WO2020164435A1 - Magnetic resonance radio frequency power amplifier device and magnetic resonance system - Google Patents

Abstract

A magnetic resonance radio frequency power amplifier device and a magnetic resonance system, wherein the magnetic resonance radio frequency power amplifier device comprises: a first balun, configured to receive a feedforward signal and transmit same to a processor by means of a first analog-to-digital converter; a second balun, configured to receive a feedback signal and transmit same to the processor by means of a second analog-to-digital converter; and the processor, configured to perform digital frequency conversion and signal synchronization on both the received feedforward signal and feedback signal, train a radio frequency power amplifier inverse function estimator by means of a neural network indirect learning structure, and copy the weight coefficient of the radio frequency power amplifier inverse function estimator to a predistorter. The output signal of the predistorter enters a radio frequency power amplifier module through a third balun, a digital-to-analog converter and an analog up-conversion module in sequence. The radio frequency power amplifier module amplifies the received signal and divides same into two paths.

Description

Magnetic resonance radio frequency power amplifier device and magnetic resonance system Technical field

The present disclosure belongs to the field of magnetic resonance, and particularly relates to a magnetic resonance radio frequency power amplifier device and a magnetic resonance system.

Background technique

The statements in this section merely provide background information related to the present disclosure, and do not necessarily constitute prior art.

In a magnetic resonance system, a radio frequency power amplifier is required to amplify the radio frequency pulses emitted by the magnetic resonance spectrometer to tens to hundreds of watts and output them to the transmitting coil to excite the experimental samples. Therefore, the RF power amplifier is an indispensable part of the magnetic resonance system. However, like general power amplifiers, RF power amplifiers have non-linear characteristics. When the RF power amplifiers work in the saturation or cut-off region, they will cause non-linear distortion of the pulse signal. The greater the power of the input signal, the more serious the distortion of the RF power amplifier. This kind of distortion includes amplitude and phase distortion within the frequency band and spectrum expansion outside the frequency band, which interferes with the signal transmission of adjacent channels, and finally makes the MRI system imaging artifacts, distortions, picture quality degradation, and interference with disease diagnosis and scientific research. Therefore, it is necessary to realize the linearization of the power amplifier while ensuring the working efficiency of the radio frequency power amplifier.

Methods to solve the nonlinearity of RF power amplifier include the following methods:

(1) The most direct method is to manufacture high-precision and high-linearity RF power amplifiers. However, the inventor found that its manufacturing process is complicated and expensive, and it is generally not suitable for high-power amplifiers.

(2) Power back-off is the easiest way to improve the linearity of the power amplifier. By reducing the operating point of the power amplifier, the power amplifier can work at a place where it is back 10 to 15 dB from the saturation point. The power back-off method is simple to implement, but the inventor finds that its disadvantage is that it will greatly reduce the working efficiency of the power amplifier, thereby increasing the maintenance cost of the system.

(3) The direct feedback method uses the output signal to directly suppress the input signal. The inventor found that its disadvantage is that it is difficult to estimate the delay of the output signal to the input signal, which makes the system stability poor. Therefore, the indirect negative feedback method is more often used in engineering. The indirect negative feedback method means that the output and input signals are compared through an indirect connection. Therefore, the advantages of the indirect negative feedback method are high precision, mature technology and low price. The inventor found that its shortcomings are that the feedback loop delay is difficult to control, the system is not stable enough, and it is not suitable for applications with a wide frequency band.

(4) The basic principle of the feedforward method is as follows: the interference signal is separated by the cancellation loop to superimpose and cancel the delayed power amplifier output signal, and then realize the linearization of the power amplifier. This method is faster, has better linearization, and uses a wider bandwidth, but the inventor finds that its structure is complex, cost is high, efficiency is low, and adaptability is poor.

(5) The basic idea of predistortion technology is to insert a predistorter with reciprocal curve characteristics and power amplifier curve characteristics in front of the power amplifier, and realize the linearization of the power amplifier through the combined use of the predistorter and the power amplifier. The predistorter of predistortion technology is mostly realized by analog circuit. Its advantages are low cost, simple circuit structure, and wide adaptable bandwidth. However, the inventor found that its disadvantage is that the predistortion effect is very limited, especially for high-order component distortion. More difficult.

In summary, the inventor found that the current radio frequency power amplifiers have low working efficiency and poor predistortion effects during nonlinear processing.

Summary of the invention

In order to solve the above-mentioned problems, the first aspect of the present disclosure provides a magnetic resonance radio frequency power amplifier device, which has the advantages of high working efficiency and good predistortion effect during nonlinear processing.

The technical solution of a magnetic resonance radio frequency power amplifier device in the first aspect of the present disclosure is:

A magnetic resonance radio frequency power amplifier device of the present disclosure includes:

The first balun is configured to receive the feedforward signal and transmit it to the processor via the first analog-to-digital converter;

The second balun is configured to receive the feedback signal and transmit it to the processor via the second analog-to-digital converter;

The processor is configured to perform digital frequency conversion and signal synchronization processing on the received feedforward signal and feedback signal, and then train the RF power amplifier inverse function estimator through the neural network indirect learning structure, and the RF power amplifier inverse function estimator The weight coefficient of is copied to the predistorter; the structure of the RF power amplifier inverse function estimator and the predistorter are the same;

The output signal of the predistorter enters the RF power amplifier module through the third balun, the digital-to-analog converter and the analog up-conversion module in sequence;

The radio frequency power amplifier module is configured to amplify the received signal and divide it into two channels, one of which is transmitted, and the other is processed by the coupling module and the analog down-conversion module to obtain the feedback signal.

Further, the process in which the processor performs digital frequency conversion on both the received feedforward signal and the feedback signal is:

Digitally down-convert the feed-forward signal into a baseband signal, and then digitally up-convert the feed-forward signal to a signal whose center frequency matches the center frequency of the processor;

The feedback signal is digitally down-converted into a baseband signal, and then the feedback signal is digitally up-converted into a signal whose center frequency matches the center frequency of the processor.

The advantage of the above scheme is that by digitally converting the received feedforward signal and feedback signal, the center frequencies of the received feedforward signal and feedback signal are in line with the center frequency signal of the processor, which can improve the working efficiency of the processor , Thereby improving the working stability of the entire magnetic resonance radio frequency power amplifier device.

Further, the predistorter is composed of a linear part and a non-linear part.

Further, the linear part is a FIR filter structure; the non-linear part is a neural network structure.

Further, the neural network structure is a shallow learning neural network or a neural network based on deep learning.

Among them, the predistorter not only needs to fit the non-linear distortion of the RF power amplifier, but also needs to fit the memory effect of the RF power amplifier. At this time, the neural network predistorter must also have a memory effect.

It is worth noting that the neural network here can be a traditional shallow learning neural network, such as back propagation (BP) neural network, multilayer perceptual neural network, etc., or it can be a neural network based on deep learning, such as deep neural Network, recurrent neural network, convolutional neural network.

The training algorithm used by the neural network can be gradient descent method, additional momentum method, conjugate gradient method, Newton algorithm, Levenberg-Marquardt algorithm, etc.

Further, the process for the processor to train the RF power amplifier inverse function estimator through the neural network indirect learning structure is:

According to the output difference between the predistorter and the RF power amplifier inverse function estimator, adaptively calculate and update the weight coefficient of the RF power amplifier inverse function estimator;

When the output difference between the predistorter and the RF power amplifier inverse function estimator is less than a certain value, the weight coefficient of the RF power amplifier inverse function estimator is copied to the predistorter.

The advantage of the above solution is that the predistorter copies the weight coefficients of the radio frequency power amplifier inverse function estimator to the predistorter when converging, which can improve the stability of the entire magnetic resonance radio frequency power amplifier device.

Further, the processor is also configured to:

According to the changes of the RF power amplifier module, the weight coefficient of the predistorter is periodically updated, and the performance changes of the RF power amplifier module are monitored and tracked in real time to realize the adaptive predistortion of the RF power amplifier module.

Further, the analog up-conversion module includes: a first digitally controlled oscillator, a band-pass filter, and a first mixer; the first digitally controlled oscillator is used to generate two quadrature signals of sine and cosine, and pass the first The mixer completes the frequency spectrum relocation together; the band-pass filter is used to filter out the unnecessary frequency spectrum generated during up-conversion mixing.

Further, the analog down-conversion module includes: a second digitally controlled oscillator, a digital finite filter, and a second mixer; the second digitally controlled oscillator is used to generate two quadrature signals of sine and cosine, and pass the second The mixers jointly complete the relocation of the frequency spectrum; the digital finite filter is used to filter out the redundant frequency spectrum generated during down-conversion mixing.

Further, the coupling module includes: a power divider and a signal attenuator, the power divider is used to extract a part of the radio frequency signal from the radio frequency power amplifier, and this part of the radio frequency signal is processed by the signal attenuator and the analog down conversion module in turn , Get the feedback signal.

In order to solve the above-mentioned problem, the second aspect of the present disclosure provides a magnetic resonance system, which has the advantages of high working efficiency and good predistortion effect during nonlinear processing.

A technical solution of a magnetic resonance system in the second aspect of the present disclosure is:

A magnetic resonance system of the present disclosure includes:

A magnetic resonance signal generating device, which is configured to generate a magnetic resonance pulse signal;

The above-mentioned magnetic resonance radio frequency power amplifier device is configured to perform predistortion amplification of the magnetic resonance pulse signal and output it to the magnetic resonance signal transmitting device;

The magnetic resonance signal transmitting device is configured to transmit a predistorted and amplified magnetic resonance signal.

The beneficial effects of the present disclosure are:

(1) The present disclosure uses a neural network to fit the nonlinearity of the radio frequency power amplifier of the magnetic resonance system, and predistort the radio frequency power amplifier, thereby improving the working efficiency of the radio frequency power amplifier device of the magnetic resonance system and the predistortion effect in the nonlinear processing process .

(2) The present disclosure uses the filter part to fit the memory effect of the radio frequency power amplifier, and uses the neural network part to fit the nonlinear distortion of the radio frequency power amplifier, thereby achieving a separate fitting of the nonlinearity and memory effect of the radio frequency power amplifier. Compared with the traditional memory neural network predistorter model, it not only greatly reduces the original network parameters and network scale, but also greatly reduces the amount of calculation in the iterative process of updating the weight coefficients.

Description of the drawings

The accompanying drawings of the specification constituting a part of the present disclosure are used to provide a further understanding of the present disclosure, and the exemplary embodiments and descriptions thereof are used to explain the present disclosure, and do not constitute an improper limitation of the present disclosure.

FIG. 1 is a schematic structural diagram of a magnetic resonance radio frequency power amplifier device provided by an embodiment of the present disclosure.

2 is a schematic diagram of the processor structure of a magnetic resonance radio frequency power amplifier device provided by an embodiment of the present disclosure.

Fig. 3 is a schematic diagram of a magnetic resonance radio frequency power amplifier device provided by an embodiment of the present disclosure.

Fig. 4 is a schematic diagram of an indirect learning structure based on a multilayer perceptual neural network provided by an embodiment of the present disclosure.

Fig. 5 is a schematic diagram of a multilayer perceptual neural network provided by an embodiment of the present disclosure.

Fig. 6 is an AM/AM characteristic curve before predistortion provided by an embodiment of the present disclosure.

FIG. 7 is an AM/AM characteristic curve after predistortion provided by an embodiment of the present disclosure.

Fig. 8 is an AM/PM characteristic curve before predistortion provided by an embodiment of the present disclosure.

Fig. 9 is an AM/PM characteristic curve after predistortion provided by an embodiment of the present disclosure.

FIG. 10 is the power spectrum of the original signal and the signal before and after predistortion provided by an embodiment of the present disclosure.

FIG. 11 is a schematic structural diagram of a magnetic resonance system provided by an embodiment of the present disclosure.

detailed description

It should be noted that the following detailed descriptions are all illustrative, and are intended to provide further description of the present disclosure. Unless otherwise indicated, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this disclosure belongs.

It should be noted that the terms used here are only for describing specific embodiments, and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprising" and/or "including" are used in this specification, they indicate There are features, steps, operations, devices, components, and/or combinations thereof.

As shown in FIG. 1, a magnetic resonance radio frequency power amplifier device of this embodiment includes:

The first balun is configured to receive the feedforward signal and transmit it to the processor via the first analog-to-digital converter;

The second balun is configured to receive the feedback signal and transmit it to the processor via the second analog-to-digital converter;

The processor is configured to perform digital frequency conversion and signal synchronization processing on the received feedforward signal and feedback signal, and then train the RF power amplifier inverse function estimator through the neural network indirect learning structure, and the RF power amplifier inverse function estimator The weight coefficient of is copied to the predistorter; the structure of the RF power amplifier inverse function estimator and the predistorter are the same;

The output signal of the predistorter enters the RF power amplifier module through the third balun, the digital-to-analog converter and the analog up-conversion module in sequence;

The radio frequency power amplifier module is configured to amplify the received signal and divide it into two channels, one of which is transmitted, and the other is processed by the coupling module and the analog down-conversion module to obtain the feedback signal.

Among them, balun: balun, responsible for mutual conversion between high-frequency single-ended signals and differential signals.

Analog-to-digital converter: There is a differential input port to convert analog signals into digital signals.

Processor: It can be realized by DSP or FPGA chip.

The processor mainly realizes the three functions of digital up-down conversion pre-processing, signal synchronization and digital pre-distortion of magnetic resonance radio frequency power amplifier, as shown in Figure 2.

Digital-to-analog converter: converts digital signals into analog signals and outputs analog signals.

Analog up-conversion module: Up-regulate the intermediate frequency signal output by the digital-to-analog converter to the radio frequency band.

RF power amplifier module: The up-converted signal is amplified by the RF power amplifier and then transmitted through the RF coil.

Coupling module: equivalent to a signal attenuator.

Analog down-conversion module: down-convert the RF signal.

Specifically, the analog up-conversion module includes: a first digitally controlled oscillator, a bandpass filter, and a first mixer; the first digitally controlled oscillator is used to generate two quadrature signals of sine and cosine and pass the first The mixer completes the frequency spectrum relocation together; the band-pass filter is used to filter out the unnecessary frequency spectrum generated during up-conversion mixing.

The analog down-conversion module includes: a second digitally controlled oscillator, a digital finite filter, and a second mixer; the second digitally controlled oscillator is used to generate two quadrature signals of sine and cosine, and pass the second mixer Jointly complete the relocation of the frequency spectrum; the digital finite filter is used to filter out the unnecessary frequency spectrum generated during down-conversion and mixing.

The coupling module includes: a power divider and a signal attenuator, the power divider is used to extract a part of the radio frequency signal from the radio frequency power amplifier, and this part of the radio frequency signal is processed by the signal attenuator and the analog down conversion module in turn to obtain feedback signal.

In a specific implementation, the process in which the processor performs digital frequency conversion on both the received feedforward signal and the feedback signal is as follows:

Digitally down-convert the feed-forward signal into a baseband signal, and then digitally up-convert the feed-forward signal into a signal whose center frequency matches the center frequency of the processor;

The feedback signal is digitally down-converted into a baseband signal, and then the feedback signal is digitally up-converted into a signal whose center frequency matches the center frequency of the processor.

The advantage of the above scheme is that by digitally converting the received feedforward signal and feedback signal, the center frequencies of the received feedforward signal and feedback signal are in line with the center frequency signal of the processor, which can improve the working efficiency of the processor , Thereby improving the working stability of the entire magnetic resonance radio frequency power amplifier device.

Among them, the neural network indirect learning structure is also implemented on a DSP or FPGA chip, as shown in Figure 3. In this structure, the structure of the predistorter and the RF power amplifier inverse function estimator are exactly the same. The predistorter is composed of a linear part and a nonlinear part.

The linear part is a FIR filter structure; the nonlinear part is a neural network structure.

The neural network structure is a shallow learning neural network or a neural network based on deep learning.

Among them, the predistorter not only needs to fit the non-linear distortion of the RF power amplifier, but also needs to fit the memory effect of the RF power amplifier. At this time, the neural network predistorter must also have a memory effect.

It is worth noting that the neural network here can be a traditional shallow learning neural network, such as a back propagation (BP) neural network, a multilayer perceptual neural network, etc., or it can be a neural network based on deep learning, such as a deep neural network. Network, recurrent neural network, convolutional neural network.

The training algorithm used by the neural network can be gradient descent method, additional momentum method, conjugate gradient method, Newton algorithm, Levenberg-Marquardt algorithm, etc.

Further, the process for the processor to train the RF power amplifier inverse function estimator through the neural network indirect learning structure is:

According to the output difference between the predistorter and the RF power amplifier inverse function estimator, adaptively calculate and update the weight coefficient of the RF power amplifier inverse function estimator;

When the output difference between the predistorter and the RF power amplifier inverse function estimator is less than a certain value, the weight coefficient of the RF power amplifier inverse function estimator is copied to the predistorter.

The advantage of the above solution is that the predistorter copies the weight coefficients of the radio frequency power amplifier inverse function estimator to the predistorter when converging, which can improve the stability of the entire magnetic resonance radio frequency power amplifier device.

In a specific implementation, the processor is further configured to:

According to the changes of the RF power amplifier module, periodically update the weight coefficient of the predistorter, monitor and track changes in the modular performance of the RF power amplifier in real time, and realize the adaptive predistortion of the RF power amplifier module.

The following takes an indirect learning structure based on a multilayer perceptual neural network, as shown in Figure 4 as an example to illustrate in detail:

The predistorter is a predistortion system composed of linear and nonlinear components. The linear part is an FIR filter, and the non-linear part is a three-layer ordinary multi-layer perceptual neural network with double input and double output. The filter part is used to fit the memory effect of the radio frequency power amplifier, and the neural network part is used to fit the nonlinear distortion of the radio frequency power amplifier, and then the nonlinearity and memory effect of the radio frequency power amplifier are fitted separately.

Compared with the traditional memory neural network predistorter model, the indirect learning structure based on the multilayer perceptual neural network not only greatly reduces the original network parameters and network scale, but also greatly reduces the amount of calculation in the iterative process of updating the weight coefficients.

For the multilayer perceptual neural network in the embodiment of Fig. 5, the feedforward calculation is as follows:

In formula (1),

Is the output of the jth neuron in the first layer of the multilayer perceptual neural network, M is the memory depth,

Is the connection weight coefficient between the jth neuron of the first layer and the i input of the input layer, and x ^-i is the i input of the neural network;

In formula (2),

Is the output of the jth neuron in the second layer, the hidden layer activation function

Is the tanh function,

Is the connection weight coefficient of the j-th neuron in the second layer and the i-th neuron in the first layer,

Is the output of the jth neuron in layer 1,

Is the bias coefficient of the jth neuron in the second layer;

In formula (3),

Is the output of the kth neuron in the third layer, l ₁ is the number of hidden layer nodes,

Is the connection weight coefficient between the kth neuron in the third layer and the jth neuron in the second layer,

Is the bias coefficient of the kth neuron in the third layer. Choose f(x)=x as the output layer neuron activation function.

The Bayesian-Levenberg-Marquardt optimization algorithm is used as the neural network training algorithm, and the objective function expression is:

F(X)=αE _W +βE _D (4)

Where

Where N is the number of rounds of the neural network weight update iteration; s2 is the number of nodes in the output layer of the neural network; q is the number of training samples per round,

When the number of training samples is q, the square of the error value of the neural network predictor and the inverse function estimator; v _j (x) is the square of the error value of the predictor and the inverse function estimator during the jth iteration; w _j is Network weight coefficient; α, β are coefficients; m is the total number of weight coefficients in the network, which is a positive integer greater than or equal to 1.

It can be seen that through the new objective function, the network can ensure that the network output error is as small as possible during the training process, while ensuring that the network has a small network weight coefficient. The iterative formula for Bayesian-Levenberg-Marquardt to update the neural network coefficients is:

X _k+1 ＝X _k -[αJ ^T J-(μ+β)I] ^-1 J ^T e (5)

Among them, μ and e are constant coefficients.

In the above formula, J is the Jacobi matrix, and its expression is

Among them, X is the weight coefficient vector of the network. The optimal values of the coefficients α and β are α ^MP and β ^{MP respectively} , and are given by the following formula:

Where γ=m-2α ^MP · tr(H ^MP ) ^-1 represents the number of effective weight coefficients, ranging from 0 to m. m is the total number of weight coefficients in the network, and H ^MP is the Hessian matrix of the objective function F(X) at its minimum point X _MP . In the calculation process, the Hessian matrix needs to be calculated. Using Gauss-Newton approximation to simplify the Hessian matrix, then:

Where J is the Jacobi matrix of E _D at point X _MP .

The operation steps of the Bayesian-LM algorithm are as follows:

⑴ Initialize the network parameters, and initialize the coefficient α=0, β=1.

⑵Using the Levenberg-Marquardt algorithm to minimize the network performance objective function F(X)=αE _W +βE _D.

(3) Use Gauss-Newton approximation method to solve H≈βJ ^T J+αI _m , and solve the number of effective parameters γ.

⑷Calculate the new estimated value of the coefficient

⑸ Repeat steps ⑵ to ⑷ until the algorithm converges.

The AM/AM characteristic diagram of the RF power amplifier before predistortion is shown in Figure 6. The AM/AM characteristic diagram after the pre-test is true is shown in Figure 7. It can be seen from Figure 6 and Figure 7 that the AM/AM characteristic of the RF power amplifier before predistortion is a nonlinear curve with hysteresis. The AM/AM characteristic of the RF power amplifier with predistorter is almost a straight line with a slope of 1 and no hysteresis, indicating that the amplitude amplification is basically linear.

The AM/PM characteristic diagram of the RF power amplifier before predistortion is shown in Figure 8. The AM/PM characteristic diagram after the pre-test is true is shown in Figure 9. It can be seen from Figure 8 and Figure 9 that before predistortion, when the input changes from 0 to 1, the phase of the AM/PM characteristic diagram of the RF power amplifier has different degrees of offset. The smaller the input, the greater the phase offset. Big. After predistortion, when the input amplitude changes in the AM/PM characteristic, the output phase offset is basically 0, indicating that the purpose of phase predistortion is basically achieved.

Figure 10 shows the power spectrum of the RF amplifier before and after predistortion. After predistortion, the in-band signal becomes flat. The predistorter structure can effectively reduce the adjacent channel power ratio by about 30dB.

The magnetic resonance radio frequency power amplifier device of this embodiment adopts a neural network to fit the nonlinearity of the radio frequency power amplifier of the magnetic resonance system, and predistorts the radio frequency power amplifier, thereby improving the working efficiency and the nonlinear processing process of the radio frequency power amplifier device of the magnetic resonance system. The predistortion effect in.

The magnetic resonance radio frequency power amplifier device of this embodiment uses the filter part to fit the memory effect of the radio frequency power amplifier, and uses the neural network part to fit the nonlinear distortion of the radio frequency power amplifier, so as to achieve a separate fitting of the nonlinearity and memory of the radio frequency power amplifier. effect. Compared with the traditional memory neural network predistorter model, it not only greatly reduces the original network parameters and network scale, but also greatly reduces the amount of calculation in the iterative process of updating the weight coefficients.

FIG. 11 is a schematic structural diagram of a magnetic resonance system provided by an embodiment of the present disclosure.

As shown in FIG. 11, a magnetic resonance system of this embodiment includes:

A magnetic resonance signal generating device, which is configured to generate a magnetic resonance pulse signal;

The magnetic resonance radio frequency power amplifier device shown in FIG. 1 is configured to perform predistortion amplifying and output the magnetic resonance pulse signal to the magnetic resonance signal transmitting device;

The magnetic resonance signal transmitting device is configured to transmit a predistorted and amplified magnetic resonance signal.

Specifically, the magnetic resonance signal generating device can be realized by a magnetic resonance spectrometer.

The magnetic resonance signal transmitting device can be realized by the radio frequency coil of the magnetic resonance system.

The magnetic resonance system of this embodiment adopts a neural network to fit the nonlinearity of the radio frequency power amplifier of the magnetic resonance system, and predistorts the radio frequency power amplifier, which improves the working efficiency of the radio frequency power amplifier device of the magnetic resonance system and the pre-processing in the nonlinear processing process. Distortion effect.

The magnetic resonance system of this embodiment uses the filter part to fit the memory effect of the radio frequency power amplifier, and uses the neural network part to fit the nonlinear distortion of the radio frequency power amplifier, so as to achieve a separate fitting of the nonlinearity and memory effect of the radio frequency power amplifier. Compared with the traditional memory neural network predistorter model, it not only greatly reduces the original network parameters and network scale, but also greatly reduces the amount of calculation in the iterative process of updating the weight coefficients.

Although the specific embodiments of the present disclosure are described above in conjunction with the accompanying drawings, they do not limit the scope of protection of the present disclosure. Those skilled in the art should understand that on the basis of the technical solutions of the present disclosure, those skilled in the art do not need to make creative efforts. Various modifications or variations that can be made are still within the protection scope of the present disclosure.

Claims (10)

1. A magnetic resonance radio frequency power amplifier device is characterized by comprising:

   The first balun is configured to receive the feedforward signal and transmit it to the processor via the first analog-to-digital converter;

   The second balun is configured to receive the feedback signal and transmit it to the processor via the second analog-to-digital converter;

   The processor is configured to perform digital frequency conversion and signal synchronization processing on the received feedforward signal and feedback signal, and then train the RF power amplifier inverse function estimator through the neural network indirect learning structure, and the RF power amplifier inverse function estimator The weight coefficient of is copied to the predistorter; the structure of the RF power amplifier inverse function estimator and the predistorter are the same;

   The output signal of the predistorter enters the RF power amplifier module through the third balun, the digital-to-analog converter and the analog up-conversion module in sequence;

   The radio frequency power amplifier module is configured to amplify the received signal and divide it into two channels, one of which is transmitted, and the other is processed by the coupling module and the analog down-conversion module to obtain the feedback signal.
2. The magnetic resonance radio frequency power amplifier device according to claim 1, wherein the process of the processor performing digital frequency conversion on both the received feedforward signal and the feedback signal is:

   Digitally down-convert the feed-forward signal into a baseband signal, and then digitally up-convert the feed-forward signal into a signal whose center frequency matches the center frequency of the processor;

   The feedback signal is digitally down-converted into a baseband signal, and then the feedback signal is digitally up-converted into a signal whose center frequency matches the center frequency of the processor.
3. The magnetic resonance radio frequency power amplifier device according to claim 1, wherein the predistorter is composed of a linear part and a non-linear part.
4. 5. The magnetic resonance radio frequency power amplifier device according to claim 3, wherein the linear part is a FIR filter structure; the non-linear part is a neural network structure.
5. The magnetic resonance radio frequency power amplifier device according to claim 4, wherein the neural network structure is a shallow learning neural network or a neural network based on deep learning.
6. The magnetic resonance radio frequency power amplifier device according to claim 1, wherein the process of the processor training the inverse function estimator of the radio frequency power amplifier through the neural network indirect learning structure is:

   According to the output difference between the predistorter and the RF power amplifier inverse function estimator, adaptively calculate and update the weight coefficient of the RF power amplifier inverse function estimator;

   When the output difference between the predistorter and the RF power amplifier inverse function estimator is less than a certain value, the weight coefficient of the RF power amplifier inverse function estimator is copied to the predistorter.
7. The magnetic resonance radio frequency power amplifier device according to claim 1, wherein the processor is further configured to:

   According to the changes of the RF power amplifier module, the weight coefficient of the predistorter is periodically updated, and the performance changes of the RF power amplifier module are monitored and tracked in real time to realize the adaptive predistortion of the RF power amplifier module.
8. The magnetic resonance radio frequency power amplifier device according to claim 1, wherein the analog up-conversion module comprises: a first digital control oscillator, a band pass filter, and a first mixer; and a first digital control The oscillator is used to generate two quadrature signals of sine and cosine, and the frequency spectrum is moved together through the first mixer; the band-pass filter is used to filter out the unwanted spectrum generated during up-conversion mixing;

   Or the analog down-conversion module includes: a second digitally controlled oscillator, a digital finite filter, and a second mixer; the second digitally controlled oscillator is used to generate two quadrature signals of sine and cosine, and pass the second mixer The frequency converters complete the relocation of the frequency spectrum; the digital finite filter is used to filter out the unnecessary frequency spectrum generated during down-conversion mixing.
9. The magnetic resonance radio frequency power amplifier device according to claim 1, wherein the coupling module comprises: a power divider and a signal attenuator, and the power divider is used to extract a part of the radio frequency signal from the radio frequency power amplifier , This part of the RF signal is processed by the signal attenuator and analog down-conversion module in turn to obtain the feedback signal.
10. A magnetic resonance system is characterized in that it comprises:

    A magnetic resonance signal generating device, which is configured to generate a magnetic resonance pulse signal;

    9. The magnetic resonance radio frequency power amplifier device according to any one of claims 1-9, which is configured to perform predistortion amplification of the magnetic resonance pulse signal and output it to the magnetic resonance signal transmitting device;

    The magnetic resonance signal transmitting device is configured to transmit a predistorted and amplified magnetic resonance signal.

Images