Machine learning assisted metasurface loading decoupling patch antenna design method

Abstract

The invention discloses a machine learning-assisted metasurface loading and decoupling patch antenna design method, particularly relates to the technical field of antenna design, and solves the problem that in the prior art, a metasurface loading and decoupling patch antenna design method can only realize port decoupling but cannot realize directional diagram decoupling. And the design method needs complex theoretical analysis calculation and time-consuming parameter optimization. According to the technical scheme, a data set composed of a pixelated metasurface and electromagnetic response is used for training an agent model, and a parameter optimization algorithm is combined with the trained agent model to update parameters according to a cost function; according to the method, the problem of simultaneous decoupling of the port and the directional diagram can be solved, and complex theoretical analysis calculation and time-consuming parameter optimization can be avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of antenna design, and in particular to a machine learning-assisted metasurface loaded decoupled patch antenna design method. Background Art

[0002] The advent of the Internet of Everything era has made large-scale multi-input multi-output antenna technology widely used in mobile communication systems, but the multipath effect caused by the mutual coupling between multiple antennas will affect the normal radiation of the antenna unit. Therefore, the decoupling technology between antennas has gradually become a hot topic in the field of antenna research. Due to its advantages such as low profile, flexible design, flexible beam control, low cost, and light weight, the metasurface structure has shown great application potential in antenna decoupling. However, the metasurface-loaded decoupling patch antenna requires complex theoretical analysis and time-consuming parameter optimization to accurately control the amplitude and phase of each unit beam to achieve the decoupling function. With the rapid development of artificial intelligence technology, the antenna modeling and design method assisted by machine learning can avoid complex theoretical analysis and reduce the time-consuming parameter optimization process, and realize efficient and accurate antenna design. Therefore, the modeling and design of low-mutual-coupling patch antennas loaded on metasurfaces assisted by machine learning has certain research value.

[0003] There is no design method based on machine learning for the existing metasurface-loaded decoupling patch antenna. The currently reported design methods for metasurface-loaded decoupling patch antennas are mainly divided into two types: the first method is to use double-sided metal structures of different shapes to form a metasurface with band-stop characteristics, and suppress the coupled waves in the working frequency band through the metasurface to achieve the reduction of port coupling; the second method is to use a metasurface composed of single-sided metal structures of different shapes to control the amplitude and phase of the electromagnetic wave, and cancel the amplitude and phase of the electromagnetic wave in the original coupling path, so as to achieve low port coupling between antenna units. The above-mentioned metasurface-loaded low mutual coupling patch antenna design methods can only achieve port decoupling but not pattern decoupling, and the design process is complicated, requiring rich design experience and long-term parameter optimization. Therefore, it is necessary to propose a design method for metasurface-loaded low mutual coupling patch antenna, which can achieve the decoupling of ports and patterns while avoiding complex theoretical analysis and calculation and time-consuming parameter optimization. Summary of the invention

[0004] Therefore, the present invention solves the technical problems that the design method of the metasurface loaded decoupling patch antenna in the prior art can only realize port decoupling but cannot realize directional pattern decoupling, and the design method requires complex theoretical analysis calculations and time-consuming parameter optimization; the present invention provides a machine learning-assisted metasurface loaded decoupling patch antenna design method, which solves the difficult problem of simultaneous decoupling of ports and directional patterns, and can avoid complex theoretical analysis calculations and time-consuming parameter optimization.

[0005] The present invention provides a machine learning assisted metasurface loaded decoupled patch antenna design method, which consists of two parts. The first part is to establish a proxy model and train the proxy model using a data set D, and the second part is parameter optimization.

[0006] Furthermore, the first part is mainly divided into three steps. Step 1 is to pixelate the metasurface structure and form a parameter matrix w; Step 2 is to use the parameter matrix w to model and simulate the electromagnetic response r through full-wave simulation software, and the parameter matrix w and the electromagnetic response r are combined to form a data set D; Step 3 is to build a proxy model through an artificial neural network (ANN) and train the ANN using the data set D.

[0007] Furthermore, a metal design area is planned on the dielectric substrate, and the metal design area is divided into 6 rows, 3 columns and 18 sub-areas for the design of the metasurface unit. Then, the metasurface unit is pixelated and divided into a pixel grid of 8 rows and 8 columns. In order to meet the requirements of symmetrical radiation of the antenna, the pixel grid of 8 rows and 8 columns can be simplified to 4 rows and 4 columns symmetrically about the central symmetry line. Finally, the 4 rows and 4 columns of the pixel grid are represented by a binary parameter matrix w, where "1" indicates that the grid is filled with metal and "0" indicates that the grid is filled with air. The electromagnetic response r is mainly composed of three key parameters, namely, the antenna's |S 11 |Parameters, |S 21 |Parameters and gain parameters of the radiation pattern at the operating frequency from -45° to 45°. In the process of training the ANN agent model using the data set D composed of the parameter matrix w and the electromagnetic response r, the parameter matrix w is the input data and the electromagnetic response r is the output data.

[0008] Furthermore, the second part consists of four steps. Step 1 defines the maximum number of iterations of the optimization algorithm L. max , the target cost function C target , and randomly generate particle parameters; Step 2 uses the proxy model to predict the electromagnetic response through the randomly generated particle parameters and calculate the cost function C; Step 3 compares the cost function C calculated in Step 2 with the target cost function C target Compare, if C is less than C target , then output the existing result, otherwise, before reaching the maximum number of iterations, proceed to step 4, update the particle parameters, and repeat step 2. The cost function C can be expressed by formula (1):

[0009]

[0010] Among them, S 11 |S is the operating frequency 11 |Parameter, S 21|S is the operating frequency 21 |Parameter, S (k) and S (-k) are the gains of the radiation patterns at the operating frequency at k° and (-k)° respectively, and p is the weight coefficient.

[0011] Finally, guided by the machine learning design method, a metasurface-loaded patch antenna with simultaneous decoupling of ports and radiation patterns was designed without complex theoretical analysis and calculations and time-consuming parameter optimization processes.

[0012] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0013] 1. The present invention provides a machine learning-assisted metasurface-loaded decoupled patch antenna design method, which trains a proxy model through a data set consisting of a pixelated metasurface and an electromagnetic response. A parameter optimization algorithm is combined with the trained proxy model to update parameters according to a cost function, and finally a metasurface-loaded patch antenna with simultaneous decoupling of ports and radiation patterns is realized.

[0014] 2. The present invention provides a machine learning-assisted metasurface loaded decoupled patch antenna design method, the cost function is designed as It can ensure that the final output result of the optimization algorithm can not only pay attention to the response requirements of the S parameters, but also achieve directional pattern decoupling.

[0015] 3. The present invention provides a machine learning-assisted metasurface loaded decoupled patch antenna design method. Compared with the prior art, the present invention can simultaneously achieve port decoupling and pattern decoupling, and can avoid complex theoretical analysis calculations and time-consuming parameter optimization processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0017] Figure 1 This is a flow chart of the method for designing a decoupled patch antenna loaded with machine learning assistance of the present invention;

[0018] Figure 2 It is a schematic diagram of the parameter matrix design process of the present invention;

[0019] Figure 3 Schematic diagram of the structure of the low mutual coupling patch antenna loaded on the metasurface, where (a) is a side view, (b) is an antenna array diagram, (c) is a metasurface structure diagram, and (d) is a metasurface unit and binary matrix T diagram;

[0020] Figure 4 The predicted and simulated S-parameter graphs for the antenna;

[0021] Figure 5 Predict and simulate the H-plane radiation pattern of the antenna at 5GHz;

[0022] Description of reference numerals:

[0023] 1. Metal design area; 2. Super surface unit. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0025] Embodiment 1:

[0026] This embodiment provides a machine learning-assisted metasurface loaded decoupled patch antenna design method, the flow chart is as follows Figure 1 As shown in FIG, the design method consists of two parts. The first part is to establish a proxy model and train the proxy model using the data set D. The second part is to optimize the parameters.

[0027] In this embodiment, the first part is mainly divided into three steps. Step 1 is to pixelate the metasurface structure and form a parameter matrix w; Step 2 is to use the parameter matrix w to model and simulate the electromagnetic response r through full-wave simulation software, and the parameter matrix w and the electromagnetic response r are combined to form a data set D; Step 3 is to build a proxy model through an artificial neural network (ANN) and use the data set D to train the ANN.

[0028] In this embodiment, Figure 2 Figure 1 is a schematic diagram of the parameter matrix design process, (a) is the metasurface design area, (b) is the unit pixelation, and (c) is the parameter matrix w. The metal design area 1 is planned on the dielectric substrate, and the metal design area is divided into 18 sub-areas with 6 rows and 3 columns for the design of the metasurface unit 2. Next, the metasurface unit is pixelated and divided into a pixel grid with 8 rows and 8 columns. In order to meet the requirement of symmetrical radiation of the antenna, the pixel grid with 8 rows and 8 columns can be simplified to 4 rows and 4 columns symmetrically about the central symmetry line. Finally, the 4 rows and 4 columns of the pixel grid are represented by the binary parameter matrix w, where "1" indicates that the grid is filled with metal and "0" indicates that the grid is filled with air. The electromagnetic response r is mainly composed of three key parameters, namely, the antenna's |S 11 |Parameters, |S 21|Parameters and the gain parameters of the radiation pattern at the working frequency from -45° to 45°. Then, the ANN agent model is trained using the data set D consisting of the parameter matrix w and the electromagnetic response r, where the parameter matrix w is the input data and the electromagnetic response r is the output data.

[0029] In this embodiment, the second part consists of four steps. Step 1 defines the maximum number of iterations L of the optimization algorithm. max , the target cost function C target , and randomly generate particle parameters; Step 2 uses the proxy model to predict the electromagnetic response through the randomly generated particle parameters and calculate the cost function C; Step 3 compares the cost function C calculated in Step 2 with the target cost function C target Compare, if C is less than C target , then output the existing result, otherwise, before reaching the maximum number of iterations, proceed to step 4, update the particle parameters, and repeat step 2. The cost function C can be expressed by formula (1):

[0030]

[0031] Among them, S 11 |S is the operating frequency 11 |Parameter, S 21 |S is the operating frequency 21 |Parameter, S (k) and S (-k) are the gains of the radiation patterns at the operating frequency at k° and (-k)° respectively, and p is the weight coefficient.

[0032] Embodiment 2:

[0033] This embodiment provides an antenna designed by the above method, and its antenna structure schematic diagram is as follows: Figure 3 As shown, the center spacing and side spacing of the two patch antenna units are 0.35λ0 and 0.05λ0 respectively (λ0 is the free space wavelength corresponding to the center frequency). Figure 4 The S parameter curves of the traditional 1×2 patch antenna and the metasurface loaded 1×2 patch antenna with the same center spacing and side spacing. Figure 4 It can be seen that the mutual coupling level of the traditional 1×2 patch antenna is -15.7dB, while the mutual coupling level of the metasurface-loaded 1×2 patch antenna of the present invention within the matching frequency band is lower than -20dB, which shows that the overall decoupling effect is significantly improved. The center frequency of this implementation case is 5GHz, the -10dB impedance matching frequency band is 4.84GHz to 5.16GHz, the relative bandwidth is 6.4%, and the antenna has a |S 21 |All below -20dB. Figure 5The unit H-plane radiation pattern of the traditional 1×2 patch antenna and the metasurface-loaded 1×2 patch antenna at 5GHz. It can be seen that the radiation direction of the metasurface-loaded 1×2 patch antenna points to the side-ray direction, and the problem of radiation pattern distortion has been significantly improved. The dielectric substrate used in this case is RO5880 substrate, and the radiator size is 0.88λ0×0.47λ0×0.21λ0.

[0034] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A machine learning-assisted metasurface loaded decoupled patch antenna design method, characterized in that: The steps include: S1: Get data set D; S2: Building antenna proxy model through artificial neural network; S3: Train the proxy model using the dataset D to obtain the trained proxy model; S4: Optimize the parameters of the trained proxy model.

2. The method for designing a metasurface loaded decoupled patch antenna assisted by machine learning according to claim 1, characterized in that: The acquisition of the data set D in step S1 includes the following steps: S11: pixelate the metasurface structure and form a parameter matrix w; S12: Use the parameter matrix w to perform modeling and simulation to obtain the electromagnetic response r, and combine the parameter matrix w and the electromagnetic response r to form a data set D.

3. The machine learning-assisted metasurface loaded decoupled patch antenna design method according to claim 2, characterized in that: The parameter matrix design in step S11 includes the following steps: S111: planning a metal design area on the antenna dielectric substrate, dividing the metal design area (1) into 18 sub-areas of 6 rows and 3 columns for designing the metasurface unit (2); S112: pixelating the metasurface structure into a pixel grid of 8 rows and 8 columns, and simplifying the pixel grid of 8 rows and 8 columns symmetrically about a central symmetry line into 4 rows and 4 columns; S113: The binary parameter matrix w represents a 4-row 4-column pixel grid, where "1" indicates that the grid is filled with metal and "0" indicates that the grid is filled with air.

4. The method for designing a metasurface loaded decoupled patch antenna assisted by machine learning according to claim 3, characterized in that: The electromagnetic response r in step S12 is mainly composed of three key parameters, namely, the antenna's |S 11 |Parameters, |S 21 |Parameters and gain parameters of the radiation pattern from -45° to 45° at the operating frequency.

5. The method for designing a metasurface loaded decoupled patch antenna assisted by machine learning according to claim 4, characterized in that: In the process of training the proxy model using the data set D in step S3, the parameter matrix w is the input data and the electromagnetic response r is the output data.

6. The method for designing a metasurface loaded decoupled patch antenna assisted by machine learning according to claim 5, characterized in that: The optimization of parameters in step S4 includes the following steps: S41: Define the maximum number of iterations L of the optimization algorithm max , the target cost function C target , and randomly generate particle parameters; S42: using the proxy model to predict the electromagnetic response through randomly generated particle parameters and calculate the cost function C; S43: The cost function C calculated in step S42 and the target cost function C target Compare, if C is less than C target , then output the existing result, otherwise proceed to the next step before reaching the maximum number of iterations; S44: Update the particle parameters and repeat step S42.

7. The method for designing a metasurface loaded decoupled patch antenna assisted by machine learning according to claim 6, characterized in that: The cost function C in step S42 is expressed by the following formula: Among them, S 11 |S is the operating frequency 11 |Parameter, S 21 |S is the operating frequency 21 |Parameter, S (k) and S (-k) are the gains of the radiation patterns at the operating frequency at k° and (-k)° respectively, and p is the weight coefficient.

8. The method for designing a metasurface loaded decoupled patch antenna assisted by machine learning according to claim 7, characterized in that: In step S12, full-wave simulation software is used to model and simulate the electromagnetic response r using the parameter matrix w.

Images