Signal optimization method of ADAM algorithm assisted multi-stage logarithmic amplifier

Abstract

The invention provides a signal optimization method of an ADAM algorithm assisted multistage logarithmic amplifier, which comprises the following steps of: training an initialized optimal parameter group, and obtaining optimal parameters of the multistage logarithmic amplifier in an iterative updating mode. The iterative updating process comprises the following steps: setting a scale factor and a range of a pole number; the optimal score is initialized to be infinite, and the optimal parameter is set to be 0; traversing the cyclic scale factor and each parameter combination in the pole number range, and generating a test data set according to each combination; predicting the test data set by using a deep learning model, and obtaining a current score according to a mean absolute error between the test data set and a real result; updating the current score as the optimal score, and updating the scale factor and the pole number; and obtaining an optimal parameter after the traversal is finished. According to the method, the recovery of the multi-stage logarithmic amplifier to the original signal can be optimized, and the parameter combination with the optimal compensation effect can be found.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a signal optimization method for a multi-stage logarithmic amplifier assisted by an ADAM algorithm. Background Art

[0002] Free-space optics (FSO) uses lasers as a carrier to transmit data through the atmosphere. This technology combines the advantages of traditional wireless microwave communications and fiber-optic communications and is widely used in satellite communications, military communications, and disaster response. In the defense and military fields, information warfare is a major form of warfare in the future. The communication method used plays a crucial role in determining the outcome of a war, even directly affecting its trajectory. FSO, due to its high confidentiality and reliability, is widely used in the military. FSO technology has also been widely adopted in the commercial sector, including wireless video surveillance, ground transmission, last-mile solutions, space communications, radio astronomy, and remote sensing.

[0003] However, FSO also has some drawbacks: during atmospheric transmission, it is easily affected by atmospheric molecules and other gas molecules, leading to signal attenuation, multi-beam effects, light intensity flicker and drift, and other problems, thus affecting link reliability. To address this issue, researchers at home and abroad have proposed various compensation methods, such as threshold detection, aperture averaging, coding and modulation, and adaptive optics. Taking threshold detection as an example, two currently popular threshold detection technologies are adaptive threshold detection (ATD) and fixed threshold detection (FTD). However, because fixed threshold detection cannot accurately recover the baseband signal at the receiving end, a multi-stage logarithmic amplifier is used to compensate the gain of the received signal. Fixed threshold detection is then used on the compensated signal to restore the original signal.

[0004] Multi-stage logarithmic amplifiers continuously adjust various parameter combinations to find the optimal performance parameter set for various turbulence intensities. However, this parameter optimization process is achieved manually, so it is inevitable that the optimal parameter set may be missed. If the optimal parameter set is missed, it can lead to signal distortion during the recovery process, signal loss or incomplete recovery, insufficient or excessive signal amplification, and excessive attenuation or amplification of certain frequency components. Summary of the Invention

[0005] In order to overcome the deficiencies of the prior art, the present invention aims to provide a signal optimization method for a multi-stage logarithmic amplifier assisted by an ADAM algorithm, which can optimize the restoration of the original signal by the multi-stage logarithmic amplifier and find the parameter combination with the best compensation effect.

[0006] To achieve the above object, the present invention provides the following solution: a signal optimization method for a multi-stage logarithmic amplifier assisted by an ADAM algorithm, comprising the following steps:

[0007] S1. Using a multi-stage logarithmic amplifier to obtain an optimal performance parameter group for a signal, and performing a parameter initialization operation on the optimal parameter group;

[0008] S2. Using a deep learning model, combined with an adaptive learning rate algorithm and a momentum method, the initialized optimal parameter group is trained, and optimal parameters of the multi-stage logarithmic amplifier are obtained through iterative updating. The iterative updating process includes:

[0009] S21. Based on the multi-stage logarithmic amplifier, the scale factor is set to a range between 1 and 2, and the number of poles is set to a range between 1 and 20;

[0010] S22, initializing the best score to infinity and the optimal parameter to 0;

[0011] S23, traversing and looping each parameter combination within the range of the scale factor and the number of poles, and generating a test data set according to each combination;

[0012] S24. Use the deep learning model to predict the test data set to obtain a prediction result, calculate the mean absolute error between the prediction result and the true result, and obtain the current score;

[0013] S25. If the current score is better than the historical score, update the current score to the best score, and update the proportional factor and the number of poles;

[0014] S26. Repeat steps S23-S25 until all parameter combinations of the scale factor and the number of poles are traversed to obtain the optimal parameters of the multi-stage logarithmic amplifier.

[0015] Optionally, using a deep learning model in combination with an adaptive learning rate algorithm and a momentum method to train the initialized optimal parameter set includes:

[0016] Initializing the weights and biases in the neural network and the first-order moment estimates, second-order moment estimates, and time steps of the optimal parameter set to 0;

[0017] Combined with the loss function, calculate the gradient under the current parameters to get the current gradient;

[0018] Performing mean estimation and square mean estimation on the current gradient to complete the update of the first-order moment estimation and the second-order moment estimation;

[0019] The updated first-order moment estimate and second-order moment estimate are corrected for deviations, and then parameter update training is performed. Optionally, the update expression of the first-order moment estimate is:

[0020] m t =β1m t-1 +(1-β1)g t

[0021] Among them, m t is the mean variable, β1 is a hyperparameter set to 0.9, and t is the time step.

[0022] Optionally, the update expression for the second-order moment estimate is:

[0023]

[0024] Among them, v t is the square mean variable, β2 is a hyperparameter set to 0.999, and t is the time step.

[0025] Optionally, the bias correction expression is:

[0026]

[0027] in, is the corrected first-order moment estimate, is the corrected second-order moment estimate, and t is the time step.

[0028] Optionally, the expression for parameter update is:

[0029]

[0030] Among them, θ t is the parameter value of the current time step, θ t+1 is the updated parameter value, α is the learning rate used to control the parameter update step size, and ∈ is a constant.

[0031] The present invention provides a signal optimization method for a multi-stage logarithmic amplifier assisted by an ADAM algorithm, and discloses the following technical effects:

[0032] 1. Adaptive adjustment: The ADAM algorithm automatically finds the optimal parameter combination without manual debugging one by one, greatly improving the efficiency and accuracy of parameter optimization.

[0033] 2. Comprehensive optimization: The ADAM algorithm can traverse more parameter combinations, reduce the omission of optimal parameter combinations, and ensure the best signal recovery effect under different turbulence intensities.

[0034] 3. Accurate compensation: Using the ADAM algorithm to process the received signal can more accurately compensate for the impact of turbulence on the signal and improve the quality of signal recovery.

[0035] 4. Strong adaptability: The method of the present invention can adapt to the signal processing requirements under different turbulence intensities and has strong environmental adaptability and robustness.

[0036] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 A schematic diagram of a method flow chart provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0040] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] like Figure 1 As shown, the present invention provides a signal optimization method for a multi-stage logarithmic amplifier assisted by an ADAM algorithm, comprising the following steps:

[0042] S1. Utilize a multi-stage logarithmic amplifier to obtain an optimal performance parameter group for a signal, and perform a parameter initialization operation on the optimal parameter group.

[0043] S2. Using a deep learning model, combined with an adaptive learning rate algorithm and a momentum method (to obtain an ADAM algorithm), the initialized optimal parameter group is trained, and the optimal parameters of the multi-stage logarithmic amplifier are obtained through iterative updating.

[0044] (1) The ADAM algorithm implements adaptive learning rate adjustment and model parameter optimization through the following steps:

[0045] Step 1: Initialize parameters: Initialize the parameters θ0, first-order moment estimate m0 = 0, second-order moment estimate v0 = 0, and time step t = 0. The parameters here can be weights and biases in the neural network.

[0046] Step 2: Calculate the gradient: For the given input data and loss function, calculate the current parameter θ t The gradient g t The loss function is used to assess the degree of inconsistency between a model's predictions and the true values. It is also the objective function for neural network optimization. The process of neural network training or optimization is to minimize the loss function. The smaller the loss function, the closer the model's predictions are to the true values, and the more robust the model is. The gradient represents the rate of change of the loss function with respect to the parameters and is used to guide the direction of parameter updates.

[0047] Step 3: Update the first-order moment estimate: According to the formula m t =β1m t-1 +(1-β1)g t Update the first-order moment estimate. The first-order moment estimate is also called the mean estimate, which is an estimate of the mean of the gradient, using the variable m t β1 is a hyperparameter between 0 and 1 that controls the exponential decay rate of the first-order moment estimate, typically set to 0.9. It determines the contribution of past gradients to the current first-order moment estimate. The first-order moment estimate can be understood as a gradient accumulation, which takes into account information from past gradients and helps accelerate convergence and escape from local optima.

[0048] Step 4: Update the second-order moment estimate: According to the formula Update the second-order moment estimate v t The second-order moment estimation is to estimate the mean of the square of the gradient, using the variable v t β2 is also a hyperparameter between 0 and 1 that controls the exponential decay of the second-order moment estimate, typically set to 0.999. It determines the contribution of past squared gradients to the current second-order moment estimate. The second-order moment estimate measures the magnitude of the gradient change and is used to adjust the learning rate, allowing the algorithm to adaptively adjust the update step size for different parameters.

[0049] Step 5: Bias correction: Since one of the design goals of the algorithm is to maintain the stability of the algorithm in non-stationary and online learning scenarios, in the initial stage, m t and v t A bias towards 0 helps the algorithm learn more robustly in the early stages of training. To eliminate this bias, bias correction is performed.

[0050] Corrected first-order moment estimate:

[0051]

[0052] Corrected second moment estimate

[0053]

[0054] The purpose of bias correction is to estimate the mean and variance of the gradient more accurately in the early stages of the algorithm, so that the algorithm can perform parameter updates more stably in the initial stage.

[0055] Step 6: Parameter update: With the corrected first-order moment estimate and second-order moment estimate, the parameters can be updated. The update formula is:

[0056]

[0057] Among them, θ t is the parameter value at the current time step t, θ t+1 is the updated parameter value, α is the learning rate, and ∈ is a small constant (usually 10 -8 ) is primarily used to prevent the denominator from being zero, thus stabilizing the parameter update calculation. The parameter update process calculates new parameter values ​​based on the current parameter values, the learning rate, and the first-order and second-order moment estimates. The learning rate controls the update step size, the first-order moment estimate influences the update direction, and the second-order moment estimate adjusts the learning rate, allowing the algorithm to adaptively adjust the update step size based on gradient changes.

[0058] (2) Algorithm structure

[0059] RMSProp algorithm: RMSProp (Root Mean Square Propagation) is an adaptive learning rate algorithm used to optimize the parameters of machine learning models. It dynamically adjusts the learning rate by taking an exponentially weighted average of the squared gradients, enabling the model to converge to the optimal solution more quickly during training.

[0060] Momentum method: The basic idea of ​​the momentum method is to consider not only the current gradient but also the update direction of the previous iteration in each iteration, and perform a weighted average of the current gradient and the update direction of the previous iteration to obtain a new update direction. This allows the model to traverse flat areas faster and move towards the optimal solution while avoiding falling into a local optimal solution.

[0061] (3) The iterative update process includes:

[0062] S21. Based on the multi-stage logarithmic amplifier, the range of the scale factor b is set to be between 1 and 2, and the range of the number of poles N is set to be between 1 and 20;

[0063] S22, initialize the best score bestScore to infinity, and set the optimal parameters optimB and optimN to 0;

[0064] S23, traversing and looping each parameter combination within the range of the scale factor and the number of poles, and generating a test data set according to each combination;

[0065] S24. Use the deep learning model to predict the test data set to obtain a prediction result, calculate the mean absolute error between the prediction result and the true result, and obtain the current score currentScore;

[0066] S25. If the current score is better than the historical score, update the current score to the best score bestScore, and update the scale factor b and the number of poles N;

[0067] S26. Repeat steps S23-S25 until all parameter combinations of the proportional factor b and the number of poles N are traversed, and the optimal parameters of the multi-stage logarithmic amplifier (optimal parameter combination b and N) are obtained. Finally, the optimal parameters optimB and optimN are output.

[0068] Therefore, the present invention provides a signal optimization method for a multi-stage logarithmic amplifier assisted by an ADAM algorithm, which can optimize the restoration of the original signal by the multi-stage logarithmic amplifier and find the parameter combination with the best compensation effect.

[0069] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0070] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A signal optimization method for a multi-stage logarithmic amplifier assisted by an ADAM algorithm, characterized in that: The following steps are involved: S1. Using a multi-stage logarithmic amplifier to obtain an optimal performance parameter group for a signal, and performing a parameter initialization operation on the optimal parameter group; S2. Using a deep learning model, combined with an adaptive learning rate algorithm and a momentum method, the initialized optimal parameter group is trained, and optimal parameters of the multi-stage logarithmic amplifier are obtained through iterative updating. The iterative updating process includes: S21. Based on the multi-stage logarithmic amplifier, the scale factor is set to a range between 1 and 2, and the number of poles is set to a range between 1 and 20; S22, initializing the best score to infinity and the optimal parameter to 0; S23, traversing and looping each parameter combination within the range of the scale factor and the number of poles, and generating a test data set according to each combination; S24. Use the deep learning model to predict the test data set to obtain a prediction result, calculate the mean absolute error between the prediction result and the true result, and obtain the current score; S25. If the current score is better than the historical score, update the current score to the best score, and update the proportional factor and the number of poles; S26. Repeat steps S23-S25 until all parameter combinations of the scale factor and the number of poles are traversed to obtain the optimal parameters of the multi-stage logarithmic amplifier.

2. The signal optimization method of a multi-stage logarithmic amplifier assisted by an ADAM algorithm according to claim 1, characterized in that: The initialized optimal parameter set is trained using a deep learning model in combination with an adaptive learning rate algorithm and a momentum method, including: Initializing the weights and biases in the neural network and the first-order moment estimates, second-order moment estimates, and time steps of the optimal parameter set to 0; Combined with the loss function, calculate the gradient under the current parameters to get the current gradient; Performing mean estimation and square mean estimation on the current gradient to complete the update of the first-order moment estimation and the second-order moment estimation; The updated first-order moment estimate and second-order moment estimate are corrected for deviations, and then parameter update training is performed.

3. The signal optimization method of a multi-stage logarithmic amplifier assisted by an ADAM algorithm according to claim 2, characterized in that: The update expression of the first-order moment estimate is: m t =β1m t-1 +(1-β1)g t Among them, m t is the mean variable, β1 is a hyperparameter set to 0.9, and t is the time step.

4. The signal optimization method of a multi-stage logarithmic amplifier assisted by an ADAM algorithm according to claim 3, characterized in that: The update expression of the second-order moment estimate is: Among them, v t is the square mean variable, β2 is a hyperparameter set to 0.999, and t is the time step.

5. The signal optimization method of a multi-stage logarithmic amplifier assisted by an ADAM algorithm according to claim 4, characterized in that: The expression for bias correction is: in, is the corrected first-order moment estimate, is the corrected second-order moment estimate, and t is the time step.

6. The signal optimization method of a multi-stage logarithmic amplifier assisted by an ADAM algorithm according to claim 5, characterized in that: The expression for parameter update is: Among them, θ t is the parameter value of the current time step, θ t+1 is the updated parameter value, α is the learning rate used to control the parameter update step size, and ∈ is a constant.

Images