A method for modeling radio frequency nonlinear behavior based on spectral mapping

Abstract

The application provides a radio frequency nonlinear behavior modeling method based on spectrum mapping, comprising: approximating input signals and output signals by a set of superimposed harmonic signals, so as to convert the nonlinear mapping relationship between the input signals and the output signals into the mapping relationship between the input signal spectrum and the output signal spectrum; generating a set of fixed harmonic signal frequencies of the input signals and the output signals based on spectrum discretization, so as to establish a spectrum mapping neural network model with a unified structure, and train the spectrum mapping neural network model; and obtaining the mapping relationship between the input signals and the output signals by using the trained spectrum mapping neural network model. The application is suitable for the nonlinear behavior model construction of radio frequency units at different levels such as devices, IPs, components, etc. The behavior model constructed based on the application can not only accurately represent various radio frequency nonlinear characteristics, but also has low model complexity and high calculation efficiency, and can meet the modeling and simulation requirements of multi-level radio frequency systems.

Description

Technical Field

[0001] This invention relates to the field of radio frequency nonlinear modeling and simulation technology, and more specifically, to a modeling method for radio frequency nonlinear behavior based on spectrum mapping. Background Technology

[0002] As the integration and complexity of electronic devices continue to increase, behavioral models are becoming increasingly important in multi-level simulations of RF systems, encompassing devices, components, and systems. Compared to equivalent circuit models and physical models, behavioral models do not require extensive knowledge of internal structures and working mechanisms, resulting in lower model complexity and greater flexibility in use.

[0003] In microwave and radio frequency (RF) device behavior modeling techniques, nonlinear modeling is both a research hotspot and a challenge. Traditional modeling methods mainly include polynomial models, Volterra series models, and polyharmonic distortion (PHD) models. Keysight Technologies has proposed the X-parameter model based on the PHD model, which has become the industry standard for large-signal nonlinear behavior models in engineering applications. In recent years, the development of machine learning technology has made the construction of RF behavior models based on artificial neural network algorithms an important research direction.

[0004] When performing multi-level RF system simulation, it is necessary not only to model individual devices but also to construct behavioral models for RF units such as IPs and components formed by interconnecting multiple devices and integrated process models. In system-level simulation, RF unit behavioral models can not only shield system designers from low-level circuit details but also improve system-level simulation efficiency. However, traditional modeling techniques often focus on building models for individual devices, and the extraction of model parameters such as X-parameters and Volterra series is complex and computationally intensive, making it difficult to meet the needs of multi-level RF system modeling and simulation. Summary of the Invention

[0005] The present invention aims to provide a modeling method for radio frequency nonlinear behavior based on spectrum mapping, so as to solve the problem that traditional methods are difficult to adapt to the simulation of multi-level radio frequency systems.

[0006] This invention provides a method for modeling radio frequency nonlinear behavior based on spectrum mapping, comprising the following steps:

[0007] Step 1: Use a superposition of a set of simple harmonic signals to approximate the input and output signals, thereby converting the nonlinear mapping relationship between the input and output signals into a mapping relationship between the spectrum of the input signal and the spectrum of the output signal;

[0008] Step 2: Generate a set of fixed simple harmonic signal frequencies for the input and output signals based on spectrum discretization, thereby establishing a spectrum mapping neural network model with a unified structure, and training the spectrum mapping neural network model;

[0009] Step 3: Use the trained spectrum mapping neural network model to obtain the mapping relationship between the input signal and the output signal.

[0010] Furthermore, step one includes:

[0011] Input signal V in (t) and output signal V out (t) can be approximated by a set of simple harmonic signals:

[0012]

[0013]

[0014] According to equations (1) and (2), the input signal V in (t) and output signal V out The mapping relationship of (t) is characterized by the following equation:

[0015]

[0016] Among them, V in_k ω in_k , (k = 0, 1…M) represent the amplitude, frequency, and phase of each harmonic signal used to characterize the input signal, respectively, V out_k ω out_k , (k = 0, 1, ..., N) represent the amplitude, frequency, and phase of each harmonic signal used to characterize the output signal, respectively.

[0017] Furthermore, step two includes:

[0018] First, the input signal spectrum and output signal spectrum of the object to be modeled are discretized at equal intervals, as shown in equations (4) and (5):

[0019] ω in_k =ω in_start +k*Δω in k = 0, 1…M (4)

[0020] ω out_k =ω out_start +k*Δω out k = 0, 1…N (5)

[0021] Where, Δω in and Δω outThe frequency resolutions of the input signal spectrum and the output signal spectrum, respectively, are expressed as:

[0022]

[0023]

[0024] Then, substituting equations (4) and (5) into equations (1) and (2) yields:

[0025]

[0026]

[0027] Using the set of fixed harmonic signals shown in equations (8) and (9) to characterize the input and output signals, the spectral mapping relationship shown in equation (3) is rewritten as:

[0028]

[0029] When using the BP neural network algorithm to fit the spectral mapping of the input signal and the output signal, a fixed-structure spectral mapping neural network model is constructed based on equation (10);

[0030] Finally, an input-output sample set for the spectrum mapping neural network model is established, and the spectrum mapping neural network model is trained using the input-output sample set. The trained spectrum mapping neural network model is then used for spectrum mapping of the input and output signals.

[0031] Furthermore, the input of the spectrum mapping neural network model is the amplitude and phase of the simple harmonic signal of the input signal at each frequency component, as shown on the right side of equation (10); the output of the spectrum mapping neural network model is the amplitude and phase of the simple harmonic signal of the output signal at each frequency component, as shown on the left side of equation (10).

[0032] Furthermore, step three includes:

[0033] First, if the input signal is in time-domain form, the amplitude and phase of the simple harmonic signal at each frequency in its spectrum are obtained using the Fast Fourier Transform. If the input signal is in frequency-domain form, the amplitude and phase of the simple harmonic signal at each frequency in its spectrum are obtained directly.

[0034] Then, the amplitude and phase of the simple harmonic signal at each frequency of the output signal are calculated using the spectrum mapping neural network model;

[0035] Finally, the amplitude and phase of the simple harmonic signal at each frequency are substituted into equation (9) to obtain the output signal.

[0036] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0037] This invention proposes a modeling method for radio frequency nonlinear behavior based on spectrum mapping, which is applicable to the construction of nonlinear behavior models for radio frequency units at different levels such as devices, IPs, and components. The behavior model constructed based on this invention can not only accurately characterize various radio frequency nonlinear characteristics, but also has low model complexity and high computational efficiency, which can meet the modeling and simulation requirements of multi-level radio frequency systems. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a flowchart of the radio frequency nonlinear behavior modeling method based on spectrum mapping in an embodiment of the present invention.

[0040] Figure 2 This is a schematic diagram of the spectrum mapping neural network model constructed in an embodiment of the present invention.

[0041] Figure 3 This is a block diagram illustrating the principle of the frequency conversion IP in an embodiment of the present invention.

[0042] Figure 4 This is a graph comparing the actual value of the output signal power spectrum of the frequency conversion IP in this embodiment of the invention with the predicted value based on the behavioral model of this invention.

[0043] Figure 5 This is a graph comparing the actual value of the phase spectrum of the frequency conversion IP output signal in this embodiment of the invention with the predicted value based on the behavioral model of this invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0045] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0046] Example

[0047] like Figure 1 As shown, this embodiment proposes a modeling method for radio frequency nonlinear behavior based on spectrum mapping, including the following steps:

[0048] Step 1: Use a superposition of a set of simple harmonic signals to approximate the input and output signals, thereby converting the nonlinear mapping relationship between the input and output signals into a mapping relationship between the spectrum of the input signal and the spectrum of the output signal;

[0049] Behavioral models describe the behavioral characteristics of the modeled object by establishing a mapping relationship between input and output. For radio frequency units such as devices, IPs, and components, the behavioral model of the radio frequency unit can be characterized by establishing a nonlinear mapping relationship between its input signal voltage wave and its output signal voltage wave.

[0050] From a time-domain perspective, any voltage wave signal can be constructed or approximated by superimposing a finite number of simple harmonic signals, each representing a different frequency signal component contained in the original signal. From a frequency-domain perspective, a voltage wave signal can be represented as a spectrum with a certain bandwidth, where the amplitude and phase at each frequency point represent the different frequency signal components contained in the original signal. Furthermore, the various nonlinear characteristics of the radio frequency unit can be analyzed from the output signal spectra corresponding to different input signals. Therefore, there is a one-to-one correspondence between the simple harmonic signals in the time domain and the signals at each frequency in the frequency spectrum.

[0051] From the perspective of function approximation, the input signal V in (t) and output signal V out (t) can be approximated by a set of simple harmonic signals:

[0052]

[0053]

[0054] Among them, V in_k ω in_k , (k = 0, 1…M) represent the amplitude, frequency, and phase of each harmonic signal used to characterize the input signal, respectively, V out_k ω out_k , (k = 0, 1…N) represent the amplitude, frequency, and phase of each harmonic signal used to characterize the output signal, respectively;

[0055] According to equations (1) and (2), the input signal V in (t) and output signal V out The mapping relationship of (t) is characterized by the following equation:

[0056]

[0057] The right side of equation (3) and the frequency points ω in the input signal spectrum in_k The amplitude V of the signal at the location in_k and phase (k=0,1…M) correspond one-to-one, and the left side of equation (3) corresponds to each frequency point ω in the spectrum of the output signal. out_k The amplitude V of the signal at the location out_k and phase (k=0,1…N) are in one-to-one correspondence; therefore, equation (3) also represents the mapping relationship between the input signal spectrum and the output signal spectrum.

[0058] Step 2: Generate a set of fixed simple harmonic signals for the input and output signals based on spectrum discretization, thereby establishing a spectrum mapping neural network model with a unified structure, and training the spectrum mapping neural network model;

[0059] When the low noise floor of the signal is ignored, the spectrum of the signal is not continuous. Instead, there are simple harmonic signals at specific frequency points. For example, after a single-tone signal passes through a nonlinear amplifier, the output signal only contains the fundamental signal and each harmonic signal. After passing through a mixer, the output signal will also contain the intermodulation signal of the fundamental signal and the local oscillator signal.

[0060] The frequency of the simple harmonic signal in equations (1) and (2) is not fixed. When the frequency of the input signal changes, the frequency of the simple harmonic signal will also change. This means that a neural network model with a fixed structure cannot be used to fit the mapping relationship of equation (3).

[0061] To address this problem, this invention generates a fixed set of input harmonic signals and a fixed set of output harmonic signals based on spectrum discretization, and uses these as basis functions to approximate the input and output signals, thereby establishing a spectrum mapping neural network model based on a unified structure. The specific implementation is as follows:

[0062] First, the input signal spectrum and output signal spectrum of the object to be modeled are discretized at equal intervals, as shown in equations (4) and (5):

[0063] ω in_k =ω in_start +k*Δω in k = 0, 1…M (4)

[0064] ω out_k =ω out_start +k*Δω out k = 0, 1…N (5)

[0065] Where, Δω in and Δω outThe frequency resolutions of the input signal spectrum and the output signal spectrum, respectively, are expressed as:

[0066]

[0067]

[0068] The frequency resolution directly affects the accuracy of the model and the amount of computation. Higher resolution means higher model accuracy, but it also increases the amount of computation during model training and use. Therefore, the frequency resolution should be determined based on the accuracy requirements and computing power limitations of the actual problem.

[0069] Then, substituting equations (4) and (5) into equations (1) and (2) yields:

[0070]

[0071]

[0072] When the frequency of the input signal is in [ω in_start ,ω in_end When the frequency of the output signal varies within the range of [ω], the frequency of the output signal will be [ω]. out_start ,ω out_end The range changes, and at this time the frequency of the simple harmonic signal in equations (8) and (9) remains constant. That is, the input signal and output signal can be represented by a set of fixed simple harmonic signals shown in equations (8) and (9). Therefore, the spectrum mapping relationship shown in equation (3) can be rewritten as:

[0073]

[0074] When using the BP neural network algorithm to fit the spectral mapping of the input and output signals, a fixed-structure spectral mapping neural network model is constructed based on equation (10), such as... Figure 2 As shown.

[0075] The input to the spectrum mapping neural network model is the amplitude and phase of the simple harmonic signal of the input signal at each frequency component, as shown on the right side of equation (10). Figure 2 In the middle, C represents other parameters that control the working state of the radio frequency unit, such as amplifier gain, filter cutoff frequency and other adjustable parameters of the device; the output of the spectrum mapping neural network model is the amplitude and phase of the simple harmonic signal of the output signal on each frequency component as shown on the left side of equation (10).

[0076] By changing the frequency, amplitude, phase, and other adjustable parameters C of the input signal, the amplitude and phase of each harmonic signal frequency on the corresponding output signal spectrum are obtained, thereby establishing the input and output sample set of the spectrum mapping neural network model. The spectrum mapping neural network model is then trained using the input and output sample set, and the trained spectrum mapping neural network model is used for spectrum mapping of the input and output signals.

[0077] It is worth noting that, since any signal can be constructed or approximated by the superposition of a finite number of simple harmonic signals, equation (8) can approximately characterize the frequency in the range [ω]. in_start ,ω in_end The range includes any signal, including single-tone and multi-tone signals. Therefore, the modeling method described in this invention is applicable to both single-tone and multi-tone input signals.

[0078] Step 3: Obtain the mapping relationship between the input signal and the output signal using the trained spectrum mapping neural network model: First, if the input signal is in time domain form, the amplitude and phase of the simple harmonic signal at each frequency in its spectrum are obtained using the fast Fourier transform. If the input signal is in frequency domain form, the amplitude and phase of the simple harmonic signal at each frequency in its spectrum are obtained directly. Then, the amplitude and phase of the simple harmonic signal at each frequency of the output signal are calculated using the spectrum mapping neural network model. Finally, the amplitude and phase of the simple harmonic signal at each frequency are substituted into equation (9) to obtain the output signal.

[0079] The method for modeling radio frequency nonlinear behavior based on spectrum mapping is verified by taking a certain frequency conversion IP as an example.

[0080] like Figure 3 The frequency conversion IP shown is composed of cascaded components such as amplifiers, filters, mixers, and interconnection process models. This frequency conversion IP realizes the downconversion of 6-7 GHz radio frequency signals to 1-2 GHz intermediate frequency signals.

[0081] Because this frequency conversion IP link contains nonlinear devices such as amplifiers and mixers, the radio frequency signals entering this link will undergo spectrum shifting and generate harmonics, intermodulation, and other signals. To accurately describe the nonlinear behavior characteristics of this frequency conversion IP link, this invention treats it as a single radio frequency unit, and the steps for constructing its nonlinear behavior model are as follows:

[0082] Step 1: Use a superposition of a set of simple harmonic signals to approximate the input and output signals, thereby transforming the nonlinear mapping relationship between the input and output signals into a mapping relationship between the input signal spectrum and the output signal spectrum.

[0083] From the time domain perspective, the input signal voltage wave and the output signal voltage wave of the frequency converter IP can be approximated by the superposition of a finite number of simple harmonic signals, as shown in Equation (1) and Equation (2), respectively. From the frequency domain perspective, each simple harmonic signal of the input signal and the output signal corresponds to the amplitude and phase of each frequency point on the spectrum of the input signal and the spectrum of the output signal, respectively, representing the signal components of different frequencies contained in the original signal.

[0084] Therefore, the mapping relationship between the input signal and the output signal can be converted into the mapping relationship between the spectrum of the input signal and the spectrum of the output signal, as shown in equation (3).

[0085] Since this frequency conversion IP downconverts a 6-7 GHz radio frequency signal to a 1-2 GHz intermediate frequency signal, the simple harmonic frequency ω of the input signal... in_k ∈[2π*6*10 9 2π*7*10 9 To observe the higher harmonics and intermodulation components of the output signal, this embodiment uses the spectrum within a frequency band including the 6th harmonic of the intermediate frequency signal to approximate the output signal, i.e., ω. out_k ∈[0,2π*12](GHz).

[0086] Step 2: Generate a set of fixed simple harmonic signals for the input and output signals based on spectrum discretization, thereby establishing a spectrum mapping neural network model with a unified structure, and training the spectrum mapping neural network model; the specific steps are as follows:

[0087] First, the input signal spectrum and output signal spectrum of the frequency converter IP are discretized at equal intervals, with a frequency resolution of 50MHz. This allows for spectrum discretization according to equations (4), (5), (6), and (7), resulting in the spectrum mapping relationship shown in equation (10). For this embodiment, the values ​​of each variable in the equation are Δω. in =2π*50*10 6 ,Δω in =2π*50*10 6 ω in_start =2π*6*10 9 ω in_end =2π*7*10 9 ω out_start =0, ω out_end =2π*12*10 9 M = 20, N = 240.

[0088] When using the BP neural network algorithm to fit the spectral mapping of the input and output signals, a spectral mapping neural network model for the frequency conversion IP is constructed based on equation (10), as follows: Figure 2As shown. The input of the spectrum mapping neural network model is the amplitude and phase of the simple harmonic signal of the input signal at each frequency component, as shown on the right side of equation (10), and the output of the spectrum mapping neural network model is the amplitude and phase of the simple harmonic signal of the output signal at each frequency component, as shown on the left side of equation (10).

[0089] Training samples are established using simulation results from ADS software. By changing parameters such as the frequency, amplitude, and phase of the input signal and collecting the corresponding output signal spectrum information, the input and output sample set of the spectrum mapping neural network model is established. The spectrum mapping neural network model of the input and output signals is obtained through training.

[0090] Step 3: Obtain the mapping relationship between the input and output signals using the trained spectrum mapping neural network model: In this embodiment, a single-tone signal with a frequency of 6.8 GHz and a power of -15 dBm is used as the input signal of the frequency converter IP for verification. Therefore, only the simple harmonic signal component at 6.8 GHz exists in equation (8), while the simple harmonic signal components at other frequencies can be considered to have amplitudes and phases of 0. Substituting the input signal into the spectrum mapping neural network model, the amplitude and phase of the simple harmonic signal at each frequency of the output signal spectrum can be calculated. The power spectrum and phase spectrum of the frequency converter IP output signal are as follows: Figure 4 and Figure 5 As shown, from Figure 4 and Figure 5 As can be seen from the comparison curves, the behavioral model constructed based on this invention can accurately reflect the nonlinear phenomena such as spectrum shifting, harmonics, and intermodulation generated by the frequency conversion IP on the input signal.

[0091] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for modeling radio frequency nonlinear behavior based on spectrum mapping, characterized in that, Includes the following steps: Step 1: Use a superposition of a set of simple harmonic signals to approximate the input and output signals, thereby converting the nonlinear mapping relationship between the input and output signals into a mapping relationship between the spectrum of the input signal and the spectrum of the output signal; Step 2: Generate a set of fixed simple harmonic signal frequencies for the input and output signals based on spectrum discretization, thereby establishing a spectrum mapping neural network model with a unified structure, and training the spectrum mapping neural network model; Step 3: Use the trained spectrum mapping neural network model to obtain the mapping relationship between the input signal and the output signal; Step one includes: Input signal and output signal Each of these can be approximated using a set of simple harmonic signals: According to equations (1) and (2), the input signal and output signal The mapping relationship is represented by the following formula: in, , , ( These represent the amplitude, frequency, and phase of each harmonic signal used to characterize the input signal, respectively. , , ( ) represent the amplitude, frequency, and phase of each harmonic signal used to characterize the output signal, respectively; Step two includes: First, the input signal spectrum and output signal spectrum of the object to be modeled are discretized at equal intervals, as shown in equations (4) and (5): in, and These are the frequency resolutions of the input signal spectrum and the output signal spectrum, respectively. Then, substituting equations (4) and (5) into equations (1) and (2) yields: Using the set of fixed harmonic signals shown in equations (8) and (9) to characterize the input and output signals, the spectral mapping relationship shown in equation (3) is rewritten as: When using the BP neural network algorithm to fit the spectral mapping of the input signal and the output signal, a fixed-structure spectral mapping neural network model is constructed based on equation (10); Finally, an input-output sample set for the spectrum mapping neural network model is established, and the spectrum mapping neural network model is trained using the input-output sample set. The trained spectrum mapping neural network model is then used for spectrum mapping of the input and output signals.

2. The method for modeling radio frequency nonlinear behavior based on spectrum mapping according to claim 1, characterized in that, and They are represented as follows: The frequency range of the input signal is: The frequency range of the output signal is .

3. The method for modeling radio frequency nonlinear behavior based on spectrum mapping according to claim 1, characterized in that, The input to the spectrum mapping neural network model is the amplitude and phase of the simple harmonic signal of the input signal at each frequency component, as shown on the right side of equation (10); the output of the spectrum mapping neural network model is the amplitude and phase of the simple harmonic signal of the output signal at each frequency component, as shown on the left side of equation (10).

4. The method for modeling radio frequency nonlinear behavior based on spectrum mapping according to claim 1, characterized in that, Step three includes: First, if the input signal is in time-domain form, the amplitude and phase of the simple harmonic signal at each frequency in its spectrum are obtained using the Fast Fourier Transform. If the input signal is in frequency-domain form, the amplitude and phase of the simple harmonic signal at each frequency in its spectrum are obtained directly. Then, the amplitude and phase of the simple harmonic signal at each frequency of the output signal are calculated using the spectrum mapping neural network model; Finally, the amplitude and phase of the simple harmonic signal at each frequency are substituted into equation (9) to obtain the output signal.

Images