Digital pre-distortion method and system based on multi-strategy fusion model optimization

Abstract

The invention provides a digital pre-distortion method and system based on multi-strategy fusion model optimization, relates to the technical field of radio communication, and aims to solve the problems of local optimum, difficulty in convergence, high model complexity, high computing resource consumption and the like in the prior art. The method comprises the following steps: constructing a power amplifier behavior model and a pre-distortion model, performing global search optimization on a pre-distortion model coefficient by adopting an improved particle swarm optimization algorithm, constructing a self-fitness function, preliminarily obtaining an optimization coefficient of a digital pre-distorter, selecting an optimization coefficient of a preset proportion, performing local refined optimization by adopting a hill climbing algorithm, and finally obtaining a pre-distortion model of the digital pre-distorter. Obtaining a target matching coefficient of the pre-distortion model, taking the minimum error between the actual output of the power amplifier after nonlinear compensation and the linear amplification output of an original input signal as an optimization target, and calculating and updating a pre-distortion parameter by adopting a least square algorithm to obtain a final parameter. The problems in the prior art are solved, and the precision of the digital pre-distortion model is improved.

Description

Technical Field

[0001] This invention belongs to the field of radio communication technology, and in particular to a digital predistortion method and system based on multi-strategy fusion model optimization. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In the field of wireless communication, the radio frequency power amplifier (PA) is the most critical component in base station equipment and also the most energy-intensive. With the development of broadband communication technologies such as 5G, radio frequency power amplifiers (PAs) typically need to operate in near-saturation regions to improve power amplifier efficiency, but this can lead to severe nonlinear distortion, causing problems such as adjacent channel interference, in-band distortion, and spectrum spread, which seriously affect communication quality.

[0004] Digital predistortion (DPD) is currently the most prevalent linearization technique. By configuring predistortion devices in the digital domain, it can compensate for the nonlinear distortion of power amplifiers. However, traditional digital predistortion techniques have certain shortcomings: determining the parameters in a predistortion scheme typically requires a large-scale structural search domain, resulting in high model complexity, high computational resource consumption, high hardware costs, and a lack of dynamic parameter adjustment mechanisms, leading to poor predistortion performance; using a single optimization algorithm makes it prone to getting trapped in local optima, resulting in slow convergence speed. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention provides a digital predistortion method and system based on multi-strategy fusion model optimization. It constructs a power amplifier behavior model and a predistortion model, and employs an improved particle swarm optimization algorithm and hill-climbing algorithm for global-then-local multi-strategy search to optimize the model and obtain the final parameters. This solves the problems of high-complexity nonlinear basis function modeling, large model search regions, and high computational resource consumption faced by existing digital predistortion technologies in broadband communication, thereby achieving efficient and low-complexity digital predistortion.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides a digital predistortion method based on multi-strategy fusion model optimization, comprising: Construct a power amplifier behavior model and a predistortion model; An improved particle swarm optimization algorithm is used to perform a global search optimization of the predistortion model coefficients; An adaptive degree function is constructed, and the optimization coefficients of the digital predistorter are initially obtained based on the global search optimization results; A preset proportion of optimization coefficients are selected from the optimization coefficients of the digital predistorter, and a hill-climbing algorithm is used for local fine-tuning optimization to obtain the target matching coefficients of the predistortion model. Based on the target matching coefficient of the predistortion model, with the optimization objective of minimizing the linear amplification output error between the actual output of the power amplifier after nonlinear compensation and the original input signal, the predistortion parameters are calculated and updated using the least squares algorithm to obtain the final parameters.

[0007] As one implementation method, a power amplifier behavior model and a predistortion model are constructed, and the specific process is as follows: A power amplifier behavior model is constructed using generalized memory polynomial basis functions; The inverse model of the power amplifier behavior model is used as the predistortion model; The model coefficients are optimized by minimizing the linear amplification error between the predistorter output signal and the original input signal after passing through the power amplifier. A cost function is constructed, and the basis functions are dynamically adjusted based on the input signal of the predistortion model and the working state of the power amplifier behavior model to obtain the final predistortion model.

[0008] As one implementation method, the power amplifier behavior model is formulated as follows: ; in, x ( n ) is the input signal. y GMP ( n ) is the output signal. K a , K b , K c For the largest nonlinear order, L a , L b , L c For maximum memory depth, M b The maximum lag length of the lag term. M c The maximum lead length of the leading term. a kl , b klm , c klm These are the model coefficients. Is the depth of memory? l sampling points, The lag length is m The sampling points of the lag term, For the lead length ism The leading sampling points.

[0009] As one implementation method, an improved particle swarm optimization algorithm is used to perform a global search optimization of the predistortion model coefficients. The specific process is as follows: Initialize the particle swarm, set the number of particles, the position of the particles represents the predistorter coefficients, each particle represents a candidate solution in the solution space, and set the initial position and velocity; Calculate the updated particle velocity based on the initial velocity; Set initial values ​​for inertia weights and update them dynamically; Calculate the updated particle position based on the initial position and the updated particle velocity; Update the historical best position by comparing the updated particle position with its historical best position; By using an improved particle swarm optimization algorithm, the particle velocity and position are updated iteratively, and the historical best position of each particle and the global best position of the particle swarm are recorded.

[0010] As one implementation method, an adaptive degree function is constructed, and based on the global search optimization results, the optimization coefficients of the digital predistorter are initially obtained. The specific process is as follows: The adaptive degree function is constructed using the mean square error method; Set the algorithm termination condition to stop optimization when the number of iterations reaches the set number or the adaptive function value is lower than the preset threshold. The optimal combination of predistorter coefficients is evaluated and selected by minimizing the adaptive function value.

[0011] As one implementation method, the formula for constructing the adaptive function is: ; in, It is the output of the initial input signal after passing through a linear ideal amplifier. The actual outputs of the signal after passing through the digital predistorter and the power amplifier, respectively. N It refers to the number of samples.

[0012] As one implementation method, a preset proportion of optimization coefficients are selected from the optimization coefficients of the digital predistorter, and a hill-climbing algorithm is used for local fine-tuning optimization to obtain the target matching coefficients of the predistortion model. The specific process is as follows: The top 10% of the optimization coefficients with the best adaptability are selected from the optimization coefficients of the digital predistorter and used as the initial values ​​for the hill-climbing algorithm. Based on the top 10% of optimization coefficients, the neighborhood perturbation vector is set as the search step size for iterative random optimization, and a neighborhood solution is generated in each iteration. By comparing the adaptive function values ​​of the current solution and the neighboring solutions, if the neighboring solution is smaller than the current solution, the current solution is updated to the neighboring solution; otherwise, the current solution remains unchanged. After multiple iterations, the predistorter coefficients approach the global optimal solution, thus obtaining the target configuration coefficients of the predistortion model.

[0013] As one implementation method, the predistortion parameters are calculated and updated according to the least squares algorithm, using the following formula: ; in, For the updated predistortion parameters, As a learning factor, The signal after pre-distortion processing. This is the basis function matrix of the power amplifier output signal obtained from the feedback loop.

[0014] A second aspect of the present invention provides a digital predistortion system based on a multi-strategy fusion model optimization, comprising: The model building module is used to build behavioral and predistortion models of power amplifiers. The global search optimization module is used to perform global search optimization on the predistortion model coefficients using an improved particle swarm optimization algorithm; an adaptive degree function is constructed, and the optimization coefficients of the digital predistorter are initially obtained based on the global search optimization results; The local fine-tuning module is used to select a preset proportion of optimization coefficients from the optimization coefficients of the digital predistorter, and use the hill-climbing algorithm to perform local fine-tuning optimization to obtain the target matching coefficients of the predistortion model. The update module is used to calculate and update the predistortion parameters based on the target matching coefficients of the predistortion model. The optimization objective is to minimize the linear amplification output error between the actual output of the power amplifier after nonlinear compensation and the original input signal. The least squares algorithm is used to obtain the final parameters.

[0015] As one implementation method, the model building module includes an input module, an up-conversion amplification module, a down-conversion sampling module, and a power amplifier.

[0016] The above one or more technical solutions have the following beneficial effects: In this embodiment, the model is constructed using generalized memory polynomial (GMP) basis functions and an adaptive optimization mechanism, which overcomes the shortcomings of traditional large-scale structure search domains, reduces model complexity and hardware costs, and reduces computational resource consumption. In this embodiment, a multi-strategy fusion search approach, combining global and local optimization, is employed. This approach integrates the advantages of both improved particle swarm optimization (PSO) and hill-climbing algorithms, addressing the problem of single optimization algorithms easily getting trapped in local optima and experiencing convergence difficulties. Specifically, an improved PSO algorithm is first used for global search, enabling the predistorter coefficients to quickly approach the optimal solution and selecting individuals with high fitness, thereby significantly shortening the convergence time and reducing model complexity. Then, the hill-climbing algorithm is used for local fine-tuning of the global search results to obtain the final parameters, further reducing the distortion of the power amplifier output signal and improving the accuracy of the digital predistortion model. By fusing two different optimization strategies, this approach demonstrates good adaptability and strong robustness for different types of power amplifier nonlinear characteristics.

[0017] In this embodiment, the improved particle swarm optimization algorithm incorporates a dynamic inertia weight adjustment mechanism, which can dynamically adjust particle positions to improve optimization capability and pre-distortion effect.

[0018] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0019] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0020] Figure 1 This is a flowchart of a digital predistortion method based on multi-strategy fusion model optimization according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram illustrating the optimization process of the improved particle swarm optimization algorithm and hill climbing algorithm according to Embodiment 1 of the present invention; Figure 3 This is a schematic diagram illustrating the principle of a digital predistortion system model based on multi-strategy fusion model optimization according to Embodiment 1 of the present invention; Figure 4 This is a diagram illustrating the effects of digital predistortion before and after optimization based on a multi-strategy fusion model, according to Embodiment 1 of the present invention. Figure 5 This is a schematic diagram of a digital predistortion system structure based on multi-strategy fusion model optimization according to Embodiment 2 of the present invention. Detailed Implementation

[0021] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0022] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0023] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0024] Terminology Explanation: Radio frequency power amplifier (PA): A radio frequency power amplifier is an electronic device specifically designed to amplify radio frequency signals. It is an indispensable key component in wireless communication systems, ensuring the effective transmission and reception of signals, and is also an important part of various wireless transmitters.

[0025] Digital pre-distortion (DPD): Digital pre-distortion (DPD) is an important method for suppressing nonlinear distortion in power amplifiers. This method adds the corresponding distortion information to the waveform shape of the transmitted signal in advance to achieve distortion cancellation, and plays a key role in modern communication systems.

[0026] Particle Swarm Optimization (PSO): The PSO algorithm is swarm-based, moving individuals within the swarm to better regions based on their fitness to the environment. However, it does not use evolutionary operators on individuals; instead, it treats each individual as a particle (point) with no volume in a D-dimensional search space, moving at a certain speed that is dynamically adjusted based on its own motion experience and that of its peers.

[0027] Hill Climbing Algorithm (HC): The hill climbing algorithm is a local optimization method that uses a heuristic approach. It is an improvement on depth-first search and uses feedback information to help generate solutions.

[0028] Example 1 This embodiment discloses a digital predistortion method based on multi-strategy fusion model optimization.

[0029] To more clearly illustrate this embodiment, a digital predistortion implementation process based on multi-strategy fusion model optimization can be specifically described as follows: A digital predistortion method based on multi-strategy fusion model optimization includes: S1. Construct a power amplifier behavior model and a predistortion model; S2. An improved particle swarm optimization algorithm is used to perform a global search optimization of the predistortion model coefficients; S3. Construct an adaptive function and, based on the global search optimization results, initially obtain the optimization coefficients of the digital predistorter; S4. Select a preset proportion of optimization coefficients from the optimization coefficients of the digital predistorter, and use the hill climbing algorithm to perform local fine-tuning optimization to obtain the target matching coefficients of the predistortion model. S5. Based on the target matching coefficient of the predistortion model, with the optimization objective of minimizing the linear amplification output error between the actual output of the power amplifier after nonlinear compensation and the original input signal, the predistortion parameters are calculated and updated using the least squares algorithm to obtain the final parameters.

[0030] like Figure 1 As shown, in step S1, a power amplifier behavior model and a predistortion model are constructed.

[0031] In this embodiment, the power amplifier behavior model and predistortion model are constructed, and the specific process is as follows: (1) Use generalized memory polynomial basis functions to construct a power amplifier behavior model.

[0032] We select generalized memory polynomial (GMP) basis functions that can better describe the nonlinearity and memory characteristics of RF power amplifiers to construct the power amplifier behavior model and predistortion model. The power amplifier behavior model is the same as the power amplifier behavior model.

[0033] The input signal is x ( n When ), the output signal of the power amplifier behavior model is: (1) in, x ( n ) is the input signal. y GMP ( n ) is the output signal. K a , K b , K c For the largest nonlinear order, L a , L b , L c For maximum memory depth, M b The maximum lag length of the lag term. M c The maximum lead length of the leading term. a kl , b klm , c klm These are the model coefficients. Is the depth of memory? l sampling points, The lag length is m The sampling points of the lag term, The lead length is m The leading sampling points.

[0034] (2) The inverse model of the power amplifier behavior model is used as the predistortion model.

[0035] The inverse model of a power amplifier is the predistortion model.

[0036] (3) The model coefficients are optimized based on minimizing the linear amplification output error between the actual output of the power amplifier after nonlinear compensation and the original input signal.

[0037] After multiple iterations and adjustments of the coefficients, the output signal of the predistorter, after passing through the power amplifier, is made as close as possible to the original input signal. The most suitable predistortion model is then searched based on the input signal and the operating state of the power amplifier.

[0038] (4) Construct a cost function and dynamically adjust the basis function according to the input signal of the predistortion model and the working state of the power amplifier behavior model to obtain the final predistortion model.

[0039] Establish an appropriate cost function, where the cost function is represented by the Error Vector Magnitude (EVM), which is an important indicator for measuring the accuracy of predistortion modeling. Its expression is as follows: (2) in, s ( n )and sideal ( n The values ​​() represent the actual output signal and the ideal output signal after demodulation using the corresponding demodulation methods, respectively. The smaller the EVM value, the higher the DPD modeling accuracy of the system and the better the nonlinear compensation effect.

[0040] Based on the obtained search space content, the search space range is narrowed, and the basis function is adaptively optimized. That is, the basis function is adjusted or optimized according to the actual information such as the power amplifier characteristics and the characteristics of the input signal, which effectively reduces the complexity of the problem.

[0041] Through the above steps, a model is constructed using generalized memory polynomial (GMP) basis functions to capture more effective dynamic memory effects in the system, more accurately fit complex nonlinear behaviors, and determine model parameters through offline training. This overcomes the shortcomings of traditional large-scale structure search domains and reduces computational resource consumption.

[0042] like Figure 1 , Figure 2 As shown, in step S2, an improved particle swarm optimization algorithm is used to perform a global search optimization on the predistortion model coefficients.

[0043] In this embodiment, an improved particle swarm optimization algorithm with strong global search capabilities is used for global search. It searches for possible optimal solutions in a large solution space, locating the approximate region of the optimal solution. This compensates for the nonlinearity, memory effect, and inter-branch crosstalk effect of the MIMO system's transmitter power amplifier.

[0044] The specific process is as follows: (1) Initialize the particle swarm, set the number of particles, the position of the particles represents the predistorter coefficient, each particle represents a candidate solution in the solution space, and set the initial position and velocity.

[0045] The particle swarm optimization algorithm performs initial optimization on the coefficients of the digital predistortion model. The number of particles is set to 30, and the position of the particle represents the coefficient of the predistorter. Each particle is treated as a candidate solution in the solution space.

[0046] Based on the input path planning data, the particle swarm is initialized, and its initial position and velocity are set.

[0047] (2) Calculate the updated particle velocity based on the initial velocity.

[0048] The speed update formula is: (3) in, λ 1 and λ 2 is the learning factor, all set to 1.5; and These are random numbers, and their values ​​are all in the range [0,1]. This represents the historical best position of the i-th particle; It is the globally optimal position of the entire particle swarm; α Inertial weights; For the first i The particle in the first b In 3D space t The position at that moment; For the first i The particle in the first b In 3D space t The speed of time.

[0049] (3) Set the initial value of the inertia weight and update it dynamically.

[0050] The initial value of the inertia weight is set to 0.7, and the inertia weight update formula is: (4) in, It follows a standard normal distribution. It is a random number between 0 and 1. , , All are constants.

[0051] After the above steps, the dynamic inertia weight adjustment mechanism can dynamically adjust the particle position, thereby improving the optimization capability and pre-distortion effect.

[0052] (4) Calculate the updated particle position based on the initial position and the updated particle velocity.

[0053] The particle velocity and position are calculated according to the particle velocity update formula (3) and the inertia weight update formula (4). The particle position update formula is as follows: (5) in, The updated particle position, i.e., the [number]th [particle position]. i The particle in the first b In 3D space t+ Position at time 1; For the updated particle velocity, i.e., the... i The particle in the first b In 3D space t+ The velocity at moment 1.

[0054] (5) Update the historical best position by comparing the updated particle position with its historical best position.

[0055] The current position of each particle is compared with its historical best position. If the current position is better, the historical best position is updated.

[0056] (6) By using the improved particle swarm algorithm, the particle velocity and position are updated iteratively, and the historical best position of each particle and the global best position of the particle swarm are recorded.

[0057] Through the above steps, the predistorter coefficients can quickly approach the optimal solution, select individuals with high fitness, and thus greatly shorten the convergence time and reduce the model complexity.

[0058] like Figure 1 , Figure 2 As shown, in step S3, an adaptive function is constructed, and the optimization coefficients of the digital predistorter are initially obtained based on the global search optimization results.

[0059] To evaluate the combination of basis function parameters, a fitness function is constructed: The specific process is as follows: (1) The adaptive degree function is constructed using the mean square error method.

[0060] The fitness function is set as follows: in predistortion, the objective fitness function is to minimize the error between the original linearly amplified input signal and the nonlinearly compensated power amplifier output signal. The predistorter coefficients for each particle are calculated. A typical error metric is mean square error (MSE). The fitness function is then... E The formula is: (6) in, It is the output of the initial input signal after passing through a linear ideal amplifier. The actual outputs of the signal after passing through the digital predistorter and the power amplifier, respectively. N It refers to the number of samples.

[0061] (2) Set the algorithm termination condition. When the number of iterations reaches the set number or the adaptive function value is lower than the preset threshold, the optimization will stop.

[0062] Set a termination condition, here set to a maximum number of iterations of 50, or a fitness value below a certain threshold, then stop the search and output the optimal solution; if the termination condition is met, the optimization coefficients of the digital predistorter are initially obtained.

[0063] (3) The optimal combination of predistorter coefficients is evaluated and selected by minimizing the adaptive function value.

[0064] Through the above steps, the digital predistorter achieves a balanced optimization of efficiency and accuracy, shortening the predistorter's startup and calibration time while ensuring coefficient accuracy. This enables controllability of the optimization process, avoids ineffective iterations, and reduces implementation complexity.

[0065] like Figure 1 , Figure 2 As shown, in step S4, a preset proportion of optimization coefficients are selected from the optimization coefficients of the digital predistorter, and a hill-climbing algorithm is used for local fine-tuning optimization to obtain the target matching coefficients of the predistortion model.

[0066] The specific process is as follows: (1) Select the top 10% of the optimization coefficients with the best adaptability from the optimization coefficients of the digital predistorter as the initial values ​​of the hill climbing algorithm.

[0067] The top 10% of historically optimal optimization coefficients obtained after iterative iterations of the particle swarm optimization algorithm are selected as the initial values ​​for local optimization in the hill-climbing algorithm. Further refinement of the optimization process involves iteratively approaching the optimal solution from the neighborhood of the current solution. This global-then-local search approach combines the advantages of both algorithms.

[0068] (2) Based on the top 10% of optimization coefficients, the neighborhood perturbation vector is set as the search step size for iterative random optimization, and a neighborhood solution is generated in each iteration.

[0069] Set the total number of iterations for the hill-climbing algorithm. m The predistorter coefficient vector is Neighborhood perturbation vector The random search step size is set to 0.01.

[0070] Calculate the current solution and the neighborhood solutions. and The objective function value. The objective function needs to quantify the predistortion effect, using mean squared error (MSE) as the objective function. According to formula (6), the objective function value of the current solution is: (7) in, This is the actual output of the power amplifier after predistortion. It is an ideal linear output.

[0071] The objective function value of the neighborhood solution is: (8) (3) Compare the objective function values ​​of the current solution and the neighboring solutions. If the objective function value of the neighboring solution is smaller than that of the current solution, then the objective function is satisfied. If the condition is met, the current solution is updated to the neighborhood solution; otherwise, the current solution remains unchanged.

[0072] (4) After multiple iterations, the predistorter coefficients approach the global optimal solution, and the target configuration coefficients of the predistorter model are obtained.

[0073] Through the above steps, the distortion of the power amplifier output signal is further reduced, the accuracy of the digital predistortion model is improved, and two different optimization strategies are integrated. This approach has good adaptability and strong robustness for different types of power amplifier nonlinear characteristics.

[0074] like Figure 1 As shown, in step S5, based on the target matching coefficient of the predistortion model, with the optimization objective of minimizing the linear amplification output error between the actual output of the power amplifier after nonlinear compensation and the original input signal, the predistortion parameters are calculated and updated using the least squares algorithm to obtain the final parameters.

[0075] The model is trained using the model training module. When the switch is closed and the system begins to work, the signal after pre-distortion processing can be represented as follows: (9) in, The basis function matrix of the power amplifier output signal obtained from the feedback loop is... It is the predistortion parameter.

[0076] The optimization objective is to minimize the error between the actual output of the power amplifier after nonlinear compensation and the linear amplification output of the original input signal. The predistortion parameters are calculated and updated using the least squares algorithm to obtain the final parameters, as shown in the formula: (10) in, For the updated predistortion parameters, As a learning factor, The signal after pre-distortion processing. This is the basis function matrix of the power amplifier output signal obtained from the feedback loop.

[0077] Through the above steps, the predistorter is precisely matched to the inverse characteristics of the power amplifier, achieving comprehensive compensation for nonlinear distortion, significantly reducing the bit error rate, and improving signal reliability.

[0078] like Figure 3 As shown, in this embodiment, the working steps of the digital predistortion system model optimized based on the multi-strategy fusion model are as follows: (1) The optimal model structure is searched by fusion, with switch 1 in the open state and switch 2 in the closed state. First, the particle swarm optimization algorithm is used to perform a global search and select individuals with higher fitness.

[0079] (2) Based on step (1), the global historical best individual in the particle swarm algorithm is used as the initial value of the hill climbing algorithm, and the surrounding area is refined and optimized. The final optimal model structure is denoted as Dbest.

[0080] (3) Disconnect switch 2, turn on switch 1, and use Dbest to train the model parameters of DPD based on the indirect learning structure.

[0081] (4) Update the iterative model and close switches 1 and 2 at the same time. When the actual working state of the power amplifier changes significantly, update the DPD model in a timely manner.

[0082] Digital predistortion operations are performed on a digital predistortion system model optimized based on a multi-strategy fusion model, such as... Figure 4 The test results of power spectral density before and after predistortion for the multi-strategy fusion model optimization digital predistortion method of this embodiment and the traditional single-model search method are provided. As can be seen from the figure, the power spectral density of the traditional single-model search predistortion method is optimized by about 14dB compared with the method without predistortion, while the proposed multi-strategy fusion model search predistortion method is optimized by about 29dB, which significantly improves the nonlinear correction effect.

[0083] Example 2 like Figure 5 As shown, the purpose of this embodiment is to provide a digital predistortion system based on a multi-strategy fusion model optimization, including: The model building module is used to build behavioral and predistortion models of power amplifiers. The global search optimization module is used to perform global search optimization on the predistortion model coefficients using an improved particle swarm optimization algorithm; an adaptive degree function is constructed, and the optimization coefficients of the digital predistorter are initially obtained based on the global search optimization results; The local fine-tuning module is used to select a preset proportion of optimization coefficients from the optimization coefficients of the digital predistorter, and use the hill-climbing algorithm to perform local fine-tuning optimization to obtain the target matching coefficients of the predistortion model. The update module is used to nonlinearly compensate for the linear amplification error between the actual output of the power amplifier and the original input signal, based on the target matching coefficients of the predistortion model. most To optimize the parameters, the least squares algorithm is used to calculate and update the predistortion parameters, thus obtaining the final parameters.

[0084] Based on a digital predistortion system optimized by a multi-strategy fusion model, the method steps in Embodiment 1 are implemented.

[0085] The model building module includes an input module, an up-conversion amplification module, a down-conversion sampling module, and a power amplifier.

[0086] The input module is used to perform inverse power amplifier model processing on the input signal in the predistorter to obtain the predistorted signal.

[0087] The up-conversion amplifier module takes the predistorter output signal, converts it from digital to analog and then up-converts it before inputting it into the power amplifier for signal amplification.

[0088] The downconversion sampling module performs downconversion, band-limited filtering, and low-speed sampling of the power amplifier output signal, and finally obtains the power amplifier output spectrum.

[0089] The global search optimization module, local fine-tuning optimization module, and update module construct a fusion model coefficient extraction module. The fusion model coefficient extraction module, based on the multi-strategy fusion basis function optimization search method, determines the final optimized basis function parameter combination after the global search of the particle swarm algorithm and the local fine-tuning search of the hill climbing algorithm, and then extracts the model coefficients by combining the indirect learning structure of the predistortion system.

[0090] The steps and methods involved in the apparatus of the above embodiments correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1.

[0091] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0092] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A digital predistortion method based on multi-strategy fusion model optimization, characterized in that, include: Construct a power amplifier behavior model and a predistortion model; An improved particle swarm optimization algorithm is used to perform a global search optimization of the predistortion model coefficients; An adaptive degree function is constructed, and the optimization coefficients of the digital predistorter are initially obtained based on the global search optimization results; A preset proportion of optimization coefficients are selected from the optimization coefficients of the digital predistorter, and a hill-climbing algorithm is used for local fine-tuning optimization to obtain the target matching coefficients of the predistortion model. Based on the target matching coefficient of the predistortion model, with the optimization objective of minimizing the linear amplification output error between the actual output of the power amplifier after nonlinear compensation and the original input signal, the predistortion parameters are calculated and updated using the least squares algorithm to obtain the final parameters.

2. The digital predistortion method based on multi-strategy fusion model optimization as described in claim 1, characterized in that, The specific process for constructing the power amplifier behavior model and predistortion model is as follows: A power amplifier behavior model is constructed using generalized memory polynomial basis functions; The inverse model of the power amplifier behavior model is used as the predistortion model; The model coefficients are optimized by minimizing the error between the predistorter output signal and the original input signal after passing through the power amplifier. A cost function is constructed, and the basis functions are dynamically adjusted based on the input signal of the predistortion model and the working state of the power amplifier behavior model to obtain the final predistortion model.

3. The digital predistortion method based on multi-strategy fusion model optimization as described in claim 2, characterized in that, The power amplifier behavior model, the formula is: ； in, x ( n ) is the input signal. y GMP ( n ) is the output signal. K a , K b , K c For the largest nonlinear order, L a , L b , L c For maximum memory depth, M b The maximum lag length of the lag term. M c The maximum lead length of the leading term. a kl , b klm , c klm These are the model coefficients. Is the depth of memory l sampling points, The lag length is m The sampling points of the lag term, For the lead length is m The leading sampling points.

4. The digital predistortion method based on multi-strategy fusion model optimization as described in claim 1, characterized in that, An improved particle swarm optimization algorithm is used to perform a global search optimization of the predistortion model coefficients. The specific process is as follows: Initialize the particle swarm, set the number of particles, the position of the particles represents the predistorter coefficients, each particle represents a candidate solution in the solution space, and set the initial position and velocity; Calculate the updated particle velocity based on the initial velocity; Set initial values ​​for inertia weights and update them dynamically; Calculate the updated particle position based on the initial position and the updated particle velocity; Update the historical best position by comparing the updated particle position with its historical best position; By using an improved particle swarm optimization algorithm, the particle velocity and position are updated iteratively, and the historical best position of each particle and the global best position of the particle swarm are recorded.

5. The digital predistortion method based on multi-strategy fusion model optimization as described in claim 1, characterized in that, An adaptive degree function is constructed, and based on the global search optimization results, the initial optimization coefficients of the digital predistorter are obtained. The specific process is as follows: The adaptive degree function is constructed using the mean square error method; Set the algorithm termination condition to stop optimization when the number of iterations reaches the set number or the adaptive function value is lower than the preset threshold. The optimal combination of predistorter coefficients is evaluated and selected by minimizing the adaptive function value.

6. The digital predistortion method based on multi-strategy fusion model optimization as described in claim 5, characterized in that, The formula for constructing the adaptive function is: ； in, It is the output of the initial input signal after passing through a linear ideal amplifier. The actual outputs of the signal after passing through the digital predistorter and the power amplifier, respectively. N It refers to the number of samples.

7. The digital predistortion method based on multi-strategy fusion model optimization as described in claim 1, characterized in that, A preset proportion of optimization coefficients are selected from the optimization coefficients of the digital predistorter, and a hill-climbing algorithm is used for local fine-tuning optimization to obtain the target matching coefficients of the predistortion model. The specific process is as follows: The top 10% of the optimization coefficients with the best adaptability are selected from the optimization coefficients of the digital predistorter and used as the initial values ​​for the hill-climbing algorithm. Based on the top 10% of optimization coefficients, the neighborhood perturbation vector is set as the search step size for iterative random optimization, and a neighborhood solution is generated in each iteration. By comparing the adaptive function values ​​of the current solution and the neighboring solutions, if the neighboring solution is smaller than the current solution, the current solution is updated to the neighboring solution; otherwise, the current solution remains unchanged. After multiple iterations, the predistorter coefficients approach the global optimal solution, thus obtaining the target configuration coefficients of the predistortion model.

8. The digital predistortion method based on multi-strategy fusion model optimization as described in claim 1, characterized in that, The predistortion parameters are calculated and updated using the least squares algorithm, as follows: ； in, For the updated predistortion parameters, As a learning factor, The signal after pre-distortion processing. This is the basis function matrix of the power amplifier output signal obtained from the feedback loop.

9. A digital predistortion system based on a multi-strategy fusion model optimization, characterized in that, include: The model building module is used to build behavioral and predistortion models of power amplifiers. The global search optimization module is used to perform global search optimization on the predistortion model coefficients using an improved particle swarm optimization algorithm; an adaptive degree function is constructed, and the optimization coefficients of the digital predistorter are initially obtained based on the global search optimization results; The local fine-tuning module is used to select a preset proportion of optimization coefficients from the optimization coefficients of the digital predistorter, and use the hill-climbing algorithm to perform local fine-tuning optimization to obtain the target matching coefficients of the predistortion model. The update module is used to calculate and update the predistortion parameters based on the target matching coefficients of the predistortion model. The optimization objective is to minimize the linear amplification output error between the actual output of the power amplifier after nonlinear compensation and the original input signal. The least squares algorithm is used to obtain the final parameters.

10. A digital predistortion system based on multi-strategy fusion model optimization as described in claim 9, characterized in that, The model building module includes an input module, an up-conversion amplification module, a down-conversion sampling module, and a power amplifier.

Images