A satellite signal nonlinear distortion compensation method based on neural network

Abstract

This invention presents a neural network-based method for compensating for nonlinear distortion in satellite signals. The method establishes a nonlinear distortion model of the satellite signal and, for the scaling factor in the model, establishes a post-distortion compensation model. A backpropagation neural network (BPNN) is used to estimate the scaling factor of the post-distortion compensation model. A particle swarm optimization method is introduced to optimize the initial weights and biases of the BPNN. Satellite signal training samples are generated to train the BPNN. The trained network is then used to generate the scaling factor of the post-distortion compensation model, thereby achieving nonlinear distortion compensation for the satellite signal. This method requires no complex mathematical derivations or assumptions, reducing the impact of nonlinear distortion on the system with relatively low complexity.

Description

Technical Field

[0001] This invention belongs to the field of communication signal detection, and in particular relates to a method for compensating for nonlinear distortion of satellite signals based on neural networks. Background Technology

[0002] Currently, common methods for compensating for nonlinear distortion include transmitter predistortion and receiver post-compensation. At the transmitter, signal predistortion can be used to linearize the characteristics of the amplifier, involving related analog electronics to achieve the inverse of the amplifier characteristics. However, signal predistortion introduces out-of-band spread, and this method is not used in satellite communication due to the strong nonlinearity of the traveling wave tube amplifier (TWTA). On the other hand, digital predistortion is a purely digital technique that preserves the signal spectrum. By modifying the transmitter constellation, the centroid of the receiver coincides with the original transmitter constellation. However, predistortion technology has strict requirements on the size and power consumption of the transmitter. Therefore, in research on compensating for nonlinear distortion in satellite return links, post-distortion compensation techniques at the receiver end are more commonly used.

[0003] In research on solutions to nonlinear distortion at the receiver end, numerous studies have employed iterative algorithms to address system nonlinear distortion problems. However, these studies all require complex calculations and assumptions, leading to issues such as slow computation speed and low computational accuracy. Summary of the Invention

[0004] Purpose of the invention: In order to solve the problems existing in the prior art, the present invention provides a method for nonlinear distortion compensation of satellite signals based on neural networks.

[0005] Technical Solution: This invention provides a method for compensating for nonlinear distortion of satellite signals based on neural networks, specifically including the following steps:

[0006] Step 1: Establish a nonlinear distortion model A1 for satellite signals;

[0007] Step 2: For the scaling factor in model A1, establish a distortion compensation model A2;

[0008] Step 3: Establish a backpropagation neural network (BPNN) and use the particle swarm optimization algorithm to calculate the initial weights and biases of the BPNN network;

[0009] Step 4: Solve model A2 to obtain the optimal scaling factor corresponding to the nonlinear distortion signal, and iteratively train the BPNN network;

[0010] Step 5: Use the scaling factor output by the trained BPNN network to perform nonlinear distortion compensation of the satellite signal based on the scaling factor.

[0011] Furthermore, the expression for the satellite signal nonlinear distortion model A1 is as follows:

[0012] z(t)=ku(t)

[0013] Where k represents the scaling factor of the transmitted signal u(t), and t represents time. Where j is the imaginary unit, and r(t) represents the amplitude of the signal u(t). The expressions for the phase of signal u(t), A(r(t)), Φ(r(t)), and u(t) are shown below:

[0014]

[0015] Where, α α β α , and All are parameters, x n Let T be the signal from the transmitter sampled at the nth time, p(t) be the impulse response function of the pulse shaping filter, and T be the signal from the transmitter sampled at the nth time. s This represents the sampling period, where N is the total number of transmitted signals sampled.

[0016] Furthermore, the expression for the distortion compensation model A2 is:

[0017]

[0018] Where K = [k0, k1, ..., k n ,…,k N-1 ] T Let k be a vector of complex scaling factors for nonlinear distortion. n For x n The corresponding nonlinear distortion complex scaling factor; x is a complex time-domain signal sequence, x = [x0, x1, ..., x...]. n ,…,x N-1 ] T m = [m0, m1, ..., m n ,…,m N-1 ] T m represents the signal vector after nonlinear distortion compensation. n Let m be the signal after nonlinear distortion compensation for the nth nonlinear distortion. n Expressed as:

[0019]

[0020] Among them, y n This is the nth signal sampled by the receiver.

[0021] Furthermore, step five specifically involves: inputting the signal from the receiving end into the trained neural network; the trained neural network outputs the corresponding nonlinear distortion complex scaling factor; and then, according to the formula... Nonlinear distortion compensation is performed on the signal received at the receiving end.

[0022] Furthermore, the calculation of the initial weights and biases of the BPNN network using the particle swarm optimization algorithm in step three is specifically as follows:

[0023] Step 3.1: Initialize the initial positions P = [p1, p2, ..., p] of the population consisting of l particles. i …,p l The position of the i-th particle is represented as p. i =(p i1 ,p i2 ,…,p iD ), p iD This represents the D-th dimension of the position of the i-th particle, where D = D w +D d D w D represents the number of weights in the neural network. w Set the number of biases for the neural network; initialize the global optimum and the individual optimum;

[0024] Step 3.2: Update the particle velocity at the current iteration;

[0025] Step 3.3: Update the particle position based on the updated particle velocity;

[0026] Step 3.4: Calculate the fitness of the particle in the current iteration and update the individual optimal value and the global optimal value; if the number of iterations reaches the maximum number of iterations, output the optimal particle position information to obtain the initial weights and biases of the neural network; otherwise, proceed to the next iteration; go to step 3.2.

[0027] Furthermore, the step 3.2 of updating the particle velocity specifically involves: first calculating the particle velocity v at the current tp-th iteration. id,tp If v id,tp Greater than the preset maximum speed v max Then v id,tp =v max If v id,tp Less than the preset minimum speed w min Then v id,tp =v min , i represents the i-th particle, d represents the dimension, d = 1, 2, ..., D.

[0028] Beneficial effects: This invention is based on data-driven approaches, requiring no complex mathematical derivations or assumptions, and reduces the impact of nonlinear distortion on the system with low complexity. Detailed Implementation

[0029] The illustrative embodiments and descriptions of the present invention are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0030] To better illustrate the method of the present invention, more detailed examples are provided below:

[0031] In a satellite return link, the nonlinear distortion caused by the traveling wave tube amplifier (TWTA) on the satellite is considered, and the nonlinear distortion is compensated at the receiver.

[0032] A method for compensating for nonlinear distortion of satellite signals based on neural networks includes the following steps:

[0033] Step A: Establish a nonlinear distortion model A1 for the satellite signal. Model A1 is as follows:

[0034] z(t)=ku(t)

[0035] in,

[0036]

[0037] Where r(t) represents the amplitude of the signal u(t), The expressions for A(r(t)) and Φ(r(t)) representing the phase of signal u(t) are as follows:

[0038]

[0039] Where, α α β α , and These are the Saleh model parameters, which are set to 2.1587, 1.1517, 4.0033, and 9.104 in this embodiment.

[0040] u(t) is the transmitted signal, expressed as:

[0041]

[0042] Where, x n Let N be the nth sampled symbol (i.e., the nth sampled signal, which is a sample of the transmitted signal), where N is the symbol sequence length and T is the symbol period. s p(t) is the impulse response function of the pulse shaping filter; k is a complex scaling factor related to the amplitude of the input signal; in this embodiment, the complex scaling factor for nonlinear distortion corresponding to the i-th symbol is k. i .

[0043] Step B: Establish the distortion compensation model A2. Model A2 is as follows:

[0044]

[0045] Where ||·|| is the 2-norm, which is the square root of the sum of the squares of the absolute values ​​of the vector elements. To make ||mx|| 2 The smallest value for K; Expanding the above model in detail, we obtain:

[0046]

[0047] Where K = [k0, k1, ..., k i ,…,k N-1 ] T Let k be the vector of nonlinear distortion complex scaling factors corresponding to the symbol sequence. n Let x be the complex scaling factor for the nonlinear distortion corresponding to the nth symbol, and let x be a complex time-domain signal sequence, x = [x0, x1, ..., xn]. n ,…,x N-1 ] T m = [m0, m1, ..., m n ,…,m N-1 ] T m represents the sign vector after nonlinear distortion compensation. n Let m be the symbol after compensation for the nth nonlinear distortion. n Expressed as:

[0048]

[0049] Among them, y n This refers to the nth symbol sampled at the receiver (i.e., the nth signal sampled at the receiver).

[0050] Step C: Establish a backpropagation neural network (BPNN). The input layer of the network has two input nodes, corresponding to the real and imaginary parts of the complex received signal. The output layer has two output nodes, corresponding to the real and imaginary parts of the complex scaling factor k. The initial weights and biases of the neural network are optimized using the particle swarm optimization algorithm. The specific process is as follows:

[0051] Step C-1: Perform initialization in a D-dimensional search space, where D = D w +D d D w D represents the number of weights in the neural network. w The number of biases in the neural network; the initial positions of the population consisting of l particles, P = [p1, p2, ..., p...]. l The position of the i-th particle is represented as p. i =(p i1 ,p i2 ,…,p iD ), p iDInitialize the global optimum G for the D-th dimension of the position of the i-th particle. best =inf, where inf is infinity, initializes the individual's optimal value G. id =inf.

[0052] Step C-2: Iteratively update the particle position and velocity. The particle velocity update formula is:

[0053] v i,tp =w p ×v i,tp-1 +c1×ψ1×(G i,tp-1 -p i,tp-1 )+c2×ψ2×(G best,tp -p i,tp-1 )

[0054] Where i represents the i-th particle, tp represents the tp-th iteration, and w p c1 and c2 are inertia weight coefficients, c1 and c2 are population learning factors, ψ1 and ψ2 are random numbers on [0,1], and w p The expression is:

[0055]

[0056] T max w represents the maximum number of iterations. min and w max These are the minimum inertia factor and the maximum inertia factor, respectively.

[0057] Speed ​​update: If v id,tp >v max Then v id,tp =v max If v id,tp <v min Then v id,tp =v min Where d represents dimension, d = 1, 2, ..., D, v id,tp For vector v i,tp The d-th element.

[0058] The particle position update formula is:

[0059] p i,tp =p i,tp-1 +v i,tp ,

[0060] Calculate the fitness of the current particle and update the individual optimal value and the global optimal value; when the number of iterations reaches the maximum number of iterations, output the optimal particle position information to obtain the initial weights and biases of the neural network; otherwise, proceed to the next iteration.

[0061] Step C-3: Iteratively train the BPNN network; calculate the mean square error E. t Update the network weights and biases; when E t The iteration ends when the error is less than the target error or the maximum number of iterations is reached.

[0062] Step D: Detect satellite signal distortion and perform satellite signal distortion compensation. The specific steps are as follows:

[0063] Step D-1: Using the trained network, input the nonlinear distortion symbol y of the receiver to be detected. n The network output is the corresponding nonlinear distortion complex scaling factor;

[0064] Step D-2: According to Compensation is performed for each received symbol.

[0065] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

Claims

1. A method for compensating for nonlinear distortion of satellite signals based on neural networks, characterized in that, Specifically, the steps include the following: Step 1: Establish a nonlinear distortion model A1 for satellite signals; Step 2: For the scaling factor in model A1, establish a distortion compensation model A2; Step 3: Establish a backpropagation neural network (BPNN) and use the particle swarm optimization algorithm to calculate the initial weights and biases of the BPNN network; Step 4: Solve model A2 to obtain the optimal scaling factor corresponding to the nonlinear distortion signal, and iteratively train the BPNN network; Step 5: Use the scale factor output by the trained BPNN network to perform nonlinear distortion compensation of the satellite signal based on the scale factor; The expression for the nonlinear distortion model A1 of the satellite signal is as follows: ； Where k represents the signal at the transmitting end. The scaling factor, where t represents time. Where j is the imaginary unit, Indicates signal The range, Indicates signal phase, , and The expression is as follows: ； ； ； in, , , and All are parameters. For the first The signal from the transmitting end of the second sample, Let be the impulse response function of the pulse shaping filter. This represents the sampling period, where N is the total number of transmitted signals sampled. The expression for the distortion compensation model A2 is: ； in, Let be a vector of complex scaling factors for nonlinear distortion, where for The corresponding nonlinear distortion complex scaling factor; It is a complex time-domain signal sequence. , This represents the signal vector after nonlinear distortion compensation. For the nth nonlinear distortion compensated signal, Expressed as: ； in, This is the nth signal sampled by the receiver.

2. The method for compensating for nonlinear distortion of satellite signals based on neural networks according to claim 1, characterized in that, Step five specifically involves: inputting the signal from the receiving end into the trained neural network; the trained neural network outputs the corresponding nonlinear distortion complex scaling factor; and then, according to the formula... Nonlinear distortion compensation is performed on the signal received at the receiving end.

3. The satellite signal nonlinear distortion compensation method based on neural networks according to claim 1, characterized in that, The process of calculating the initial weights and biases of the BPNN network using the particle swarm optimization algorithm in step three is as follows: Step 3.1: Initialize by The initial position of a population of particles , No. The position of each particle is represented as , The D-th dimension represents the position of the i-th particle. , This represents the number of weights in the neural network. Set the number of biases for the neural network; initialize the global optimum and the individual optimum; Step 3.2: Update the particle velocity at the current iteration; Step 3.3: Update the particle position based on the updated particle velocity; Step 3.4: Calculate the fitness of the particle in the current iteration and update the individual optimal value and the global optimal value; if the number of iterations reaches the maximum number of iterations, output the optimal particle position information to obtain the initial weights and biases of the neural network; otherwise, proceed to the next iteration; go to step 3.

2.

4. The satellite signal nonlinear distortion compensation method based on neural networks according to claim 3, characterized in that, The specific steps for updating the particle velocity in step 3.2 are as follows: First, calculate the particle velocity at the current tp-th iteration. ,like Greater than the preset maximum speed ,but ,like Less than the preset minimum speed ,but , i represents the i-th particle, d represents the dimension, d=1,2,…,D.

Images