A Design Method of RF Low-Noise Amplifier Circuit Based on Hybrid Neural Network

Abstract

The present invention provides a design method for a radio frequency low-noise amplifier circuit based on a hybrid neural network. The method includes: S1 Selecting appropriate devices according to performance index requirements, and determining corresponding operating states and bias conditions; S2 Using the constructed hybrid neural network to obtain the scattering parameters and noise parameters of the transistor, and obtaining the S2P file of the device under specified conditions; S3 Performing stability analysis based on the obtained S2P file, and further completing the design of the low-noise amplifier. The present invention helps to improve the efficiency and applicability of analyzing and designing low-noise amplifiers.

Description

Technical Field

[0001] The present invention relates to the field of 5G technology, and particularly to a method for designing a radio frequency low-noise amplifier circuit based on a hybrid neural network. Background Art

[0002] With the advent of the 5G era, obtaining a high-quality and high-efficiency wireless communication system has been increasingly emphasized. As the front end of a wireless receiver, the performance analysis and design of a low-noise amplifier are particularly important. A low-noise amplifier (LNA) is generally used as a high-frequency or intermediate-frequency preamplifier for various radio receivers, as well as an amplifier circuit for high-sensitivity electronic detection devices. It is located at the front end of the receiving system, linearly amplifies the weak radio frequency signals received by the antenna, suppresses various noise interferences, and improves the system sensitivity. Due to the special position and role of the LNA in the receiving system, the design of this component plays a key role in the performance indicators of the entire receiving system.

[0003] In the prior art, the design method of a low-noise amplifier can only be studied based on a clear device model and equivalent circuit to obtain its scattering parameters (S-parameters) and noise parameters. In the case of strong device performance nonlinearity, unclear internal structure, limited range and accuracy of test instruments and equipment, the measurement of these parameters becomes a complex and time-consuming process, making the working process cumbersome and inefficient, and also very unfavorable for the research of low-noise amplifiers. The main problems are as follows:

[0004] (1) The traditional design of a low-noise amplifier is based on a clear device model and equivalent circuit. When a new semiconductor is developed, there is no clear equivalent circuit model for its internal structure, and it is also difficult to obtain an accurate device model, which makes the design of a low-noise amplifier very difficult.

[0005] (2) The parameters of the physical model of a low-noise amplifier for microwave / millimeter-wave band devices are strong functions of frequency, and the extraction is often very cumbersome and complex. Only a small amount of data can be obtained through limited performance tests. When the frequency and bias conditions change, it is impossible to quickly and accurately obtain the parameter data of the transistor. Summary of the Invention

[0006] In view of the above problems, the present invention aims to combine the ANN model with the GaN HEMT noise physical model, use the hybrid neural network model to obtain the scattering and noise parameters of the transistor under different frequencies and bias conditions, and fully demonstrate the model using the prepared actual devices and actual measurement data. Through these data, a complete S2P file of the transistor can be established, and then the S2P file can be used to design a low-noise amplifier. And a method for designing a radio frequency low-noise amplifier circuit based on a hybrid neural network is proposed.

[0007] The object of the present invention is achieved by the following technical solutions:

[0008] In a first aspect, the present invention discloses a design method for a radio frequency low-noise amplifier circuit based on a hybrid neural network, including:

[0009] S1 Select appropriate devices according to performance index requirements, and determine corresponding operating states and bias conditions;

[0010] S2 Use the constructed hybrid neural network to obtain the scattering parameters and noise parameters of the transistor, and obtain the S2P file of the device under specified conditions;

[0011] S3 Perform stability analysis based on the obtained S2P file, and further complete the design of the low-noise amplifier.

[0012] In one embodiment, the noise amplifier is a two-port network structure.

[0013] In one embodiment, in step S1, the determined operating state includes temperature;

[0014] The bias conditions include frequency, drain-source voltage, and drain-source current of the device.

[0015] In one embodiment, in step S2, the constructed hybrid neural network includes 3 neural network structures, and its inputs are drain-source current (Ids), drain-source voltage (Vds), and frequency (f). Among them, the first neural network structure ANN1 is used to obtain and output scattering parameters according to the input drain-source current (Ids), drain-source voltage (Vds), and frequency (f). The scattering parameters include S 11 , S 12 , S 21 , S 22 The amplitude and angle of, where S 11 Represents the input reflection coefficient when the output is terminated in a match, S 21 Represents the forward transmission gain when the output is terminated in a match, S 22 Represents the output reflection coefficient when the input is terminated in a match, S 12 Represents the reverse transmission gain when the input is terminated in a match; the second neural network structure ANN2 is used to obtain and output the intrinsic parameters and equivalent temperature of the EPC equivalent circuit model according to the input frequency (f); the third neural network structure ANN3 is used to obtain and output noise parameters according to the input drain-source current (Ids), drain-source voltage (Vds), and frequency (f). The noise parameters include the minimum noise figure (NF min ), the optimum reflection source coefficient (Mag(Γ opt ), Ang(Γ opt )) and the thermal resistance R n .

[0016] In one embodiment, the first neural network structure ANN1, the second neural network structure ANN2, and the third neural network structure ANN3 each include an input layer, a hidden layer, and an output layer. The hidden layer structure of the first neural network structure ANN1 is 9-8; the hidden layer structure of the third neural network structure ANN3 is 9-7.

[0017] In one embodiment, before step S2, the method further includes:

[0018] Sb1 training the hybrid neural network, including:

[0019] Selecting the data directly measured by the AlGaN HEMT device on the Si substrate as samples for neural network training. The samples include training samples and test samples, and the samples contain the scattering parameters measured under different biases of the device and the noise parameters measured at different frequencies;

[0020] Among them, the values of the scattering parameters in the samples are obtained by rapid measurement using an instrument;

[0021] The acquisition of the noise parameters in the samples includes:

[0022] Analyzing the device to be measured using the noise temperature, and the noise temperature is represented in the equivalent small-signal model. The equivalent small-signal EPC noise model contains the device intrinsic parameters C gs , r gs , g m , g ds , C gd , and the equivalent temperature parameters include τ, T g , T d , T c , where T g represents the gate noise voltage source introduced at the gate Gate, T d represents the drain noise current source, T c represents the correlation temperature between the two, and τ represents the correlation coefficient;

[0023] Calculating the noise parameters according to the device intrinsic parameters and the equivalent temperature parameters.

[0024] In one embodiment, step S3 specifically includes:

[0025] S31 Performing stability analysis according to the obtained S2P file, and designing a negative feedback circuit according to the stability analysis result;

[0026] S32 Designing a matching circuit, where the matching circuit includes input matching and output matching;

[0027] S33 Overall optimizing the circuit to complete the design of the low-noise amplifier.

[0028] In one embodiment, step S31 includes:

[0029] Calculating the stability factor K of the radio frequency noise amplifier circuit, where where S 11 represents the input reflection coefficient under the condition of output terminal termination matching, S 21 represents the forward transmission gain under the condition of output terminal termination matching, S 22 represents the output reflection coefficient under the condition of input terminal termination matching, S 12 represents the reverse transmission gain under the condition of input terminal termination matching;

[0030] When K < 1, introduce negative feedback to make the circuit reach a stable state, where the negative feedback is realized by connecting inductors in series at two source electrodes; by continuously adjusting the size of the inductor, so that the circuit satisfies K > 1 to meet the stability requirements.

[0031] In one embodiment, step S32 includes:

[0032] Perform matching for the minimum noise figure at the input end, select the center point of the minimum noise figure circle as the optimal reflection source coefficient and match it to 50Ω; perform conjugate matching at the output end to 50Ω for maximum gain matching.

[0033] In a second aspect, the present invention discloses a radio frequency low noise amplifier circuit design device based on a hybrid neural network, which is used to implement the radio frequency low noise amplifier circuit design method based on a hybrid neural network shown in any one of the above first aspects.

[0034] The beneficial effects of the present invention are as follows: The present invention proposes a construction method for a low noise amplifier and a neural network model, and combines the artificial neural network model with the low noise amplifier model. It can accurately and quickly obtain the scattering and noise parameters of the low noise amplifier under different frequencies, temperatures, and bias conditions through direct and limited test data, and can also be used to expand the device data sheet. These advantages greatly improve the efficiency of analyzing and designing low noise amplifiers, and are beneficial to establishing an efficient communication system. Description of the Drawings

[0035] The present invention is further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to the following drawings without creative efforts.

[0036] Figure 1 It is a schematic flowchart of a radio frequency low noise amplifier circuit design method based on a hybrid neural network shown in an embodiment of the present invention;

[0037] Figure 2 Schematic diagram of a two-port microwave network and S-parameters for an exemplary embodiment of the present invention;

[0038] Figure 3 I-V curve diagram of a device for an exemplary embodiment of the present invention;

[0039] Figure 4 Schematic diagram of a noise parameter test scheme for an exemplary embodiment of the present invention;

[0040] Figure 5 Schematic diagram of a hybrid neural network model for an exemplary embodiment of the present invention;

[0041] Figure 6 Schematic diagram of the construction process of a hybrid neural network for an exemplary embodiment of the present invention;

[0042] Figure 7 For an exemplary embodiment of the present invention, the bias condition is V ds = 3V, I ds = 70 mA, S-parameter fitting schematic diagram;

[0043] Figure 8 For an exemplary embodiment of the present invention, the bias condition is V ds = 3V, I ds = 90 mA, S-parameter fitting schematic diagram;

[0044] Figure 9 For an exemplary embodiment of the present invention, the bias condition is V ds = 4V, I ds = 70 mA, S-parameter fitting schematic diagram;

[0045] Figure 10 Schematic diagram of the design steps of a low-noise amplifier for an exemplary embodiment of the present invention;

[0046] Figure 11 Schematic diagram of the structure of a low-noise amplifier designed with stable matching for an exemplary embodiment of the present invention;

[0047] Figure 12 Schematic diagram of the circuit of a low-noise amplifier designed by ADS for an exemplary embodiment of the present invention;

[0048] Figure 13 Schematic diagram of the stability result for an exemplary embodiment of the present invention;

[0049] Figure 14 Schematic diagram of the input matching for an exemplary embodiment of the present invention;

[0050] Figure 15 Schematic diagram of the output matching for an exemplary embodiment of the present invention;

[0051] Figure 16 This is a schematic diagram of the S-parameters of an exemplary embodiment of the present invention when the operating frequency f of the radio frequency low-noise amplifier circuit is 3.5 GHz. Detailed implementation manners

[0052] The present invention will be further described in conjunction with the following application scenarios.

[0053] Refer to Figure 1 a circuit design method for a radio frequency low-noise amplifier based on a hybrid neural network shown in the embodiment, wherein

[0054] The noise amplifier targeted by the present invention has a two-port network structure. A two-port network is a circuit or device with 2 ports. The ports are connected to the internal network of the circuit, and it can represent the whole or a part of the circuit with their corresponding external characteristic parameters.

[0055] The method includes:

[0056] S1 Select appropriate devices according to the performance index requirements, and determine the corresponding operating states and bias conditions.

[0057] Among them, the selected appropriate devices include selecting devices that meet the requirements according to the noise figure and gain requirements of the low-noise amplifier to be designed, including the physical chips. Determining the corresponding operating states includes the device operating temperature, operating frequency, and the device DC bias conditions, mainly including the drain-source voltage Vds, drain-source current Ids, etc.

[0058] In one implementation manner, in step S1, the determined operating state includes temperature;

[0059] The bias conditions include frequency, the drain-source voltage and drain-source current of the device.

[0060] In one implementation, the present invention designs a radio frequency low noise amplifier based on a GaN-based high electron mobility transistor (GaN HEMT). A high electron mobility transistor (HEMT), also known as a modulation-doped field effect transistor (MODFET), is a type of field effect transistor that uses two materials with different energy bandgaps to form a heterojunction to provide a channel for carriers, unlike a metal oxide semiconductor field effect transistor that directly uses a doped semiconductor to form a conductive channel. In recent years, the development of high electron mobility transistors (HEMTs) has provided low noise and low power consumption for monolithic microwave integrated circuits (MMICs). Its operating frequency can range from 1 to 100 GHz, and this wide range of frequencies has been widely studied and applied in fields such as radar, communication satellites, wireless local area networks (WLANs), and cellular phones. The third-generation semiconductor materials are represented by materials such as gallium nitride (GaN) and silicon carbide (SiC). Among them, GaN is a new material with a relatively large bandgap. Compared with the first two generations of semiconductor materials, GaN materials have better linear characteristics (higher dynamic range at the same noise figure); greater broadband characteristics, suitable for making broadband devices; and can withstand higher input power, increasing the anti-interference ability of the whole machine.

[0061] S2 uses the constructed hybrid neural network to obtain the scattering parameters and noise parameters of the transistor, and obtains the S2P file of the device under specified conditions.

[0062] Among them, the S2P file specifically records the S and noise parameters of the HEMT device under fixed bias and temperature conditions. The S parameters mainly include S 11 , S 12 , S 21 , S 22 The amplitude values and phase values of four parameters, and the noise parameters mainly include the minimum noise figure NFmin at the operating frequency, the thermal noise resistance value Rn, the optimal matching amplitude value, the optimal matching phase, etc.

[0063] Specifically, the scattering parameters (S parameters): are an important set of parameters in microwave transmission. Obtaining accurate S parameters of the channel is a very important step. Through the S parameters, the present invention can see almost all the characteristics of the transmission channel. Most of the problems related to signal integrity, such as signal reflection, crosstalk, and loss, can find useful information from the S parameters.

[0064] A microwave network is an equivalent circuit of non-uniformity in a microwave transmission system. Microwave networks can be classified into single-port, two-port, and multi-port networks according to the number of channels of the transmission system they are connected to. Among them, the two-port microwave network is the most commonly used, and the present invention will construct a low-noise amplifier based on this network structure. In addition, starting from actual measurements, open and short circuits at the ports of the RF circuit will cause total signal reflection, and open and short circuit conditions with a wide frequency band are also not easy to achieve in the RF circuit. There is a need for a parameter to describe the network characteristics under the condition of network port matching. Each element of the scattering parameter matrix is obtained on the premise that a "matched load" is connected to a certain port and can be directly measured. Therefore, in order to accurately describe the operating characteristics of a transistor, the present invention uses scattering parameters to analyze the microwave two-port network, so as to design a compliant LNA.

[0065] Figure 2 The block diagram showing the two-port microwave network and its S parameters is presented. a1 = incident wave at the input port, b1 = reflected wave at the input port, a2 = incident wave at the output port, b2 = reflected wave at the output port.

[0066] S 11 = b1 / a1, S 11 represents the input reflection coefficient under the condition that the output end is terminated with a match, and is usually referred to as the return loss.

[0067] S 21 = b2 / a1, S 21 represents the forward transmission gain (coefficient) under the condition that the output end is terminated with a match, and is usually referred to as the insertion loss.

[0068] S 22 = b2 / a2, S 22 represents the output reflection coefficient under the condition that the input end is terminated with a match.

[0069] S 12 = b1 / a2, S 12 represents the reverse transmission gain (coefficient) under the condition that the input end is terminated with a match.

[0070] In addition to describing the operating characteristics of a transistor, scattering parameters can also reflect the reflection, transmission, isolation, etc. characteristics of a low-noise amplifier. When the internal structure of a device or model is unknown, its characteristics can be directly understood through scattering parameters. These advantages are all beneficial for the present invention to design the LNA and analyze its performance, and quickly judge whether it meets the application requirements.

[0071] Specifically, noise parameters: The performance of a low-noise amplifier mainly includes noise figure, reasonable gain, stability, etc., which are collectively referred to as noise parameters. Among them, the noise parameters of the present invention refer to the minimum noise figure (NFmin), the optimum reflection source coefficient (Ang(Γopt ) and Mag(Γ opt ), and a thermal resistor (Rn).

[0072] Noise is defined as an unwanted and irregular perturbation that interferes with a useful signal. The magnitude of noise lies in its ability to interfere with the useful signal. Therefore, the concept of signal-to-noise ratio is usually used to measure the degree of influence of noise on the useful signal, and its definition is the power of the useful signal divided by the power of the noise. For transistors and amplifiers, it is always desirable to increase the power of the useful signal as much as possible when the signal is amplified by the amplifier, while reducing or even suppressing the useless interference signal, that is, to increase the signal-to-noise ratio at the output as much as possible. Therefore, the noise figure is commonly used in the art to measure the noise characteristics of a transistor, and its definition formula is:

[0073]

[0074] In the above formula, P Si is the input signal power; P Ni is the input noise power; P So is the output signal power; P No is the output noise power.

[0075] In a wireless communication system, sensitivity is an important factor to be considered. To increase the communication distance of the receiver, the sensitivity of the receiver needs to be improved to enhance the overall performance index of the receiver. The sensitivity calculation formula of the receiver is:

[0076] S = -174 + NF + 10log(BW) + S / N(2)

[0077] In the above formula, NF is the noise figure, BW is the bandwidth of the signal, and S / N is the signal-to-noise ratio of the input signal. Only the noise figure in the formula can be optimized. Therefore, optimizing the noise figure is the most direct means to improve the sensitivity of the receiver. For RF engineers, improving the measurement efficiency and accuracy of the noise figure can reduce the design margin and relieve the design pressure. For component manufacturers, the flexibility and applicability of noise measurement can expand the application range of products and enhance the competitiveness of products. The improvement of the performance of wireless communication, satellite communication, and radar systems requires devices, subsystems, and systems to have a low noise figure, which requires more rapid and accurate measurement and judgment of the noise figure. The existing instrument measurement methods are cumbersome and time-consuming, and need to be re-measured when changing the bias or frequency. The present invention proposes a method that can quickly and accurately obtain scattering and noise parameters under different conditions, which is more conducive to the design and application of LNA.

[0078] In one implementation, before step S2, the method further includes:

[0079] Sb1 training the hybrid neural network, including:

[0080] The directly measured data of AlGaN HEMT devices on Si substrates are selected as samples for neural network training. The samples include training samples and test samples, and the samples contain the scattering parameters measured under different biases of the devices and the noise parameters measured at different frequencies.

[0081] Among them, the values of the scattering parameters in the samples are obtained by rapid measurement with an instrument.

[0082] The acquisition of the noise parameters in the samples includes:

[0083] The device under test is analyzed using the noise temperature, and the noise temperature is represented in an equivalent small-signal model. The equivalent small-signal EPC noise model contains the device intrinsic parameters C gs , r gs , g m , g ds , C gd , where C gs represents the capacitance between the gate and the source, r gs represents the gate-source resistance, g m is the transconductance, g ds represents the conductance between the drain and the source, C gd represents the capacitance between the gate and the drain, that is, Figure 4 the small-signal equivalent model marked on the left side in g , T d , T c , where T g represents the gate noise voltage source introduced at the gate Gate, T d represents the drain noise current source, T c represents the correlation temperature between the two, and τ represents the correlation coefficient;

[0084] The noise parameters are calculated based on the device intrinsic parameters and the equivalent temperature parameters.

[0085] In one scenario, when selecting sample data, the device I-V curve shown in the appendix Figure 3 is used to select the training and test samples for the neural network. The selection of the bias conditions for the training samples is shown in Table 1, and the frequencies are selected as 0.5 GHz, 1 GHz, 1.5 GHz, 2 GHz, 2.5 GHz, 3 GHz, 3.5 GHz, 4 GHz, 5 GHz, 6 GHz, 7 GHz, 8 GHz, 9 GHz, 10 GHz, a total of 169 groups of data. The test samples are respectively selected with the bias conditions (① V ds = 3 V, I ds = 70 mA, ② V ds = 3 V, I ds = 90 mA, ③ V ds = 4 V, I dsThree sets of data (i.e., = 70 mA).

[0086] Table 1: Selection of Bias Points for Training Samples

[0087]

[0088] The values of the scattering parameters can be quickly measured using an instrument, while the measurement of the noise parameters is relatively complex. The solution adopted in the present invention is as shown in the appendix Figure 4 shown;

[0089] When calculating the noise parameters, the noise temperature can be used to analyze the device under test. The noise temperature can be expressed in the equivalent small-signal model. Because there is a specific model, the noise temperature is widely used in the calculation and analysis of the noise figure.

[0090] The equivalent small-signal EPC noise model includes the intrinsic device parameters C gs , r gs , g m , g ds , C gd , and the noise source characteristic temperature parameters include τ, T g , T d , T c , whose meanings are respectively the gate noise voltage source introduced at the gate (characterized by T g ), the drain noise current source (characterized by T d ), and the correlation temperature T c between the two and the correlation coefficient τ. The noise parameters can be calculated based on the device intrinsic parameters and the noise temperature.

[0091] After obtaining the training and test samples of the scattering and noise parameters, the neural network can be constructed.

[0092] In one implementation, in step S2, the constructed hybrid neural network includes 3 neural network structures, whose inputs are the drain-source current (Ids), drain-source voltage (Vds), and frequency (f). Among them, the first neural network structure ANN1 is used to obtain and output the scattering parameters according to the input drain-source current (Ids), drain-source voltage (Vds), and frequency (f). The scattering parameters include the amplitude and angle of S 11 , S 12 , S 21 , S 22 . Among them, S 11 represents the input reflection coefficient under the condition of a matched termination at the output end, S 21 represents the forward transmission gain under the condition of a matched termination at the output end, S 22 represents the output reflection coefficient under the condition of a matched termination at the input end, S 12represents the reverse transmission gain under the condition of matching termination at the input end; the second neural network structure ANN2 is used to obtain and output the intrinsic parameters and equivalent temperature of the EPC equivalent circuit model according to the input frequency (f); the third neural network structure ANN3 is used to obtain and output the noise parameters according to the input drain-source current (Ids), drain-source voltage (Vds) and frequency (f), where the noise parameters include the minimum noise figure (NF min ), the optimal reflection source coefficient (Mag(Γ opt ), Ang(Γ opt )) and the thermal resistance R n .

[0093] Artificial Neural Networks (ANN) rely on the complexity of the system and adjust the relationships between a large number of internally connected nodes to achieve the purpose of processing information. It has the ability of self-learning and self-adaptation. It can analyze and master the potential laws between a batch of corresponding input-output data provided in advance. Finally, according to these laws, new input data is used to calculate the output result. This process is called the learning process of the neural network. This principle enables the neural network model to have generalization ability, that is, during the training process, it can also provide correct responses to input values not given by the neural network. In this way, the established model can be used to make reliable predictions for a wide range of input parameters. In the present invention, taking advantage of the characteristic that the neural network can make reliable predictions, when only limited transistor parameter data can be measured, the S and noise parameters of the transistor under other different conditions can be obtained quickly and accurately, and a complete S2P file can be obtained, so as to quickly and accurately establish a low-noise amplifier that meets the requirements. This method is very beneficial to the design and analysis of low-noise amplifiers.

[0094] In order to design the LNA circuit more precisely, the present invention constructs a new type of hybrid neural network structure, as Figure 5 shown. This structure includes 3 neural network structures. For ANN1, it is a neural network for training S parameters. The inputs are the drain-source current (Ids), drain-source voltage (Vds) and frequency (f), and the outputs are the scattering parameters, including S 11 , S 12 , S 21 , S 22 . The relationship between the input neurons and the output neurons is expressed by a formula as shown in formula (1).

[0095]

[0096] For ANN2, the input is frequency and the output is the intrinsic parameters and equivalent temperature of the EPC equivalent circuit model. ANN3 is a neural network for noise parameters. The input of ANN3 is the noise parameters initially obtained through the learning of ANN2 neural network. Then ANN3 plays a correction role and the output is the minimum noise figure (NF min ), the optimal reflection coefficient (Mag(Γ opt ), Ang(Γ opt )) and the thermal resistance R n . The relationship between the input neurons and the output neurons is expressed by formula (2).

[0097] [F min , Mag(Γ opt ), Ang(Γ opt ), R n = F 2 [f] (2)

[0098] The construction process of the above method is as shown in the appendix Figure 6 .

[0099] Among them, the training and learning of this hybrid neural network are both completed in the matlab development environment.

[0100] In one implementation, the first neural network structure ANN1, the second neural network structure ANN2, and the third neural network structure ANN3 respectively include an input layer, a hidden layer, and an output layer. Among them, the hidden layer structure of the first neural network structure ANN1 is 9-8; the hidden layer structure of the third neural network structure ANN3 is 9-7.

[0101] The present invention uses the mean square error to evaluate the accuracy of neural network prediction. Formula (3) is the expression of the mean square error.

[0102]

[0103] In the above formula, y i represents the i-th original data, and d i represents the i-th number obtained by neural network fitting. N represents the total number of tests.

[0104] By comparing the average mean square error of the fitting results of S and noise parameters under different hidden layer structures, it can be seen from Tables 2 and 3 that when the hidden layer structure is 9-8, the average error of the S parameter is the smallest. Therefore, the hidden layer of 9-8 is selected as the structure of the S parameter neural network; when the hidden layer structure is 9-7, the average error of the noise parameters is the smallest. Therefore, the hidden layer of 9-7 is selected as the structure of the noise neural network.

[0105] Table 2: Average mean square error of different hidden layer structures of S parameters

[0106] Hidden layer S-parameter average mean square error (%) 9--8 1.201025 9--5 5.871675 8--6 4.5631 9--6--6 4.252575 8 6.710488

[0107] Table 3: Average Mean Square Error of Different Hidden Layer Structures with Noise Parameters

[0108]

[0109] It should be noted that a neural network includes an input layer, a hidden layer, and an output layer. The hidden layer structure can have 1, 2, or 3 layers. The appropriate number of intermediate layers is selected according to the model accuracy. For example, 9-7 represents that the number of intermediate hidden layers of the neural network is 2, and the number of neurons in the two layers is 9 and 7 respectively; 9-6-6 represents that the number of intermediate hidden layers is 3, and the number of neurons in each layer is 9, 6, and 6 respectively.

[0110] The neural network is trained under the conditions of V ds = 3V, I ds = 70 mA, V ds = 3V, I ds = 90 mA and V ds = 4V, I ds = 70 mA, and the obtained S is fitted with the actual test data. The comparison results of the S-parameter ANN model data and the actual test values under different bias conditions are shown in the appendix Figures 7 - 9 As shown. It can be seen from the figure that the fitting degree of the graph is good, and the average error is less than 2%.

[0111] Among them, the appendix Figure 7 is for V ds = 3V, I ds = 70 mA, and the average errors of the S parameters are E11 = 0.93%, E21 = 1.8%, E12 = 0.3%, E22 = 1.39%

[0112] The appendix Figure 8 is for V ds = 3V, I ds = 90 mA, and the average errors of the S parameters are E11 = 0.11%, E21 = 1.12%, E12 = 0.44%, E22 = 1.04%;

[0113] The appendix Figure 9 is for V ds = 4V, I ds = 70 mA, and the average errors of the S parameters are E11 = 0.02%, E21 = 1%, E12 = 0.55%, E22 = 0.86%;

[0114] S3 performs stability analysis based on the obtained S2P file and further completes the design of the low-noise amplifier.

[0115] Among them, the S2P file contains the S and noise parameter data used to design the HEMT device of the low-noise amplifier, and can replace the transistor to design the low-noise amplifier.

[0116] See Figure 10 , which shows a schematic diagram of the design steps of a low-noise amplifier.

[0117] In one embodiment, step S3 specifically includes:

[0118] S31 Perform stability analysis based on the obtained S2P file, and design a negative feedback circuit according to the stability analysis result;

[0119] S32 Design a matching circuit, where the matching circuit includes input matching and output matching;

[0120] S33 Perform overall optimization of the circuit to complete the design of the low-noise amplifier.

[0121] In one embodiment, step S31 includes:

[0122] The stability of the radio frequency amplification circuit is an important factor for maintaining the normal operation of the communication system. In the design of the radio frequency amplification circuit, it is necessary to consider the stability of the circuit under different working conditions. The higher the stability, the better the circuit performance. When a radio frequency amplification circuit becomes unstable, the circuit will not be able to complete the normal amplification function, but instead exhibit behaviors similar to those of an oscillator circuit.

[0123] Calculate the stability factor K of the radio frequency noise amplifier circuit, where where S 11 represents the input reflection coefficient under the condition of terminating the output with a match, S 21 represents the forward transmission gain under the condition of terminating the output with a match, S 22 represents the output reflection coefficient under the condition of terminating the input with a match, S 12 represents the reverse transmission gain under the condition of terminating the input with a match;

[0124] When K < 1, introduce negative feedback to make the circuit reach a stable state, where the negative feedback is achieved by connecting two inductors in series at the source; by continuously adjusting the size of the inductor, the circuit is made to satisfy K > 1 to meet the stability requirements.

[0125] Appendix Figure 11 is the structural diagram for designing the LNA. From Appendix Figure 11 it can be seen that by continuously adjusting Ls, a suitable value can be found to make the circuit satisfy K > 1 and ensure good circuit performance.

[0126] In one embodiment, step S32 includes:

[0127] The input end is matched for the minimum noise figure, and the center point of the minimum noise figure circle is selected as the optimal reflection source coefficient and matched to 50Ω; the output end is conjugated to 50Ω for the maximum gain matching.

[0128] When designing a low-noise amplifier, the matching circuit is an essential part. By adding a matching circuit, the normal operation of the circuit can be ensured, and various parameter indicators of the low-noise amplifier can meet the production requirements. The matching circuit includes input matching and output matching, and different products have different requirements for the matching circuit. In the present invention, when designing a low-noise amplifier, the noise figure determines the performance of the system, so the noise figure should be given priority when performing circuit matching. The input end is matched for the minimum noise figure, and the center point of the minimum noise figure circle is selected as the optimal reflection source coefficient and matched to 50Ω; the output matching circuit is mainly used to improve the gain, improve the gain flatness and the output standing wave ratio, so the output end is conjugated to 50Ω for the maximum gain matching.

[0129] In one embodiment, the present application also provides an exemplary embodiment of a radio frequency low-noise amplifier setting to further illustrate the method proposed by the present invention.

[0130] Among them, the technical indicators of the low-noise amplifier to be designed meet the following conditions:

[0131] The minimum noise figure NFmin < 0.75dB;

[0132] Gain: S 21 > 12dB;

[0133] Input and output standing wave ratios: VSWRin < 1.5, VSWRout < 1.5

[0134] Bias conditions: 3V, 70mA, T = 25°C;

[0135] Frequency band range: 2GHz ≤ f ≤ 4GHz;

[0136] Center frequency: 3.5GHZ

[0137] Through the trained neural network, the S-parameters and noise parameters of GaN at 3V and 70mA are obtained, and an S2P file under this bias condition is constructed using these parameters. According to the proposed method for designing a low-noise amplifier, the circuit schematic diagram of the LNA can be constructed, as shown in the appendix Figure 12 as shown.

[0138] When Ls = 0.5nH, the circuit reaches absolute stability, and K > 1 is satisfied within the frequency band range. The results are as shown in the appendix Figure 13 as shown, Figure 13The left side shows the stability schematic diagram before adding negative feedback (i.e., the unstable state), and the right side shows the stability schematic diagram after adding negative feedback (i.e., the stable state).

[0139] Next, the present invention uses a CL matching circuit to match the input and output ports. From the attached Figure 14 It can be seen that the noise figure nf(2) of the entire circuit is equal to NF at 3.5 GHz min And is equal to 0.715 dB < 0.75 dB, indicating that the noise figure at this point has reached optimization, that is, the input matching is completed, and the design specifications are met. From the attached Figure 15 It can be seen that the output impedance is matched to 50 Ω, and the output matching is completed. Finally, the Tuning tool in ADS is used to finely tune and optimize the circuit to complete the accurate and rapid matching of the input and output ports.

[0140] The operating schematic diagram can obtain the S-parameters under this condition, as shown in the attached Figure 16 Schematic diagram of the S-parameters of the embodiment shown when the operating frequency f = 3.5 GHz of the radio frequency low-noise amplifier circuit. At 3.5 GHz, S 21 = 12.319 dB > 12 dB, VSWRin = 1.2 < 1.5, VSWRout = 1.02 < 1.5 meet the design specifications. Through the above steps, a low-noise amplifier that meets the actual requirements is designed.

[0141] The results show that as long as the S2P file of a device is known, the method mentioned in the present invention can be used to design a low-noise amplifier. In addition, when the data in the S2P file is insufficient, the neural network designed by the present invention can also be used to improve the S2P file. This makes the application scope of the technical solution of the present invention wider and has general applicability.

[0142] The present invention combines the neural network model with the low-noise amplifier model, and uses the ANN model to predict the scattering and noise parameters of the transistor at different frequencies and biases. These data can be used to establish a complete S2P file of the transistor, and then the S2P file is used to design a low-noise amplifier, so as to quickly obtain the S and noise parameters of the low-noise amplifier.

[0143] When measuring the noise parameters of a real device, engineers need to adapt the two-port network. When a certain condition changes, the adaptation process needs to be carried out again. This cumbersome process is very unfavorable for engineers to develop radio frequency circuits. By constructing a neural network of the S and noise parameters of the device, the present invention can accurately and quickly obtain the S and noise parameters of the low-noise amplifier under different frequencies and bias conditions through direct and limited test data. These advantages greatly improve the efficiency of analyzing and designing low-noise amplifiers and are beneficial to establishing an efficient communication system.

[0144] Meanwhile, the present invention also provides a radio frequency low noise amplifier circuit design device based on a hybrid neural network, which is used to implement the specific implementation manners corresponding to the method steps in the radio frequency low noise amplifier circuit design method based on a hybrid neural network as described above. This application will not repeat the description here. Figure 1 Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments described herein can be implemented by hardware, software, firmware, middleware, code, or any appropriate combination thereof. For hardware implementation, the processor can be implemented in one or more of the following units: application specific integrated circuit (ASIC), digital signal processor (DSP), digital signal processing device (DSPD), programmable logic device (PLD), field programmable gate array (FPGA), processor, controller, microcontroller, microprocessor, or other electronic units designed to implement the functions described herein, or a combination thereof. For software implementation, part or all of the processes of the embodiments can be completed by instructing relevant hardware through a computer program. When implemented, the above program can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. The computer-readable medium includes computer storage media and communication media, where the communication media includes any medium that facilitates the transmission of a computer program from one place to another. The storage media can be any available medium that can be accessed by a computer. The computer-readable medium can include, but is not limited to, RAM, ROM, EEPROM, CD-ROM, or other optical disk storage, magnetic disk storage media, or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer.

[0145] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the protection scope of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the essence and scope of the technical solutions of the present invention.

[0146] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the protection scope of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A design method for a radio frequency low-noise amplifier circuit based on a hybrid neural network, characterized in that, it includes: S1 Select appropriate devices according to performance index requirements, and determine the corresponding operating states and bias conditions; S2 uses the constructed hybrid neural network to obtain the scattering parameters and noise parameters of the transistor, and acquire the S2P file of the device under specified conditions; the constructed hybrid neural network includes 3 neural network structures, and its inputs are drain-source current, drain-source voltage and frequency. Among them, the first neural network structure ANN1 is used to obtain and output the scattering parameters according to the input drain-source current, drain-source voltage and frequency. The scattering parameters include the magnitude and angle of S 11 , S 12 , S 21 , S 22 . Among them, S 11 represents the input reflection coefficient under the condition that the output end is terminated with a match, S 21 represents the forward transmission gain under the condition that the output end is terminated with a match, S 22 represents the output reflection coefficient under the condition that the input end is terminated with a match, S 12 represents the reverse transmission gain under the condition that the input end is terminated with a match; the second neural network structure ANN2 is used to obtain and output the intrinsic parameters and equivalent temperature of the EPC equivalent circuit model according to the input frequency; the third neural network structure ANN3 is used to obtain and output the noise parameters according to the input drain-source current, drain-source voltage and frequency. The noise parameters include the minimum noise figure, the optimum reflection source coefficient and the thermal resistance; S3 Perform stability analysis based on the obtained S2P file, and further complete the design of the low-noise amplifier, specifically including: S31 Perform stability analysis based on the obtained S2P file, and design a negative feedback circuit according to the stability analysis results; S32 Design a matching circuit, where the matching circuit includes input matching and output matching; S33 Perform overall optimization of the circuit to complete the design of the low-noise amplifier.

2. The design method for a radio frequency low-noise amplifier circuit based on a hybrid neural network according to claim 1, characterized in that, in step S1, the determined operating states include temperature; the bias conditions include frequency, drain-source voltage and drain-source current of the device.

3. The design method for a radio frequency low-noise amplifier circuit based on a hybrid neural network according to claim 1, characterized in that, The first neural network structure ANN1, the second neural network structure ANN2, and the third neural network structure ANN3 respectively include an input layer, a hidden layer, and an output layer, where the hidden layer structure of the first neural network structure ANN1 is 9-8; the hidden layer structure of the third neural network structure ANN3 is 9-7.

4. The design method for a radio frequency low-noise amplifier circuit based on a hybrid neural network according to claim 1, characterized in that, before step S2, it further includes: Sb1 Train the hybrid neural network, including: Select the directly measured data of AlGaN HEMT devices on an Si substrate as samples for neural network training, where the samples include training samples and test samples, and the samples contain scattering parameters measured under different biases of the devices and noise parameters measured at different frequencies; where the values of the scattering parameters in the samples are quickly measured by an instrument; The acquisition of the noise parameters in the samples includes: Analyze the device under test using the noise temperature, which is represented in an equivalent small-signal model. The equivalent small-signal EPC noise model includes the device intrinsic parameters C gs , r gs , g m , g ds and C gd , where C gs represents the capacitance between the gate and the source, r gs represents the gate-source resistance, g m is the transconductance, g ds represents the conductance between the drain and the source, C gd represents the capacitance between the gate and the drain; the equivalent temperature parameters include τ, T g , T d and T c , where T g represents the gate noise voltage source introduced at the gate Gate, T d represents the drain noise current source, T c represents the correlation temperature between the two, and τ represents the correlation coefficient; Calculate the noise parameters according to the intrinsic parameters and equivalent temperature parameters of the device.

5. The design method for a radio frequency low-noise amplifier circuit based on a hybrid neural network according to claim 1, characterized in that, step S31 includes: Calculate the stability factor K of the radio frequency noise amplifier circuit, where where S 11 represents the input reflection coefficient under the condition of output terminal termination matching, S 21 represents the forward transmission gain under the condition of output terminal termination matching, S 22 represents the output reflection coefficient under the condition of input terminal termination matching, S 12 represents the reverse transmission gain under the condition of input terminal termination matching; When K < 1, introduce negative feedback to make the circuit reach a stable state, where the negative feedback is realized by connecting inductors in series at two sources; by continuously adjusting the size of the inductor, so that the circuit satisfies K > 1 to meet the stability requirements.

6. The design method for a radio frequency low-noise amplifier circuit based on a hybrid neural network according to claim 1, characterized in that, step S32 includes: Perform matching with the minimum noise figure at the input end, select the center point of the minimum noise figure circle as the best reflection source coefficient and match it to 50Ω; perform matching with the maximum gain at the output end by conjugate matching to 50Ω.

7. A design device for a radio frequency low-noise amplifier circuit based on a hybrid neural network, characterized in that, this device is used to implement the design method for a radio frequency low-noise amplifier circuit based on a hybrid neural network as described in any one of claims 1-6 above.

Images