
| CN109754013B (en) | Electric power system hybrid measurement fusion method based on unscented Kalman filtering | |
| CN107994903B (en) | Analog-to-digital conversion circuit and pipeline analog-to-digital converter | |
| CN110632528A (en) | Lithium battery SOH estimation method based on internal resistance detection | |
| CN111272334A (en) | Particle swarm optimization BP neural network-based multidimensional force sensor calibration decoupling method | |
| CN103078640B (en) | A kind of RLS adaptive-filtering calibration steps for ADC | |
| CN113459104B (en) | Robot Positioning Error Prediction Method and Prediction Device Based on Support Vector Regression | |
| CN112749784B (en) | Computing device and acceleration method of neural network | |
| CN111366221A (en) | Weighing method and system of strain type pressure sensor | |
| CN116840759A (en) | Rapid calibration system and method suitable for hybrid integrated circuit test system | |
| US20240120932A1 (en) | Circuits, chips, systems and methods for eliminating random perturbation | |
| CN113031434B (en) | Fractional order self-adaptive control method and device for time-lag multi-flexible swing arm system | |
| CN112731273B (en) | Low-complexity signal direction-of-arrival estimation method based on sparse Bayesian | |
| CN114095022B (en) | Split assembly line successive approximation analog-to-digital converter calibration method based on machine learning | |
| CN107587955B (en) | The scaling method of rocket engine thrust offset based on deep belief network | |
| CN115642913A (en) | Analog-to-digital converter ADC calibration method, device, equipment and storage medium | |
| WO2005041417A2 (en) | Digital input signals constructor providing analog representation thereof | |
| Li et al. | Data modeling of calibration parameter measurement based on MLP model | |
| JPH04113481A (en) | Manufacture of neural network type data processor | |
| CN118041363A (en) | TI ADC neural network calibration method based on signal derivative information | |
| Yurchenko et al. | Realization of Foreground Calibration and Correction of Pipelined ADC Based on 2.5-bit Stages | |
| CN114265049B (en) | UWB-based real-time ranging method, network structure and network node |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |