A broadband frequency doubling method and a radio frequency front-end system applying the same

Abstract

The application relates to a broadband frequency doubling method and a radio frequency front-end system applying the method, and belongs to the field of communication and radio frequency microwave technology. The method comprises the following steps: inputting an input base frequency signal into a main frequency doubling channel and an auxiliary frequency doubling channel respectively; performing nonlinear conversion on the input base frequency signal through the main frequency doubling channel to output a first mixed signal; performing nonlinear conversion on the input base frequency signal through the auxiliary frequency doubling channel to output a second mixed signal; performing vector modulation on the amplitude and phase of the second mixed signal to generate a cancellation signal; performing power synthesis on the first mixed signal and the cancellation signal, at least partially cancelling the second distortion component and the first distortion component after modulation, and outputting a frequency doubling signal; extracting a feedback signal from the frequency doubling signal, and adjusting the amplitude scaling factor and the phase offset of the vector modulation in real time according to the feedback signal. The application has the effects of reducing the intermodulation distortion component in the frequency doubling output signal and improving the spurious-free dynamic range.

Description

Technical Field

[0001] This application relates to the fields of communication and radio frequency microwave technology, and in particular to a broadband frequency doubling method and a radio frequency front-end system using the method. Background Technology

[0002] In radio frequency transceiver front-ends and test instruments, active frequency doublers are the core frequency conversion modules. They use the nonlinear characteristics of nonlinear devices to convert the input base frequency signal into a doubled output signal, which can reduce local oscillator feedthrough and improve system integration. Therefore, they are widely used in the fields of communication and test instruments.

[0003] Currently, active frequency doublers generally adopt a single-channel frequency multiplier architecture, which uses a nonlinear device (such as a Schottky diode or GaAs FET) as the frequency multiplier core, combined with an input matching network, harmonic filter, and output buffer stage to form the frequency conversion link. Some improved schemes improve the flatness of the frequency multiplication gain to a certain extent by introducing bias compensation or temperature compensation circuits in the frequency multiplication link.

[0004] However, the aforementioned single-channel frequency multiplier architecture suffers from the following problems: Nonlinear devices generate thermal noise and shot noise during operation, and their nonlinear characteristics also distort the input fundamental frequency signal, introducing intermodulation distortion. This inherent noise and intermodulation distortion cannot be eliminated by external filtering or shielding. While the frequency components of the intermodulation distortion are related to the input fundamental frequency signal, their root cause lies in the intrinsic nonlinearity of the device and cannot be avoided by optimizing the quality of the input signal. When the input signal is small, this inherent noise and intermodulation distortion lead to an increase in the noise floor of the output second harmonic signal, increased spurious signals, and a deterioration in the spurious-free dynamic range, severely limiting the sensitivity and spectral purity of the broadband high-precision frequency multiplier system. Summary of the Invention

[0005] To address the technical challenges of increased noise floor, spurious emissions, and degraded spurious-free dynamic range in second harmonic signals, which severely restrict the sensitivity and spectral purity of broadband high-precision frequency doubling systems, this application provides a broadband frequency doubling method and an RF front-end system using this method.

[0006] In a first aspect, this application provides a broadband frequency doubling method, comprising: The input base frequency signal is split into two paths and sent to the main frequency multiplication path and the auxiliary frequency multiplication path respectively; The nonlinear device in the main frequency doubling path performs a nonlinear conversion on one of the input fundamental frequency signals, and outputs a first mixed signal containing a second harmonic useful signal and a first distortion component. The pseudo-frequency doubling unit in the auxiliary frequency doubling path performs a nonlinear conversion on the other input fundamental frequency signal to output a second mixed signal containing a second distortion component. The pseudo-frequency doubling unit has the same operating state and circuit structure as the nonlinear device in the main frequency doubling path, so that the second distortion component has a preset correlation with the first distortion component. The second mixed signal is vector-modulated in terms of amplitude and phase to generate a cancellation signal; the cancellation signal contains a modulated second distortion component, which is used to cancel the first distortion component in reverse. The first mixed signal and the cancellation signal are combined by power synthesis, and the modulated second distortion component in the cancellation signal is at least partially canceled by the first distortion component, and a second harmonic signal is output. A feedback signal is extracted from the second harmonic signal, and the amplitude scaling factor and phase offset of the vector modulation are adjusted in real time according to the feedback signal to reduce the residual distortion component in the second harmonic signal, so that the cancellation error between the modulated second distortion component and the first distortion component tends to be minimized.

[0007] By adopting the above technical solution, an auxiliary path with the same operating state and circuit structure as the main path is set up, so that the distortion components generated by the auxiliary path have a preset correlation with the distortion components of the main path; by vector modulating the auxiliary path signal and then combining it with the main path signal power, the two distortion components cancel each other out at the combining node; by extracting the feedback signal from the output signal and adjusting the vector modulation parameters in real time, a closed-loop control is formed. This solution can reduce the intermodulation distortion components in the second harmonic output signal and improve the spurious-free dynamic range.

[0008] Optionally, adjusting the amplitude scaling factor and phase offset of the vector modulation in real time based on the feedback signal includes: Spectral analysis is performed on the feedback signal to obtain the power spectral density within a preset frequency band; the preset frequency band is a preset bandwidth range that includes the frequency value of the second harmonic signal. The power spectral density is compared with a preset first threshold and a second threshold to obtain a comparison result, wherein the first threshold is lower than the second threshold. Based on the comparison results, a tiered cancellation control strategy is implemented, the cancellation control strategy including: When the power spectral density is higher than the second threshold, the amplitude scaling factor and phase offset of the vector modulation are adjusted simultaneously. When the power spectral density is between the first threshold and the second threshold, the phase offset is fixed to an inverted state, and only the amplitude scaling factor is adjusted. When the power spectral density is lower than the first threshold, the amplitude scaling factor and phase offset of the vector modulation are not adjusted.

[0009] By employing the above technical solution, and through spectral analysis of the feedback signal, cancellation control is performed in stages based on the comparison results between the power spectral density within the preset frequency band and the first and second thresholds: amplitude and phase are adjusted simultaneously for strong distortion, only amplitude is adjusted for moderate distortion, and adjustment is skipped for weak distortion. This solution can allocate control resources according to the distortion intensity, reducing computational overhead while ensuring cancellation effectiveness.

[0010] Optionally, the real-time adjustment of the amplitude scaling factor and phase offset of the vector modulation includes calculating the updated values ​​of the amplitude scaling factor and the phase offset using a least mean square iterative algorithm; The method further includes: Obtain the signal-to-noise ratio of the second harmonic signal; The current signal condition is determined based on the signal-to-noise ratio. When the condition is determined to be a small signal condition, the minimum mean square iteration algorithm is executed with a preset first iteration step size. When the condition is determined to be a large signal condition, the minimum mean square iteration algorithm is executed with a preset second iteration step size, wherein the second iteration step size is greater than the first iteration step size.

[0011] By adopting the above technical solution, the signal condition is determined by obtaining the signal-to-noise ratio of the second harmonic signal. For small signals, a smaller iteration step size is used to ensure convergence stability, and for large signals, a larger iteration step size is used to ensure convergence speed.

[0012] Optionally, the method further includes: The sample signal is extracted by coupling from the output of the primary frequency harmonic path or from the second frequency harmonic signal; Envelope detection is performed on the sample signal to extract the envelope of the distortion component; The envelope of the distorted component is compared with the reference envelope to generate an envelope error signal; A first calibration control quantity is generated based on the envelope error signal; The first calibration control value is superimposed on the current amplitude scaling factor and / or current phase offset of the vector modulation to update the control parameters of the vector modulator; Specifically, the step of adjusting the amplitude scaling factor and phase offset of the vector modulation in real time according to the feedback signal at a first update rate is performed, and the step of generating a first calibration control quantity according to the envelope error signal at a second update rate is performed, wherein the second update rate is less than the first update rate.

[0013] By employing the above technical solution, a sample signal is extracted from the main path output or the second harmonic signal. The distortion component envelope is then extracted via envelope detection and compared with a reference envelope to generate an envelope error signal. This error signal is then generated at a second update rate slower than the main closed loop and superimposed on the current control parameters of the vector modulator. This solution can periodically correct cancellation offsets caused by temperature drift or device aging without interfering with the real-time response of the main closed loop.

[0014] Optionally, the method further includes: After each least mean square iteration algorithm reaches convergence, the signal characteristics of the current input fundamental frequency signal, the amplitude scaling factor and phase offset obtained from the convergence of this iteration are recorded and stored in the historical sample library as empirical samples. Based on multiple empirical samples accumulated in the historical sample library, a continuous mapping model is established from the signal feature quantity to the amplitude scaling factor and the phase offset; Before the least mean square iterative algorithm is started or when a dynamic change in the input base frequency signal is detected, the current signal feature quantity of the current input base frequency signal is obtained; The continuous mapping model is queried using the current signal features to generate predicted values ​​for the amplitude scaling factor and the phase offset; The predicted value is used as the initial value of the amplitude scaling factor and phase offset of the least mean square iterative algorithm for iteration.

[0015] By employing the above technical solution, after each LMS iteration converges, the current signal features and the cancelling parameters obtained upon convergence are recorded as empirical samples. A continuous mapping model from signal features to the optimal cancelling parameters is established. Before LMS starts or when the signal undergoes dynamic changes, this model is queried to generate predicted values ​​as initial values ​​for iteration. This solution utilizes historical convergence experience to accelerate iterative convergence after large dynamic signal transitions.

[0016] Optionally, the method further includes: Before splitting the input reference signal into two paths, the input reference signal is amplified by variable gain. Obtain the power of the input fundamental frequency signal and the cancellation error in the second harmonic signal; The gain value of the variable gain amplifier is determined jointly based on the power and the cancellation error, so that the nonlinear devices in the main frequency doubling path and the auxiliary frequency doubling path operate in the linear region; Specifically, when the power increases and the cancellation error increases, the gain value is reduced first to allow the nonlinear device to exit the saturation region, and then the cancellation is fine-tuned by adjusting the amplitude scaling factor and the phase offset; when the power reduction causes the signal-to-noise ratio of the second harmonic signal to be lower than a preset threshold, the gain value is increased and then the minimum mean square iteration algorithm is executed.

[0017] By adopting the above technical solution, a variable gain amplifier is set before the power divider, and the gain value is determined jointly based on the input signal power and cancellation error, so that the nonlinear device operates in the linear region. When the power increases and saturation occurs, the gain is preferentially reduced; when the power decreases and the signal-to-noise ratio is insufficient, the gain is increased before LMS is performed. This scheme can maintain the preset correlation between the distortion components of the main and auxiliary paths from being destroyed by saturation, and ensure that the LMS iteration has a sufficient signal-to-noise ratio under small signal conditions.

[0018] Optionally, the method further includes: Obtain the current signal characteristics of the input fundamental frequency signal; The continuous mapping model is queried using the current signal feature quantity to generate the predicted value of the gain value, as well as the predicted values ​​of the amplitude scaling factor and the phase offset; When the transient rate of change of the signal characteristic quantity of the input base frequency signal exceeds a preset threshold, the gain of the variable gain amplification is set with the predicted value of the gain value, and the predicted values ​​of the amplitude scaling factor and the phase offset are used as the initial values ​​of the minimum mean square iteration algorithm, and the gain setting and iteration initialization are executed synchronously. After the least mean square iterative algorithm converges, the continuous mapping model is updated based on the amplitude scaling factor and phase offset obtained from the convergence, and the mapping parameters related to gain prediction in the continuous mapping model are updated based on the convergence speed and convergence accuracy of this iteration.

[0019] By adopting the above technical solution, the predicted values ​​of gain and cancellation parameters are generated synchronously by querying the continuous mapping model with the current signal characteristics. When the transient rate of change of the signal exceeds the threshold, gain setting and iterative initialization are performed synchronously. After LMS convergence, the mapping model is updated based on the convergence result. This solution eliminates the serial waiting time of adjusting the gain first and then the cancellation parameters, and improves the prediction accuracy with the accumulation of running time.

[0020] Secondly, this application provides a radio frequency front-end system, including, A power divider is used to split the input baseband signal into two paths. The main frequency multiplier path, connected to one output terminal of the power divider, includes a nonlinear device for performing nonlinear conversion on one input base frequency signal, and outputting a first mixed signal containing a second harmonic useful signal and a first distortion component; An auxiliary frequency multiplication path is connected to another output terminal of the power divider and includes a pseudo frequency multiplication unit for performing nonlinear conversion on the other input base frequency signal to output a second mixed signal containing a second distortion component. The pseudo frequency multiplication unit has the same operating state and circuit structure as the nonlinear device so that the second distortion component has a preset correlation with the first distortion component. A vector modulator, connected to the output of the auxiliary frequency multiplication path, is used to modulate the amplitude and phase of the second mixed signal to generate a cancellation signal. The cancellation signal contains a modulated second distortion component, which is used to cancel the first distortion component in the opposite direction. A power combiner, connected to the output of the main frequency multiplication path and the output of the vector modulator, is used to combine the first mixed signal and the cancellation signal, wherein the modulated second distortion component in the cancellation signal at least partially cancels the first distortion component, and outputs a second frequency multiplication signal; A feedback coupling branch is connected to the output of the power combiner and is used to extract a feedback signal from the second harmonic signal. The controller, connected to the feedback coupling branch and the vector modulator respectively, is used to adjust the amplitude scaling factor and phase offset of the vector modulation in real time according to the feedback signal, so as to reduce the residual distortion component in the second harmonic signal and minimize the cancellation error between the modulated second distortion component and the first distortion component.

[0021] Thirdly, this application provides a radio frequency front-end device, including a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described in any of the first aspects.

[0022] Fourthly, this application provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described in any of the first aspects.

[0023] In summary, this application includes the following beneficial technical effects: In this application, by setting up an auxiliary path with the same operating state and circuit structure as the main path, the distortion components generated by the auxiliary path have a preset correlation with the distortion components of the main path. By vector modulating the auxiliary path signal and then combining it with the main path signal power, the two distortion components cancel each other out at the combining node. By extracting the feedback signal from the output signal and adjusting the vector modulation parameters in real time, a closed-loop control is formed. This scheme can reduce the intermodulation distortion components in the second harmonic output signal and improve the spurious-free dynamic range. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating the broadband frequency doubling method disclosed in the embodiments of this application.

[0026] Figure 2 This is a structural block diagram of the radio frequency front-end system disclosed in the embodiments of this application. Detailed Implementation

[0027] The following is in conjunction with the appendix Figure 1-2 This application will be described in further detail.

[0028] This application discloses a broadband frequency doubling method, the execution subject of which is a radio frequency front-end system. The following will combine... Figure 1 The detailed execution process of the broadband frequency doubling method by the radio frequency front-end system is described in detail.

[0029] S101 divides the input base frequency signal into two paths, which are then sent to the main frequency multiplication path and the auxiliary frequency multiplication path, respectively.

[0030] S102 performs nonlinear conversion on one input base frequency signal through a nonlinear device in the main frequency doubling path, and outputs a first mixed signal containing a second harmonic useful signal and a first distortion component.

[0031] S103, the pseudo-frequency doubling unit in the auxiliary frequency doubling path performs nonlinear conversion on the other input fundamental frequency signal and outputs a second mixed signal containing a second distortion component. The pseudo-frequency doubling unit has the same working state and circuit structure as the nonlinear device in the main frequency doubling path, so that the second distortion component has a preset correlation with the first distortion component.

[0032] S104, the second mixed signal is vector-modulated in terms of amplitude and phase to generate a cancellation signal; the cancellation signal contains a modulated second distortion component, which is used to cancel the first distortion component in reverse.

[0033] S105, the first mixed signal and the cancellation signal are combined by power synthesis, and the modulated second distortion component in the cancellation signal is at least partially canceled by the first distortion component, and a second harmonic signal is output.

[0034] S106 extracts the feedback signal from the second harmonic signal, and adjusts the amplitude scaling factor and phase offset of the vector modulation in real time according to the feedback signal to reduce the residual distortion component in the second harmonic signal, so that the cancellation error between the modulated second distortion component and the first distortion component tends to be minimized.

[0035] S106, specifically the step of "adjusting the amplitude scaling factor and phase offset of vector modulation in real time based on the feedback signal," includes the following sub-steps: S1061, Perform spectrum analysis on the feedback signal to obtain the power spectral density within a preset frequency band; the preset frequency band is a preset bandwidth range that includes the frequency value of the second harmonic signal. S1062, The power spectral density is compared with a preset first threshold and a second threshold respectively to obtain a comparison result, wherein the first threshold is lower than the second threshold; S1063, Based on the comparison results, a tiered cancellation control strategy is implemented. The cancellation control strategy includes: S1064, when the power spectral density is higher than the second threshold, simultaneously adjust the amplitude scaling factor and phase offset of the vector modulation; S1065, when the power spectral density is between the first threshold and the second threshold, the fixed phase offset is in the inverted state, and only the amplitude scaling factor is adjusted. S1066: When the power spectral density is below the first threshold, the amplitude scaling factor and phase offset of the vector modulation are not adjusted.

[0036] The real-time adjustment of the amplitude scaling factor and phase offset of the vector modulation includes calculating the updated values ​​of the amplitude scaling factor and phase offset using a least mean square iterative algorithm. S1067, obtain the signal-to-noise ratio of the second harmonic signal, and determine the current signal condition based on the signal-to-noise ratio; S1068, when the condition is determined to be a small signal condition, the minimum mean square iteration algorithm is executed using the preset first iteration step size; S1069, when the condition is determined to be a large signal condition, the minimum mean square iteration algorithm is executed with a preset second iteration step size, wherein the second iteration step size is greater than the first iteration step size.

[0037] In implementation, for S101, the RF front-end system first receives the input baseband signal from an external signal source through its input port. This input baseband signal is an RF carrier signal with a frequency of f0, and its frequency range can be, for example, 3 GHz to 40 GHz. The RF front-end system sends this input baseband signal to a power divider. The power divider, for example, uses a Wilkinson power divider structure, which divides the input baseband signal into two equal paths. The two signals have equal power and the same phase, and are output from the two output ports of the power divider, one path being sent to the main frequency multiplication path and the other path being sent to the auxiliary frequency multiplication path.

[0038] For S102, the main frequency doubling path contains a nonlinear device as the frequency doubling core. In one embodiment, this nonlinear device adopts a reverse-parallel Schottky diode pair structure, that is, two Schottky diodes are connected in parallel with their anodes facing each other and their cathodes facing outwards, forming a differential structure. The anode of this reverse-parallel Schottky diode pair structure is grounded, and the cathode is connected to an LC resonant matching network that resonates at the second harmonic frequency 2f0.

[0039] When the input fundamental frequency signal enters the main frequency multiplier path, it is applied across a pair of Schottky diodes connected in reverse parallel. Due to the nonlinear exponential relationship of the Schottky diode's current-voltage characteristics, the waveform of the input fundamental frequency signal undergoes nonlinear distortion as it passes through the diode pair. This distortion generates abundant harmonic components, including the fundamental frequency f0, the second harmonic 2f0, and the third harmonic 3f0. The LC resonant matching network resonates at the second harmonic frequency 2f0, presenting a low-impedance path for the second harmonic component, allowing it to pass efficiently and be output. Simultaneously, it presents a high impedance to the fundamental frequency and other harmonic components, suppressing their passage. After this nonlinear conversion and selection by the LC resonant matching network, the main frequency multiplier path outputs a first mixed signal. This first mixed signal contains the second harmonic useful signal (the desired output component at frequency 2f0) and a first distortion component. The first distortion component includes the intermodulation distortion component generated by the nonlinear device in the main frequency harmonic path during operation. This intermodulation distortion component is an additional frequency component generated due to the non-ideal characteristics of the nonlinear device, and its spectral components are located in the sideband range near the frequency of the second harmonic useful signal, for example, in the frequency band of 2f0±50MHz.

[0040] For S103, the auxiliary frequency doubling path contains a pseudo-frequency doubling unit. The structure of this pseudo-frequency doubling unit is exactly the same as the nonlinear device in the main frequency doubling path. Specifically, the pseudo-frequency doubling unit also adopts an anti-parallel Schottky diode pair structure, with its anode grounded and its cathode connected to the same LC resonant matching network as the main frequency doubling path.

[0041] Furthermore, the pseudo-frequency multiplier unit maintains consistency with the nonlinear devices in the main frequency multiplier path in the following four dimensions: devices from the same production batch, the same package specifications, the same DC bias voltage (e.g., Vbias = -0.3V), and the same quiescent operating point current (e.g., Iq = 1.2mA). This consistency ensures that the pseudo-frequency multiplier unit has the exact same operating state and circuit structure as the nonlinear devices in the main frequency multiplier path. When another input base frequency signal enters the auxiliary frequency multiplier path, the pseudo-frequency multiplier unit performs the same nonlinear conversion on the signal as the main frequency multiplier path, outputting a second mixed signal. This second mixed signal contains a second distortion component. Because the pseudo-frequency multiplier unit and the nonlinear devices in the main frequency multiplier path are completely identical in operating state and circuit structure, the distortion components generated by both under the same input signal are highly similar, i.e., the second distortion component has a preset correlation with the first distortion component. In actual engineering implementation, through the above four-fold consistency design, this correlation coefficient can reach above 0.97, ensuring that the second distortion component can reflect the characteristics of the first distortion component with high fidelity.

[0042] For S104, the output of the auxiliary frequency multiplication path is connected to a vector modulator. This vector modulator is used to modulate the amplitude and phase of the second mixed signal. The vector modulator internally includes an amplitude adjustment unit and a phase adjustment unit. The amplitude adjustment unit achieves amplitude scaling by changing the attenuation or amplification factor of the signal, and the phase adjustment unit achieves phase shifting by changing the length of the signal transmission path or through a phase-shifting network. The controller included in the RF front-end system provides two control voltages to the vector modulator via a digital-to-analog converter (DAC), controlling the amplitude scaling factor α and the phase shift θ, respectively. Initially, the controller sets the amplitude scaling factor α to a preset initial value (e.g., α = 1.0, i.e., no amplification or attenuation) and the phase shift θ to a preset initial value (e.g., θ = 0°). After modulation by the vector modulator, the second distortion component in the second mixed signal is adjusted to a modulated second distortion component. This modulated second distortion component is used in subsequent steps to cancel out the first distortion component, i.e., its amplitude is approximately equal to that of the first distortion component, and its phase differs from the first distortion component by 180°.

[0043] For S105, the output of the vector modulator and the output of the main frequency multiplier path are respectively connected to the two input ports of the power combiner. This power combiner, for example, employs an inverse multiplexed Wilkinson power divider structure, operates over a wide frequency range (e.g., 7 GHz to 40 GHz), and features low insertion loss (e.g., insertion loss ≤ 0.3 dB) and high port isolation (e.g., port isolation ≥ 28 dB).

[0044] The first mixed signal enters one input port of the power combiner from the main frequency multiplication path, while the modulated second mixed signal (i.e., the cancellation signal) enters the other input port of the power combiner from the vector modulator. Inside the power combiner, the two signals are superimposed. During this superposition process, the modulated second distortion component in the cancellation signal and the first distortion component in the first mixed signal are approximately equal in amplitude and 180° out of phase, resulting in destructive interference and at least partial cancellation. Simultaneously, the second harmonic useful signal in the first mixed signal, lacking a corresponding inverting signal in the auxiliary path, is retained and output through the power combiner. The power combiner outputs a second harmonic signal, which is the final output signal after distortion cancellation processing.

[0045] For S106, the output of the power combiner is also connected to a feedback coupling branch. This feedback coupling branch, for example, uses a directional coupler structure, which extracts a small portion of the signal energy from the main transmission line at the output of the power combiner as a feedback signal. This feedback signal contains information about the residual distortion components in the output second harmonic signal. The output of the feedback coupling branch is connected to the input port of the analog-to-digital converter (ADC) inside the controller via an RF signal transmission line. The feedback coupling branch sends the extracted feedback signal to the ADC of the controller for sampling via the RF transmission line. Before the ADC conversion, a pre-amplifier analog bandpass filter in the controller performs analog domain filtering on the feedback signal. The center frequency of this analog bandpass filter is set to the second harmonic signal frequency value 2f0, and the passband bandwidth is set to 2f0±50MHz, used to filter out out-of-band noise and spurious components that may cause aliasing before the ADC conversion. After analog domain filtering, the feedback signal is then sent to the ADC for sampling.

[0046] Subsequently, the controller performs spectral analysis on the filtered feedback signal (corresponding to S1061). Specifically, the controller integrates a high-speed analog-to-digital converter (ADC) with a sampling rate of, for example, 4.8 GS / s. The controller converts the feedback signal from the analog domain to the digital domain using the ADC, obtaining digitized feedback signal samples. The controller performs a Fast Fourier Transform (FFT) operation on the digitized feedback signal (it should be noted that the purpose of the first FFT operation on the digitized feedback signal here is to obtain the power spectral density within the preset frequency band, which is used for comparison with the first and second thresholds to achieve graded determination of distortion intensity). The number of FFT points is, for example, 1024 points, and the overlap rate is, for example, 75%. The controller uses a sliding window method, performing an FFT operation on the latest feedback signal sample every 100 μs to obtain the amplitude distribution of the feedback signal in the frequency domain. Based on this, the controller extracts the power spectral density within the preset frequency band. The preset frequency band is a preset bandwidth range centered on the second harmonic signal frequency value 2f0, that is, the frequency band range of 2f0±50MHz. This frequency band range covers the second harmonic useful signal and the distortion components that may exist in its adjacent sidebands.

[0047] For S1062, the controller internally presets a first threshold and a second threshold. The first and second thresholds are set based on the spectral mean within the preset frequency band. In one embodiment, the controller calculates the arithmetic mean of the power spectral density of all frequency points within the preset frequency band (2f0±50MHz), uses this mean as the spectral mean, adds 6dB to the spectral mean (i.e., spectral mean + 6dB) to set the first threshold, and adds 12dB to the spectral mean (i.e., spectral mean + 12dB) to set the second threshold, where the first threshold is lower than the second threshold.

[0048] The controller compares the calculated power spectral density with a first threshold and a second threshold, and executes a tiered cancellation control strategy based on the comparison results (corresponding to S1063 to S1066). Specifically: When the power spectral density is higher than the second threshold, it indicates the presence of a strong distortion component in the current frequency band. The controller simultaneously adjusts the amplitude scaling factor α and the phase offset θ of the vector modulator to quickly and comprehensively attenuate this strong distortion component. When the power spectral density is between the first and second thresholds, it indicates the presence of a moderate-intensity distortion component in the current frequency band. The controller fixes the current adjustment of the phase offset θ and does not perform additional phase adjustment, maintaining the established anti-phase cancellation relationship (keeping the phase of the modulated second distortion component 180° out of phase with the phase of the first distortion component). Only the amplitude scaling factor α is adjusted to simplify the adjustment and balance convergence speed and accuracy. When the power spectral density is lower than the first threshold, it indicates that the power level in the current frequency band is within the noise floor range and there is no significant distortion component. The controller does not adjust the amplitude scaling factor α or the phase offset θ to save the controller's computational resources.

[0049] During the aforementioned adjustment process, the controller employs the Least Mean Square (LMS) iterative algorithm to calculate the updated values ​​of the amplitude scaling factor α and the phase offset θ in real time. The controller acquires the second harmonic signal from the output of the power combiner, performs analog-to-digital conversion, and then the digital down-converter module within the controller down-converts the second harmonic signal to baseband to obtain a digital feedback signal. The controller performs a Fast Fourier Transform (FFT) on the digital feedback signal to extract the sum of squares of the amplitudes of all frequency components within the preset frequency band, excluding the second harmonic useful signal frequency point (i.e., 2f0). This sum of squares is used as the power estimate of the current residual distortion component, which is the current measure of the cancellation error and is used as the error signal e(n) of the LMS iterative algorithm. The second FFT operation performed on the same digital feedback signal here has a different purpose than the first FFT operation used for threshold comparison: the purpose of this FFT operation is to extract the sum of squares of the amplitudes of all frequency components within the preset frequency band, excluding the second harmonic useful signal frequency point, to quantify the current cancellation error value, which is used as the error signal for the LMS iterative algorithm. The two FFT operations are based on the same digital feedback signal samples, but serve different control loops.

[0050] The basic iterative formula of the LMS algorithm is w(n+1) = w(n) - μ·▽J(n), where w(n) is the current controlled parameter vector (containing α and θ), μ is the iteration step size, and ▽J(n) is the gradient direction of the error signal with respect to the parameters. Specifically, the controller calculates the partial derivatives of the error signal e(n) with respect to the amplitude scaling factor α and the partial derivatives of the error signal e(n) with respect to the phase offset θ, and uses these two partial derivatives to form the gradient direction; then, the controller updates the values ​​of α and θ respectively along the negative direction of this gradient direction according to the above iterative formula, so that the error signal e(n) changes in the direction of decreasing.

[0051] In a preferred implementation of this embodiment, since there is a coupling effect between the amplitude adjustment and phase adjustment of the vector modulator, the controller uses an alternating perturbation method for gradient estimation: in even-numbered iteration cycles, only a small perturbation is applied to the amplitude scaling factor α, and the change in the error signal is observed to estimate ∂e / ∂α and update α; in odd-numbered iteration cycles, only a small perturbation is applied to the phase offset θ, and the change in the error signal is observed to estimate ∂e / ∂θ and update θ, thereby reducing computational complexity while ensuring convergence.

[0052] For S1067 to S1069, during the execution of the LMS iterative algorithm, the controller simultaneously acquires the signal-to-noise ratio (SNR) of the second harmonic signal to determine the current signal condition. The SNR is calculated as follows: the controller acquires the second harmonic signal from the output of the power combiner, and extracts the power ratio of the useful second harmonic signal (signal power at frequency 2f0) to the in-band noise (average noise power within the preset frequency band excluding the 2f0 frequency point) through FFT operation. This ratio is used as the estimated value of the SNR.

[0053] The controller performs a signal-to-noise ratio (SNR) calculation and condition determination every closed-loop update cycle (≤10μs). Specifically, it continuously monitors the SNR of the current harmonic signal using a sliding window approach and adjusts the iteration step size in real-time based on the latest SNR in each cycle. When the SNR remains below a preset threshold for a given time window (e.g., for five consecutive cycles), this threshold is pre-calibrated by the system designer based on the device noise figure and system sensitivity requirements (e.g., set to 15dB), indicating a small-signal condition. Conversely, when the SNR remains above the preset threshold for a given time window, it indicates a large-signal condition. This hysteresis comparison method avoids frequent iteration step size switching caused by SNR fluctuations around the preset threshold.

[0054] After determining the current signal condition, the controller selects the corresponding iteration step size to execute the LMS iterative algorithm based on the condition. Specifically, when the condition is determined to be a small signal condition, the LMS iterative algorithm is executed using a preset first iteration step size. The first iteration step size is relatively small, for example, set to 0.003, so that the adjustment range of α and θ in each iteration is small, in order to ensure the stability of the iteration process under the condition of small signal and low signal-to-noise ratio, avoid iteration divergence due to excessive adjustment range in a single step, and ensure convergence accuracy.

[0055] When the signal condition is determined to be large, the LMS iterative algorithm is executed using a preset second iteration step size. The second iteration step size is relatively large, for example, set to 0.012, allowing for a larger adjustment range of α and θ in each iteration. This ensures a faster convergence speed during the iteration process under high signal-to-noise ratio conditions, enabling timely tracking of signal changes. The second iteration step size is larger than the first iteration step size. In this way, the controller can adaptively adjust the iteration step size according to the current signal condition, ensuring both convergence accuracy and response speed, allowing the system to maintain stable and efficient cancellation performance under wide dynamic range input signal conditions.

[0056] The controller uses a digital-to-analog converter (DAC) to convert the amplitude scaling factor α and phase offset θ updated in each iteration into an analog control voltage, which is then output to the control input of the vector modulator, thereby achieving real-time updates to the modulation state of the vector modulator. The entire process described above—from feedback signal extraction, spectrum analysis, threshold comparison, hierarchical cancellation control to LMS iterative calculation and updating of vector modulator parameters—constitutes a complete closed-loop feedback control loop. This closed loop continuously executes with a period ≤10μs, ensuring that residual distortion components in the second harmonic signal are suppressed in real-time and continuously, minimizing the cancellation error between the modulated second distortion component and the first distortion component. The controller monitors the change in the error signal e(n) Δe = |e(n) - e(n-1)| within each closed-loop update cycle. When Δe is less than a preset convergence threshold (e.g., 5% of the initial value of e(n)) for M consecutive periods (e.g., M=10), the controller determines that the LMS iteration has reached a stable convergence state. At this time, the current values ​​of α and θ remain unchanged, but the feedback signal is continuously monitored. When Δe exceeds the convergence threshold again (e.g., due to dynamic changes in the input signal), the controller restarts the LMS iteration update. Through the above convergence determination mechanism, the controller avoids invalid iterations while ensuring cancellation accuracy, thus reducing system power consumption.

[0057] Optionally, the broadband frequency doubling method also includes the following steps: S301, extracts sample signals by coupling from the output of the main frequency harmonic path or from the second frequency harmonic signal; S302, perform envelope detection on the sample signal and extract the envelope of the distortion component; S303, compare the envelope of the distortion component with the reference envelope to generate an envelope error signal; S304, generates a first calibration control quantity based on the envelope error signal; S305, superimpose the first calibration control value onto the current amplitude scaling factor and / or current phase offset of the vector modulation, and update the control parameters of the vector modulator; Specifically, the amplitude scaling factor and phase offset of the vector modulation are adjusted in real time according to the feedback signal at a first update rate, and a first calibration control quantity is generated according to the envelope error signal at a second update rate, wherein the second update rate is less than the first update rate.

[0058] In implementation, while executing the closed-loop feedback control described in S101 to S106 (i.e., real-time cancellation control executed at the first update rate), the RF front-end system also executes the following slow calibration step in parallel. This slow calibration step is executed cyclically at a second update rate; for example, the update period corresponding to the second update rate is 1ms to 10ms, meaning the second update rate is less than the first update rate. The numerical relationship between the first and second update rates is controlled separately by the controller's timer module; the two operate independently and in parallel.

[0059] For S301, the RF front-end system sets up a first coupling branch between the output of the main frequency multiplier path and the input port of the power combiner, or a second coupling branch at the output of the power combiner. When extracting from the output of the main frequency multiplier path, the first coupling branch couples and extracts a small portion of the signal energy as a sample signal from the main transmission line of the main frequency multiplier path output. When extracting from the second harmonic signal, the second coupling branch couples and extracts a small portion of the signal energy as a sample signal from the main transmission line of the power combiner output. One of the two coupling branches is selected, and the system designer chooses one implementation method based on the actual layout. This sample signal contains the amplitude information of the distortion component in the output signal of the main frequency multiplier path.

[0060] For the S302, the RF front-end system also includes an envelope detector, whose input is connected to the output of either the first coupling branch or the second coupling branch. After the sample signal enters the envelope detector, it demodulates the amplitude envelope of the sample signal. The envelope detector internally contains a rectifier diode and a low-pass filter: the rectifier diode performs half-wave or full-wave rectification on the input RF signal, converting the amplitude variation of the RF signal into a rectified pulsating DC signal; the low-pass filter smooths and filters this pulsating DC signal, removing the carrier frequency component and extracting the envelope signal that reflects the signal amplitude variation over time. The envelope detector converts this envelope signal into an analog voltage value, which is then sent to the controller via an analog-to-digital converter.

[0061] For the S303, the controller internally stores a reference envelope value. This reference envelope value is set as follows: During the initial system calibration phase, a known clean fundamental frequency signal is input to the RF front-end system. With both the primary and secondary frequency harmonic paths operating normally, the envelope voltage value output by the envelope detector is recorded. This voltage value corresponds to the ideal envelope reference under this input condition. Since the primary frequency harmonic path still generates inherent intermodulation distortion under this condition, the ideal envelope reference actually includes the second harmonic useful signal component and the inherent distortion component in the primary frequency harmonic path output signal. That is, the envelope voltage output by the envelope detector at this time includes the influence of the distortion component. To ensure that the reference envelope can serve as a reference without additional distortion, the controller records this envelope voltage during the calibration phase and uses it as the reference envelope for subsequent comparisons. During normal system operation, the controller compares the real-time distortion component envelope voltage output by the current envelope detector with this stored reference envelope voltage, calculates the difference between the two, and generates an envelope error signal. This envelope error signal reflects the offset direction and amount of the current distortion component relative to the calibration reference.

[0062] For S304, the controller performs a second LMS iterative operation on the envelope error signal to generate a first calibration control quantity. Unlike the first LMS iterative algorithm (i.e., the LMS algorithm used for real-time adjustment of the amplitude scaling factor and phase offset) that performs real-time cancellation adjustment in S101 to S106, the second LMS iterative operation in this step uses a smaller iteration step size and fewer iterations. Specifically, the iteration step size of the second LMS iterative operation is smaller than that of the first LMS iterative algorithm (for example, the second LMS iteration step size can be set to 0.0005 to 0.001) to ensure the convergence stability of slow calibration; the second LMS iterative operation only performs a preset number of iterations (for example, 10 to 50 times) in each slow calibration cycle before outputting the current first calibration control quantity. This first calibration control quantity includes an amplitude calibration bias and / or a phase calibration bias, corresponding to the correction amount of the amplitude scaling factor and the correction amount of the phase offset, respectively. In the second LMS iteration operation, the error signal is the envelope error signal, and the controlled parameters are the amplitude calibration bias and / or the phase calibration bias. The second LMS iteration algorithm gradually adjusts these two biases along the direction in which the envelope error signal decreases the fastest, so that the envelope error signal tends to be minimized.

[0063] For S305, at the end of each slow calibration cycle, the controller superimposes the first calibration control value output from the second LMS iteration calculation onto the amplitude scaling factor and / or the current phase offset currently used by the vector modulator. Specifically, when the first calibration control value includes an amplitude calibration bias, the controller reads the value of the amplitude scaling factor currently used by the vector modulator, adds this value to the amplitude calibration bias, and obtains the updated amplitude scaling factor; when the first calibration control value includes a phase calibration bias, the controller reads the value of the phase offset currently used by the vector modulator, adds this value to the phase calibration bias, and obtains the updated phase offset. The controller outputs the updated amplitude scaling factor and / or phase offset to the control input of the vector modulator through a digital-to-analog converter, overwriting the corresponding control parameters previously stored in the vector modulator's internal registers with the updated amplitude scaling factor and / or phase offset, thus completing one slow calibration cycle. After the overwrite is completed, the vector modulator operates according to the updated control parameters, and the control parameters written in the previous slow calibration cycle are completely replaced and no longer retained or superimposed. Subsequently, the closed-loop feedback control continues to perform real-time rapid adaptive adjustment based on the updated control parameters and at the first update rate.

[0064] It should be noted that the closed-loop feedback control described in S101 to S106 operates at a high speed with a first update rate (period ≤ 10 μs). Its function is to quickly track and cancel the real-time distortion components caused by dynamic changes in the input signal at the current moment. However, its adjustment reference point (i.e., the current amplitude scaling factor and phase offset of the vector modulator) may gradually deviate from the optimal value due to temperature drift or device aging. The slow calibration step applies a reference correction to the current control parameters of the vector modulator periodically with a second update rate (period 1 ms to 10 ms). That is, it continuously tracks the cancellation offset caused by temperature drift or device aging through envelope detection and slow LMS iteration, and adds this correction to the control parameters of the fast closed loop. The fast closed loop is responsible for real-time fine-tuning near the corrected reference point, while the slow calibration is responsible for periodically updating the reference point itself. The two are decoupled and run in parallel on the time scale without interfering with each other.

[0065] Optionally, the broadband frequency doubling method also includes the following steps: S401: After each least mean square iteration algorithm reaches convergence, record the signal characteristics of the current input fundamental frequency signal and the amplitude scaling factor and phase offset obtained from the convergence of this iteration, and store them as empirical samples in the historical sample library.

[0066] S402, based on multiple empirical samples accumulated in the historical sample library, establishes a continuous mapping model from signal features to amplitude scaling factor and phase offset.

[0067] S403: Before the least mean square iterative algorithm is started or when a dynamic change in the input base frequency signal is detected, the current signal characteristic quantity of the current input base frequency signal is obtained.

[0068] S404 queries the continuous mapping model with the current signal features to generate predicted values ​​for the amplitude scaling factor and phase offset.

[0069] S405 uses the predicted value as the initial value for the amplitude scaling factor and phase offset of the least mean square iterative algorithm for iteration.

[0070] S501 performs variable gain amplification on the input reference signal before splitting it into two paths.

[0071] S502, obtains the power of the input base frequency signal and the cancellation error in the second harmonic signal.

[0072] S503 determines the gain value of the variable gain amplifier based on the power and the cancellation error, so that the nonlinear devices in the main frequency doubling path and the auxiliary frequency doubling path operate in the linear region.

[0073] Specifically, when the power increases and the cancellation error increases, the gain value is reduced first to allow the nonlinear device to exit the saturation region, and then the cancellation is fine-tuned by adjusting the amplitude scaling factor and phase offset. When the power decreases and the signal-to-noise ratio of the second harmonic signal is lower than the preset threshold, the gain value is increased and then the minimum mean square iteration algorithm is executed.

[0074] S601, obtain the current signal characteristic quantity of the input base frequency signal.

[0075] S602 queries the continuous mapping model with the current signal characteristics to generate predicted values ​​for the gain, amplitude scaling factor, and phase offset.

[0076] S603: When the transient change rate of the signal characteristic quantity of the input base frequency signal exceeds the preset threshold, the gain of the variable gain amplification is set with the predicted value of the gain value, and the predicted values ​​of the amplitude scaling factor and the phase offset are used as the initial values ​​of the minimum mean square iteration algorithm to simultaneously perform gain setting and iteration initialization.

[0077] S604 After the least mean square iterative algorithm converges, the continuous mapping model is updated based on the amplitude scaling factor and phase offset obtained from the convergence, and the mapping parameters related to gain prediction in the continuous mapping model are updated based on the convergence speed and convergence accuracy of this iteration.

[0078] In implementation, based on the closed-loop feedback control described in S101 to S106 and the adaptive step size switching described in S1067 to S1069, the RF front-end system further performs the following experience accumulation and parameter prediction steps.

[0079] For S401, the controller performs a convergence state determination after each LMS iteration converges. The convergence state determination method is as follows: the controller monitors the change in the error signal e(n) Δe = |e(n) - e(n-1)| within each closed-loop update cycle. When Δe is less than a preset convergence threshold (e.g., 5% of the initial value of e(n)) for M consecutive cycles (M is, for example, 10), the controller determines that the current LMS iteration has reached a stable convergence state. At this time, the controller acquires the following data: the signal characteristics of the current input base frequency signal, and the amplitude scaling factor α and phase offset θ obtained from the final convergence of this iteration. The signal characteristics include the instantaneous power P_in of the current input base frequency signal and its power-time difference dP / dt (i.e., the difference between the power at the current moment and the power at the previous moment divided by the time interval). The controller combines the above data into an empirical sample according to a preset data structure format, which is represented as {P_in, dP / dt, α, θ}. The controller stores this empirical sample in a historical sample library in non-volatile memory. The historical sample database is stored in the form of a queue, with a queue length set to, for example, 1000 entries. When the number of samples exceeds the queue length, the controller removes the oldest sample according to the first-in-first-out principle to control storage space and ensure the timeliness of the samples.

[0080] For the S402, the controller, based on multiple empirical samples accumulated in the historical sample library, uses a locally weighted regression method to establish a continuous mapping model between signal features (input power P_in and power-time difference dP / dt) and optimal cancellation parameters (amplitude scaling factor α and phase offset θ). The process of establishing this mapping model is as follows: The controller treats the signal features of each empirical sample in the historical sample library as discrete observation points in a two-dimensional input space (the horizontal axis represents instantaneous power P_in, and the vertical axis represents the power-time difference dP / dt), and the corresponding α and θ are considered as the observed values ​​at that observation point. When it is necessary to predict the optimal parameters under a new input condition, the controller performs the following local weighted regression operation: First, the controller calculates the similarity between the current signal feature (P_current, dP / dt_current) and the signal feature of each empirical sample in the historical sample library. The similarity is calculated as follows: the absolute value of the instantaneous power difference |P_current-P_in| and the absolute value of the power-time difference |dP / dt_current-dP / dt| are calculated respectively, the two differences are normalized and then summed to obtain the distance metric. Regarding the specific implementation of the normalization process: The controller employs a dynamic normalization method, using the difference between the maximum and minimum instantaneous power values ​​of all experience samples in the historical sample library as the normalization benchmark for the power dimension, and the difference between the maximum and minimum power time differences of all experience samples as the normalization benchmark for the difference dimension. Each difference is divided by its corresponding normalization benchmark to obtain a dimensionless normalized difference, thereby eliminating the influence of dimensional differences on distance measurement calculations. The smaller the distance, the higher the similarity.

[0081] Secondly, the controller selects the K empirical samples with the smallest distance metric values ​​(K is, for example, 20). Next, the controller uses the reciprocal of the distance metric value of each selected empirical sample as a weight, assigning higher weights to samples that are closer in distance, and performs a weighted average of α and θ among the K samples. The result of the weighted average is the predicted value of the amplitude scaling factor α_pred and the predicted value of the phase offset θ_pred corresponding to the current signal feature. After obtaining the weighted average result, the controller further calculates the standard deviation of the amplitude scaling factor α among the K empirical samples. If the standard deviation is less than a preset dispersion threshold (for example, the standard deviation of α is less than 0.05), the weighted average result is output as the predicted value; if the standard deviation is greater than the preset dispersion threshold, it indicates that the consistency of empirical samples similar to the current operating condition in the historical sample library is poor, the controller reduces the confidence of the predicted value, and uses the weighted sum of the predicted value and a preset default initial value as the actual initial value used for iteration, where the weight of the predicted value is positively correlated with the confidence level. The aforementioned local weighted regression calculation is performed by the controller within each prediction period. The more samples in the historical sample library and the denser their distribution, the higher the prediction accuracy. In embodiments without a Gaussian kernel function, the controller directly uses the reciprocal of the distance metric as the weight. In another embodiment, the controller uses a Gaussian kernel function as the weight allocation function, with its kernel width parameter σ controlling the decay rate of the weight as distance increases. The weight corresponding to distance d is K(d) = exp(-d² / (2σ²)), and the smaller the distance, the higher the weight. The weighted average in this step is implemented using any of the above weight allocation methods.

[0082] For the S403, the controller triggers the parameter prediction process in two scenarios: First, during system startup or recovery from standby, it is triggered before the initial startup of the LMS iterative algorithm. Second, during normal system operation, the controller continuously monitors the signal characteristics of the input baseband signal. When the rate of change of instantaneous power exceeds a preset threshold (e.g., a power change rate greater than 3 dB / μs), it determines that the input signal has undergone dynamic change and triggers the parameter prediction process. After triggering, the controller obtains the instantaneous power P_current of the input baseband signal at the current moment through the power detection module and calculates the power-time difference dP / dt_current at the current moment as the current signal characteristic.

[0083] For S404, the controller inputs the current signal characteristics (P_current, dP / dt_current) into the continuous mapping model established by S402, performs local weighted regression operation, and generates the amplitude scaling factor prediction value α_pred and the phase offset prediction value θ_pred.

[0084] For the S405, the controller uses α_pred and θ_pred as the initial control parameters for the vector modulator, outputting them to the vector modulator via a digital-to-analog converter, so that the vector modulator starts operating with these predicted values ​​as the initial state. Subsequently, the controller initiates the LMS iterative algorithm, performing iterative fine-tuning from this initial value. Since the predicted value is close to the optimal parameters under the current operating conditions, the number of convergence steps required for the LMS iteration is significantly reduced.

[0085] For the S501, the RF front-end system incorporates a variable gain amplifier connected in series between the input port and the power divider. The input of this variable gain amplifier is connected to the signal input port of the RF front-end system, and its output is connected to the input port of the power divider. The gain of the variable gain amplifier is controlled by a control voltage output from the controller. The controller provides this gain control voltage to the variable gain amplifier via a digital-to-analog converter, and this control voltage corresponds to the gain value of the variable gain amplifier.

[0086] For S502, the controller obtains the power P_in of the input fundamental frequency signal through a power detection module located before the input of the variable gain amplifier. Simultaneously, the controller calculates the cancellation error e(n) in the current second harmonic signal using the feedback signal extracted in S106. This cancellation error e(n) is the power estimate of the current residual distortion component, and its specific calculation method has been described in detail in S106.

[0087] For the S503, when the controller detects an increase in input power and a synchronous increase in the cancellation error e(n) (e.g., P_in increases beyond a preset threshold and the growth rate of e(n) exceeds a preset ratio threshold within N consecutive sampling periods), the controller determines that the nonlinear device has entered the saturation region. At this time, the controller prioritizes a gain reduction operation: gradually reducing the gain value of the variable gain amplifier by a preset first gain adjustment step size (e.g., 0.5dB), thereby reducing the signal amplitude input to the main and auxiliary frequency multiplication paths, thus causing the nonlinear device to exit the saturation region. During the gain reduction operation, the controller temporarily suspends the LMS iterative algorithm's updates to the amplitude scaling factor and phase offset. After the gain adjustment stabilizes (e.g., after the gain adjustment is completed and a preset stabilization time is waited for, such as 10μs), the controller resumes the execution of the LMS iterative algorithm, performing fine-tuning of the cancellation parameters provided the nonlinear device has returned to the linear region.

[0088] When the controller detects a decrease in input power and the signal-to-noise ratio (SNR) of the second harmonic signal falls below a preset threshold (e.g., SNR below 15dB), it indicates that the current input signal is too weak, the gradient estimation accuracy of the LMS algorithm may be insufficient, and the iteration process is susceptible to noise interference and divergence. In this case, the controller prioritizes a gain boost operation: gradually increasing the gain value of the variable gain amplifier by a preset second gain adjustment step size (e.g., 0.5dB), thereby increasing the signal amplitude input to the main harmonic path and the auxiliary harmonic path, and thus improving the SNR of the second harmonic signal. Once the SNR recovers to above the preset threshold, the controller then starts or resumes the execution of the LMS iterative algorithm to ensure the stability of the iteration process.

[0089] During gain adjustment, the controller simultaneously monitors the operating status of the nonlinear device (e.g., through static operating point current monitoring) to ensure that the adjusted gain value does not cause the device to re-enter the saturation region. The gain adjustment adopts a "coarse adjustment followed by fine adjustment" strategy: in the saturation state, the gain is rapidly reduced with relatively large steps, and after exiting the saturation region, the gain and cancellation parameters are finely adjusted with smaller steps to achieve joint convergence.

[0090] For S601, based on the variable gain amplification control described in S501 to S503, the controller further acquires the signal characteristics of the current input base frequency signal. These signal characteristics include the instantaneous power P_current and the power-time difference dP / dt_current. The acquisition method is the same as in S403.

[0091] For S602, the controller queries the continuous mapping model established in S402 using the current signal characteristic (P_current, dP / dt_current). The continuous mapping model established in S402 originally only outputs predicted values ​​for the amplitude scaling factor and phase offset. In this step, the controller expands the output of the continuous mapping model: the controller extracts the K closest empirical samples to the current signal characteristic from the historical sample library, obtaining the gain value information of these empirical samples at the time of storage. This gain value information is obtained as follows: when the historical sample library is established in S402, each empirical sample, in addition to recording the signal characteristic and optimal cancellation parameters, also synchronously records the gain value used by the variable gain amplifier when the empirical sample was acquired. The controller uses the same local weighted regression method as S402, with the reciprocal of the distance metric as the weight, to perform a weighted average of the gain values ​​in the K samples, generating a predicted gain value G_pred. Simultaneously, the controller generates the predicted amplitude scaling factor α_pred and the predicted phase offset θ_pred according to the process of S402.

[0092] For S603, the controller continuously monitors the transient rate of change of signal characteristics of the input baseband signal, such as continuously calculating the power difference between the current moment and the previous moment and dividing it by the time interval to obtain the transient rate of change. When this transient rate of change exceeds a preset threshold (e.g., power change rate greater than 3dB / μs), the controller determines that the input signal has experienced a large jump. At this time, the controller performs the following operations: First, the controller outputs the corresponding control voltage to the variable gain amplifier using the gain prediction value G_pred generated by S602 via a digital-to-analog converter, setting the gain of the variable gain amplifier to G_pred. Simultaneously, the controller outputs α_pred and θ_pred generated by S602 to the vector modulator via another digital-to-analog converter, setting the amplitude scaling factor and phase offset of the vector modulator. The gain setting and vector modulator parameter setting are issued in parallel by the controller within the same control cycle and completed synchronously, without waiting for the gain adjustment to complete before executing the parameter settings.

[0093] After the gain and initial parameters are set synchronously, the controller immediately starts the LMS iterative algorithm to perform iterative fine-tuning starting from α_pred and θ_pred. The benefit of synchronous execution is that it eliminates the serial waiting time of "detecting error changes first, then adjusting the gain, and then adjusting the cancellation parameters" in existing technologies, thus significantly improving the system response speed.

[0094] For S604, after each LMS iteration converges (the convergence determination method is the same as S401), the controller performs the following update operation: On one hand, the controller combines the current signal characteristics (P_current, dP / dt_current), the converged amplitude scaling factor α and phase offset θ, and the current gain value G_current of the variable gain amplifier into a new empirical sample {P_current, dP / dt_current, α, θ, G_current}, and stores it in the historical sample library. The difference between this operation and S401 is the addition of a gain value field, enabling the continuous mapping model to simultaneously support gain prediction and cancellation parameter prediction.

[0095] On the other hand, the controller updates the mapping parameters related to gain prediction in the continuous mapping model based on the convergence speed and accuracy of this iteration. The convergence speed is measured by the number of iterations N_iter from the start to the convergence of this iteration, and the convergence accuracy is measured by the error signal e_final after final convergence. The controller compares the gain prediction value G_pred recorded in the current empirical sample with the actual gain value G_current, and calculates the gain prediction error ΔG = G_current - G_pred. Based on this gain prediction error, the controller updates the kernel width parameter σ (when using a Gaussian kernel function) related to gain prediction in subsequent local weighted regression. When ΔG is consistently positive (i.e., the actual gain is always higher than the predicted value), it indicates that the prediction model has a systematic bias in the operating region corresponding to the current signal feature, and the controller will increase the kernel width parameter σ used in the local weighted regression in this region. Specifically, the controller marks the feature space region within a preset radius as the operating region with the current signal feature as the center, and increases the kernel width parameter σ of the empirical samples in this region during subsequent prediction. The increase in the kernel width parameter σ is positively correlated with the absolute value of ΔG; that is, the greater the prediction error, the greater the increase in the kernel width parameter, thus smoothing the model in that region and avoiding overfitting to noisy samples. When the prediction error is small, the kernel width parameter is maintained or decreased to improve local fitting accuracy. The adjustment range of the kernel width parameter is between a preset upper and lower limit.

[0096] It should be noted that the experience accumulation and parameter prediction steps described in S401 to S405, the variable gain amplification control steps described in S501 to S503, and the joint prediction and synchronous execution steps described in S601 to S604 are progressively related: S401 to S405 establish a basic mapping model from signal characteristics to cancellation parameters; S501 to S503 introduces variable gain amplification control based on this mapping model, solving the problems of large signal saturation and insufficient signal-to-noise ratio for small signals; S601 to S604 further incorporates gain control into the prediction range of the mapping model, realizing the joint prediction and synchronous execution of gain and cancellation parameters, eliminating the serial delay between gain adjustment and parameter adjustment. These three steps together constitute the complete technical solution of this embodiment.

[0097] Reference Figure 2 This application also discloses a radio frequency front-end system, including, A power divider is used to split the input baseband signal into two paths. The main frequency multiplier path, connected to one output terminal of the power divider, includes a nonlinear device for performing nonlinear conversion on one input base frequency signal, and outputting a first mixed signal containing a second harmonic useful signal and a first distortion component; An auxiliary frequency multiplication path is connected to another output terminal of the power divider and includes a pseudo frequency multiplication unit for performing nonlinear conversion on the other input base frequency signal to output a second mixed signal containing a second distortion component. The pseudo frequency multiplication unit has the same operating state and circuit structure as the nonlinear device so that the second distortion component has a preset correlation with the first distortion component. A vector modulator, connected to the output of the auxiliary frequency multiplication path, is used to modulate the amplitude and phase of the second mixed signal to generate a cancellation signal. The cancellation signal contains a modulated second distortion component, which is used to cancel the first distortion component in the opposite direction. A power combiner, connected to the output of the main frequency multiplication path and the output of the vector modulator, is used to combine the first mixed signal and the cancellation signal, wherein the modulated second distortion component in the cancellation signal at least partially cancels the first distortion component, and outputs a second frequency multiplication signal; A feedback coupling branch is connected to the output of the power combiner and is used to extract a feedback signal from the second harmonic signal. The controller, connected to the feedback coupling branch and the vector modulator respectively, is used to adjust the amplitude scaling factor and phase offset of the vector modulation in real time according to the feedback signal, so as to reduce the residual distortion component in the second harmonic signal and minimize the cancellation error between the modulated second distortion component and the first distortion component.

[0098] This application also discloses a radio frequency front-end device, which includes a memory and a processor. The memory stores a computer program that can be loaded by the processor and executed as described above for the wideband frequency doubling method.

[0099] This application also discloses a computer-readable storage medium that stores a computer program that can be loaded by a processor and executed as described above in the wideband frequency doubling method. The computer-readable storage medium includes, for example, various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0100] It should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0101] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit the scope of protection of the application. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

Claims

1. A broadband frequency doubling method, characterized in that, include: The input base frequency signal is split into two paths and sent to the main frequency multiplication path and the auxiliary frequency multiplication path respectively; The nonlinear device in the main frequency doubling path performs a nonlinear conversion on one of the input fundamental frequency signals to output a first mixed signal containing a second harmonic useful signal and a first distortion component. The pseudo-frequency doubling unit in the auxiliary frequency doubling path performs a nonlinear conversion on the other input fundamental frequency signal to output a second mixed signal containing a second distortion component. The pseudo-frequency doubling unit has the same operating state and circuit structure as the nonlinear device in the main frequency doubling path, so that the second distortion component has a preset correlation with the first distortion component. The second mixed signal is vector-modulated in terms of amplitude and phase to generate a cancellation signal; the cancellation signal contains a modulated second distortion component, which is used to cancel the first distortion component in reverse. The first mixed signal and the cancellation signal are combined by power synthesis, and the modulated second distortion component in the cancellation signal is at least partially canceled by the first distortion component, and a second harmonic signal is output. A feedback signal is extracted from the second harmonic signal, and the amplitude scaling factor and phase offset of the vector modulation are adjusted in real time according to the feedback signal to reduce the residual distortion component in the second harmonic signal, so that the cancellation error between the modulated second distortion component and the first distortion component tends to be minimized.

2. The broadband frequency doubling method according to claim 1, characterized in that, The step of adjusting the amplitude scaling factor and phase offset of the vector modulation in real time based on the feedback signal includes: Spectral analysis is performed on the feedback signal to obtain the power spectral density within a preset frequency band; the preset frequency band is a preset bandwidth range that includes the frequency value of the second harmonic signal. The power spectral density is compared with a preset first threshold and a second threshold to obtain a comparison result, wherein the first threshold is lower than the second threshold. Based on the comparison results, a tiered cancellation control strategy is implemented, the cancellation control strategy including: When the power spectral density is higher than the second threshold, the amplitude scaling factor and phase offset of the vector modulation are adjusted simultaneously. When the power spectral density is between the first threshold and the second threshold, the phase offset is fixed to an inverted state, and only the amplitude scaling factor is adjusted. When the power spectral density is lower than the first threshold, the amplitude scaling factor and phase offset of the vector modulation are not adjusted.

3. The broadband frequency doubling method according to claim 1 or 2, characterized in that, The real-time adjustment of the amplitude scaling factor and phase offset of the vector modulation includes calculating the updated values ​​of the amplitude scaling factor and the phase offset using a least mean square iterative algorithm; The method further includes: Obtain the signal-to-noise ratio of the second harmonic signal; The current signal condition is determined based on the signal-to-noise ratio. When the condition is determined to be a small signal condition, the minimum mean square iteration algorithm is executed with a preset first iteration step size. When the condition is determined to be a large signal condition, the minimum mean square iteration algorithm is executed with a preset second iteration step size, wherein the second iteration step size is greater than the first iteration step size.

4. The broadband frequency doubling method according to any one of claims 1-3, characterized in that, The method further includes: The sample signal is extracted by coupling from the output of the primary frequency harmonic path or from the second frequency harmonic signal; Envelope detection is performed on the sample signal to extract the envelope of the distortion component; The envelope of the distorted component is compared with the reference envelope to generate an envelope error signal; A first calibration control quantity is generated based on the envelope error signal; The first calibration control value is superimposed on the current amplitude scaling factor and / or current phase offset of the vector modulation to update the control parameters of the vector modulator; Specifically, the step of adjusting the amplitude scaling factor and phase offset of the vector modulation in real time according to the feedback signal at a first update rate is performed, and the step of generating a first calibration control quantity according to the envelope error signal at a second update rate is performed, wherein the second update rate is less than the first update rate.

5. The broadband frequency doubling method according to claim 3, characterized in that, The method further includes: After each least mean square iteration algorithm reaches convergence, the signal characteristics of the current input fundamental frequency signal, the amplitude scaling factor and phase offset obtained from the convergence of this iteration are recorded and stored in the historical sample library as empirical samples. Based on multiple empirical samples accumulated in the historical sample library, a continuous mapping model is established from the signal feature quantity to the amplitude scaling factor and the phase offset; Before the least mean square iterative algorithm is started or when a dynamic change in the input base frequency signal is detected, the current signal feature quantity of the current input base frequency signal is obtained; The continuous mapping model is queried using the current signal features to generate predicted values ​​for the amplitude scaling factor and the phase offset; The predicted value is used as the initial value of the amplitude scaling factor and phase offset of the least mean square iterative algorithm for iteration.

6. The broadband frequency doubling method according to claim 5, characterized in that, The method further includes: Before splitting the input reference signal into two paths, the input reference signal is amplified by variable gain. Obtain the power of the input fundamental frequency signal and the cancellation error in the second harmonic signal; The gain value of the variable gain amplifier is determined jointly based on the power and the cancellation error, so that the nonlinear devices in the main frequency doubling path and the auxiliary frequency doubling path operate in the linear region; Specifically, when the power increases and the cancellation error increases, the gain value is reduced first to allow the nonlinear device to exit the saturation region, and then the cancellation is fine-tuned by adjusting the amplitude scaling factor and the phase offset; when the power reduction causes the signal-to-noise ratio of the second harmonic signal to be lower than a preset threshold, the gain value is increased and then the minimum mean square iteration algorithm is executed.

7. The broadband frequency doubling method according to claim 6, characterized in that, The method further includes: Obtain the current signal characteristics of the input fundamental frequency signal; The continuous mapping model is queried using the current signal feature quantity to generate the predicted value of the gain value, as well as the predicted values ​​of the amplitude scaling factor and the phase offset; When the transient rate of change of the signal characteristic quantity of the input base frequency signal exceeds a preset threshold, the gain of the variable gain amplification is set with the predicted value of the gain value, and the predicted values ​​of the amplitude scaling factor and the phase offset are used as the initial values ​​of the minimum mean square iteration algorithm, and the gain setting and iteration initialization are executed synchronously. After the least mean square iterative algorithm converges, the continuous mapping model is updated based on the amplitude scaling factor and phase offset obtained from the convergence, and the mapping parameters related to gain prediction in the continuous mapping model are updated based on the convergence speed and convergence accuracy of this iteration.

8. A radio frequency front-end system, characterized in that, include, A power divider is used to split the input baseband signal into two paths. The main frequency multiplier path, connected to one output terminal of the power divider, includes a nonlinear device for performing nonlinear conversion on one input base frequency signal, and outputting a first mixed signal containing a second harmonic useful signal and a first distortion component; An auxiliary frequency multiplication path is connected to another output terminal of the power divider and includes a pseudo frequency multiplication unit for performing nonlinear conversion on the other input base frequency signal to output a second mixed signal containing a second distortion component. The pseudo frequency multiplication unit has the same operating state and circuit structure as the nonlinear device so that the second distortion component has a preset correlation with the first distortion component. A vector modulator, connected to the output of the auxiliary frequency multiplication path, is used to modulate the amplitude and phase of the second mixed signal to generate a cancellation signal. The cancellation signal contains a modulated second distortion component, which is used to cancel the first distortion component in the opposite direction. A power combiner, connected to the output of the main frequency multiplication path and the output of the vector modulator, is used to combine the first mixed signal and the cancellation signal, wherein the modulated second distortion component in the cancellation signal at least partially cancels the first distortion component, and outputs a second frequency multiplication signal; A feedback coupling branch is connected to the output of the power combiner and is used to extract a feedback signal from the second harmonic signal. The controller, connected to the feedback coupling branch and the vector modulator respectively, is used to adjust the amplitude scaling factor and phase offset of the vector modulation in real time according to the feedback signal, so as to reduce the residual distortion component in the second harmonic signal and minimize the cancellation error between the modulated second distortion component and the first distortion component.

9. A radio frequency front-end device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 7.

Images