CN111200470B - A high-order modulation signal transmission control method suitable for nonlinear interference - Google Patents

Abstract

The invention relates to a method suitable for a computerThe transmission control method of the linear interference high-order modulation signal realizes the demodulation function through the neural network to reduce the nonlinear interference caused by the power amplifier to the signal, thereby recovering the signal of the transmitting end, wherein the cost function of the neural network adopts a cross entropy function; the training method adopts a back propagation algorithm; the real part and the imaginary part of the signal after passing through the matched filter and the corresponding power back-off coefficient are used as the input of the neural network, and the output result of the neural network representsQAMA probability value of 1 for each bit in the signal; and the activation functions of the hidden layer and the output layer of the neural network are respectively adoptedtanhFunction sumsigmoidA function; compared with the demodulation algorithm of the high-order modulation signal in the existing nonlinear scene, the whole design scheme not only improves the performance, but also reduces the complexity of the algorithm.

Description

A high-order modulation signal transmission control method suitable for nonlinear interference

technical field

The invention relates to a high-order modulation signal transmission control method suitable for nonlinear interference, and belongs to the technical field of mobile communication.

Background technique

With the continuous increase in the number of mobile users, people's demand for communication technology has become higher and higher. In the past 40 years or so, the mobile communication system has developed from the first-generation mobile communication system (1G) in the 1970s to the fifth-generation mobile communication system (5G), which is currently widely concerned. At present, the industry's research on 5G wireless access technology mainly focuses on three application scenarios of 5G: enhanced mobile broadband (eMBB), massive model communication (mMTC), and ultra-reliable and low-latency communication (URLLC). Mobile broadband is a comprehensive enhancement of the current 4G data transmission rate, delay, and user capacity. Mass communication is for the Internet of Things, and ultra-reliable and low-latency communication is for the Internet of Vehicles and other fields. In order to realize the above scenarios, the bandwidth and transmission rate required by 5G will be much greater than that of 4G. Compared with 4G, the transmission rate of 5G will reach 1Gb/s. To achieve such a high transmission rate, the most direct way is to use Higher order QAM modulation. The advantage of high-order QAM modulation is that it can greatly improve the transmission rate of the communication system and increase the frequency band utilization. However, the high-order QAM modulation signal is prone to nonlinear interference during the transmission process, especially the nonlinear interference caused by the power amplifier to the signal. As a result, the signal will be distorted.

Around this technology, a variety of technologies have been proposed at home and abroad to overcome the nonlinear interference caused by the power amplifier to the signal. These technologies can usually be divided into pre-distortion compensation technology and post-distortion compensation technology. A common pre-distortion compensation technology is digital pre-distortion (DPD) technology. The basic idea of this technology is to correct the baseband signal input to the power amplifier according to the nonlinear characteristics of the power amplifier, so as to eliminate the nonlinear effect on the signal as much as possible. interference. Since different power amplifiers have different nonlinear characteristics, it is inefficient to perform linearization correction on each power amplifier in practical engineering applications. Different from the pre-distortion compensation technology, the post-distortion compensation technology uses a specific demodulation algorithm at the receiving end of the communication system to eliminate the non-linear interference to the signal as much as possible. The current post-distortion compensation technology is highly complex.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide a high-order modulation signal transmission control method suitable for nonlinear interference, which can reduce the nonlinear interference caused by the power amplifier to the signal, thereby improving the transmission capability of the communication system

In order to solve the above technical problems, the present invention adopts the following technical solutions: The present invention designs a high-order modulation signal transmission control method suitable for nonlinear interference, which is used for binary bit streams X{x ₁ , x ₂ ,... ,x _N }, to realize the transmission control from the transmitter to the receiver; including the following steps:

获得分组数K，然后进入步骤B；其中，

表示向上取整，N表示二进制比特流X中的比特数；Step A. According to the modulation order M, the transmitter obtains the number of bits m in a single QAM signal according to m=log ₂ M, and presses

Obtain the number of groups K, and then enter step B; wherein,

Represents rounded up, N represents the number of bits in the binary bit stream X;

Step B. The transmitting end is m with the number of bits in a single QAM signal, and is grouped sequentially for each bit in the binary bit stream X, and obtaining K modulation orders is the QAM signal of M, forming the M-QAM signal S{s ₁ ,...,s _k ,...,s _K }, s _k represents the k-th QAM signal in the M-QAM signal S; among them, if the number of bits in the last QAM signal in the sequence is less than m, then at the end Fill 0 to satisfy the number of bits m, and then enter step C;

Step C. The transmitter applies a multiplier to the M-QAM signal S{s ₁ ,...,s _k ,...,s _K }, and processes and updates each QAM signal through v _k =G×s _k , obtain the signal V{v ₁ ,...,v _k ,...,v _K }, where v _k represents the kth QAM signal in the signal V, and the transmitter sends this signal to the receiver, where, G represents the preset power back-off coefficient, and then enter step D;

Step D. The receiving end receives the signal from the transmitting end, applies the real part, imaginary part, and power backoff coefficient G of each QAM signal in the received signal, as the input of the trained neural network, and each QAM signal is processed by the neural network , respectively obtain the probability that each bit in each QAM signal is 1, and then enter step E;

Step E. The processing result of the neural network at the receiving end, respectively, for each bit in each QAM signal, determine whether the probability that the bit is 1 is not less than the preset probability threshold, and if yes, then determine that the value of the bit is equal to 1; otherwise, determine that the bit is is 0; after completing the above judgment on each bit in each QAM signal, according to the order of each QAM signal and the order of each bit in the QAM signal, sort the values of all the bits, that is, obtain the value received by the receiving end. The resulting binary bit stream; wherein, if there is a 0-fill operation for the last QAM signal in the sequence in step B, delete the last corresponding bit in the resulting binary bit stream.

As a preferred technical solution of the present invention: in the step B, the transmitting end applies the constellation mapping method to sequentially group each bit in the binary bit stream X with the number of bits m in a single QAM signal.

As a preferred technical solution of the present invention: it also includes step CD-1 as follows, after obtaining the signal V{v ₁ ,...,v _k ,...,v _K } in the step C, enter the step CD- 1;

Step CD-1. The transmitting end applies a square root raised cosine filter to the signals V{v ₁ ,...,v _k ,...,v _K }, and applies u _k =Ι(v _k ) to each QAM signal, respectively. Perform processing and update to obtain the signal U{u ₁ ,...,u _k ,...,u _K }, where u _k represents the kth QAM signal in the signal U, and the transmitter sends this signal to the receiver , where Ι(·) represents the transformation function of the square root raised cosine filter to the input signal, and then enters step D.

As a preferred technical solution of the present invention: it also includes step CD-2 as follows, after obtaining the signal U{u ₁ ,...,u _k ,...,u _K } in the step CD-1, enter the step CD-2;

Step CD-2. The transmitter applies a power amplifier to the signals U{u ₁ ,...,u _k ,...,u _K }, and processes and updates each QAM signal through t _k =PA(u _k ). , obtain the signal T{t ₁ ,...,t _k ,...,t _K }, where t _k represents the kth QAM signal in the signal T, and the transmitter sends this signal to the receiver, where, PA(·) represents the transformation function of the power amplifier to the input signal, and then proceeds to step D.

As a preferred technical solution of the present invention, the step D includes the following steps:

Step D1. The receiving end receives the signal from the transmitting end, and applies a matched filter to the obtained signal R{r ₁ ,..., _rk ,...,r _K }, through y _k =1( _rk ) Each QAM signal is processed and updated to obtain the signal Y{y ₁ ,...,y _k ,...,y _K }, and then enter step D2; wherein, the transformation function in the matched filter and the square root rise The transformation functions in the cosine filter are the same, rk denotes the kth QAM signal in the signal R, and _yk denotes the _kth QAM signal in the signal Y.

Step D2. For the signal Y{y ₁ ,...,y _k ,...,y _K }, the receiving end applies the real part, imaginary part, and power backoff coefficient G of each QAM signal y _k in the signal, as For the input of the trained neural network, each QAM signal is processed by the neural network to obtain the probability that each bit in each QAM signal _yk is 1, and then step E is entered.

As a preferred technical solution of the present invention: the neural network is a fully connected neural network with a number of hidden layers L, wherein the number of neurons in each hidden layer is Z; in the application of the neural network, the first hidden layer The relationship between the output a ^l of the neural network and the output a ^l-1 of the l-1 hidden layer of the neural network is:

表示神经网络中第l-1隐藏层中第i个神经元到第l隐藏层中第j个神经元的权重值；

表示神经网络中第l隐藏层中第i个神经元的偏置，且神经网络中相同隐藏层中各个神经元的偏置彼此相同；f(·)表示神经网络中隐藏层的激活函数。In the formula,

Represents the weight value of the ith neuron in the l-1th hidden layer in the neural network to the jth neuron in the lth hidden layer;

represents the bias of the ith neuron in the lth hidden layer in the neural network, and the biases of each neuron in the same hidden layer in the neural network are the same as each other; f( ) represents the activation function of the hidden layer in the neural network.

As a preferred technical solution of the present invention, the neural network training stage includes the following steps:

Step 1. For the sample binary bit stream, through step A to step D, obtain the training sample Ψ, including the signal Y'{y' ₁ ,...,y' _k ,..., the receiving end passes through the matched filter y′ _K } and the power backoff coefficient G, thus determine that the number of nodes in the input layer of the neural network is equal to 3, which are used to receive the real part, imaginary part, and power backoff coefficient G of the QAM signal y′ _k respectively, and then enter the step II;

和

且神经网络中相同隐藏层中各个神经元的偏置彼此相同，分别针对信号Y'中的各QAM信号y'_k，神经网络的输入层分别接收QAM信号y'_k的实部、虚部、以及功率回退系数G，然后根据神经网络输出层的节点数等于m，由神经网络输出层获得该QAM信号y'_k中各比特为1的概率；如此通过步骤A至步骤E实现对神经网络的训练，确定神经网络中隐藏层数目为L，每个隐藏层的神经元的个数为Z，以及优化

和

Step II. Apply the xavier initialization method to initialize as described

and

And the biases of each neuron in the same hidden layer in the neural network are the same as each other, for each QAM signal _y'k in the signal _Y ', the input layer of the neural network receives the real part, imaginary part, and the power backoff coefficient G, then according to the number of nodes in the output layer of the neural network equal to m, the probability that each bit in the QAM signal y'k is ₁ is obtained from the output layer of the neural network; training, determine the number of hidden layers in the neural network is L, the number of neurons in each hidden layer is Z, and optimize

and

和

As a preferred technical solution of the present invention: a cost function in the form of cross entropy is used in the neural network, and a backpropagation algorithm is used to iteratively optimize the cost function in the neural network.

and

As a preferred technical solution of the present invention, the activation function of the hidden layer in the neural network is the tanh function, and the activation function of the output layer is the sigmoid function.

The method for controlling the transmission of a high-order modulated signal that is suitable for nonlinear interference according to the present invention adopts the above technical solution and has the following technical effects compared with the prior art:

The present invention is designed to be suitable for the transmission control method of high-order modulated signals subject to nonlinear interference. The demodulation function is realized through the neural network, so as to reduce the nonlinear interference caused by the power amplifier to the signal, so as to restore the signal at the transmitting end. The cost function of the network adopts the cross-entropy function; the training method adopts the back-propagation algorithm; the real part, imaginary part and the corresponding power back-off coefficient of the signal after passing through the matched filter are used as the input of the neural network, and the output of the neural network represents the The probability value of each bit in the QAM signal is 1; and the activation function of the hidden layer and the output layer of the neural network adopts the tanh function and the sigmoid function respectively; the whole design scheme is different from the demodulation of high-order modulated signals in existing nonlinear scenarios. Compared with the algorithm, not only the performance is improved, but also the complexity of the algorithm is reduced.

Description of drawings

Fig. 1 is a system schematic diagram of the present invention's design for a control method for high-order modulated signal transmission subject to nonlinear interference;

Fig. 2 is the frame schematic diagram of the neural network in the present invention's design;

Fig. 3 is the BER simulation graph of the first embodiment of the present invention's design and application;

FIG. 4 is a BER simulation curve diagram of the second embodiment of the design and application of the present invention.

Detailed ways

The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

The invention designs a high-order modulation signal transmission control method suitable for nonlinear interference, which is used to realize the transmission from the transmitter to the receiver for the binary bit stream X{x ₁ ,x ₂ ,...,x _N } Control; as shown in Figure 1, including the following steps A to E.

获得分组数K，然后进入步骤B；其中，

表示向上取整，N表示二进制比特流X中的比特数。Step A. According to the modulation order M, the transmitter obtains the number of bits m in a single QAM signal according to m=log ₂ M, and presses

Obtain the number of groups K, and then enter step B; wherein,

Represents rounded up, and N represents the number of bits in the binary bit stream X.

Step B. The transmitting end is m with the number of bits in a single QAM signal, applies the constellation mapping method, carries out sequential grouping for each bit in the binary bit stream X, obtains K modulation orders and is the QAM signal of M, constitutes M-QAM The signal S{s ₁ ,...,s _k ,...,s _K }, s _k represents the kth QAM signal in the M-QAM signal S; wherein, if the number of bits in the last QAM signal in the sequence is less than m , then add 0 at the end to satisfy the number of bits m, and then go to step C.

Step C. The transmitter applies a multiplier to the M-QAM signal S{s ₁ ,...,s _k ,...,s _K }, and processes and updates each QAM signal through v _k =G×s _k , obtain the signal V{v ₁ ,...,v _k ,...,v _K }, v _k represents the kth QAM signal in the signal V, and then enter step CD-1; wherein, G represents the preset power return Regression factor.

Step CD-1. The transmitting end applies a square root raised cosine filter to the signals V{v ₁ ,...,v _k ,...,v _K }, and applies u _k =Ι(v _k ) to each QAM signal, respectively. Perform processing update to obtain signals U{u ₁ ,...,u _k ,...,u _K }, where u _k represents the k-th QAM signal in the signal U, and then enter step CD-2; wherein, Ι(· ) represents the transform function of the square root raised cosine filter on the input signal.

Step CD-2. The transmitter applies a power amplifier to the signals U{u ₁ ,...,u _k ,...,u _K }, and processes and updates each QAM signal through t _k =PA(u _k ). , obtain the signal T{t ₁ ,...,t _k ,...,t _K }, t _k represents the kth QAM signal in the signal T, and then enter step D1; where PA( ) represents the power amplifier pair The transform function of the input signal.

Step D1. The receiving end receives the signal from the transmitting end, and applies a matched filter to the obtained signal R{r ₁ ,..., _rk ,...,r _K }, through y _k =1( _rk ) Each QAM signal is processed and updated to obtain the signal Y{y ₁ ,...,y _k ,...,y _K }, and then enter step D2; wherein, the transformation function in the matched filter and the square root rise The transformation functions in the cosine filter are the same, rk denotes the kth QAM signal in the signal R, and _yk denotes the _kth QAM signal in the signal Y.

Step D2. As shown in FIG. 2, the receiving end applies the real part, imaginary part, and power of each QAM signal y _k in the signal to the signal Y{y ₁ ,...,y _k ,...,y _K } The back-off coefficient G is used as the input of the trained neural network. The neural network processes each QAM signal to obtain the probability that each bit in each QAM signal _yk is 1, and then goes to step E.

In the specific application of the neural network, as shown in Figure 2, the neural network is a fully connected neural network with a number of hidden layers L, wherein the number of neurons in each hidden layer is Z; in the application of the neural network, the first hidden layer is hidden. The relationship between the output a l of the layer and the output a ^{l -1} of the l-1th hidden layer of the neural network is:

表示神经网络中第l-1隐藏层中第i个神经元到第l隐藏层中第j个神经元的权重值；

表示神经网络中第l隐藏层中第i个神经元的偏置，且神经网络中相同隐藏层中各个神经元的偏置彼此相同；f(·)表示神经网络中隐藏层的激活函数。In the formula,

Represents the weight value of the ith neuron in the l-1th hidden layer in the neural network to the jth neuron in the lth hidden layer;

represents the bias of the ith neuron in the lth hidden layer in the neural network, and the biases of each neuron in the same hidden layer in the neural network are the same as each other; f( ) represents the activation function of the hidden layer in the neural network.

And for the training of the neural network, the following steps I to II are included.

Step 1. For the sample binary bit stream, by step A to step D, the training sample Ψ is obtained, including the signal Y'{y' ₁ ,...,y' _k ,..., the receiving end passes through the matched filter. y' _K } and the power back-off coefficient G, thus determine that the number of input layer nodes of the neural network is equal to 3, for receiving the real part, imaginary part, and power back-off coefficient G of the QAM signal y' _k respectively, and then enter the step II.

和

且神经网络中相同隐藏层中各个神经元的偏置彼此相同，分别针对信号Y'中的各QAM信号y'_k，神经网络的输入层分别接收QAM信号y'_k的实部、虚部、以及功率回退系数G，然后根据神经网络输出层的节点数等于m，由神经网络输出层获得该QAM信号y'_k中各比特为1的概率；如此通过步骤A至步骤E实现对神经网络的训练，确定神经网络中隐藏层数目为L，每个隐藏层的神经元的个数为Z，并且其中通过反向传播算法迭代优化神经网络中的

和

Step II. Apply the xavier initialization method to initialize as described

and

And the biases of each neuron in the same hidden layer in the neural network are the same as each other, for each QAM signal _y'k in the signal _Y ', the input layer of the neural network receives the real part, imaginary part, and the power backoff coefficient G, then according to the number of nodes in the output layer of the neural network equal to m, the probability that each bit in the QAM signal y'k is ₁ is obtained from the output layer of the neural network; The number of hidden layers in the neural network is determined to be L, the number of neurons in each hidden layer is Z, and the back-propagation algorithm is used to iteratively optimize the neural network.

and

In practical applications, the cost function in the form of cross-entropy is used in the neural network, and the activation function of the hidden layer in the neural network is the tanh function, and the activation function of the output layer is the sigmoid function.

Step E. For the processing result of the neural network, the receiving end judges whether the probability that the bit is 1 is not less than the preset probability threshold for each bit in each QAM signal. In practical applications, the probability threshold is set to 0.5, and in the judgment process. , if yes, it is determined that the value of this bit is equal to 1; otherwise, the value of this bit is determined to be 0; after the above-mentioned determination of each bit in each QAM signal is completed, according to the order of each QAM signal and the order of each bit in the QAM signal, for The values of all the bits are sorted, that is, the resultant binary bit stream received by the receiving end is obtained; wherein, if there is a zero-filling operation for the last QAM signal in the sequence in step B, the last corresponding number of bits in the resultant binary bit stream is deleted. bits.

Apply the control method for high-order modulated signal transmission designed by the present invention to be applied in practice. In the first embodiment, it is necessary to build a communication system simulation platform with nonlinear interference in the transmission process, wherein the communication system simulation The main physical layer parameters of the platform are shown in Table 1 below.

Table 1

As shown in FIG. 1 , applying the neural network demodulation method designed by the present invention and suitable for high-order modulated signals subject to nonlinear interference, the length of the binary bit stream X{x ₁ ,x ₂ ,...,x _N } is 9220, that is, N=9220; the execution steps are as follows.

In step A, the modulation order M is set to 1024, and the number of bits m=10 and the number of groups K=922 in a single QAM signal are obtained.

In step B, the transmitting end applies the constellation mapping method to sequentially group each bit in the binary bit stream X to obtain 922 QAM signals to form 1024-QAM signals S{s ₁ ,...,s _k ,... ,s _K };

In step C, the transmitter applies a multiplier to the 1024-QAM signal S{s ₁ ,...,s _k ,...,s _K }, and processes each QAM signal through v _k =G×s _k Update, obtain the signal V{v ₁ ,...,v _k ,...,v _K }, where the power backoff coefficient G is selected according to different situations, when training the neural network, the power backoff coefficient G needs to be set to 0dB , -2dB, -4dB, -6dB.

In step CD-1, the transmitter applies a square root raised cosine filter to the signals V{v ₁ ,...,v _k ,...,v _K }, and applies u _k =Ι(v _k ) to each QAM signal. Process and update separately to obtain the signal U{u ₁ ,...,u _k ,...,u _K }. In practical applications, the roll-off coefficient of the square root raised cosine filter is 0.25, and the number of taps is 201 , the expression for the square root raised cosine filter is:

Wherein, v _k represents the kth QAM signal in the signal V, uk represents the _{kth QAM signal in the signal U, h k represents} the filter coefficient, C represents the length of the filter, and in this embodiment, the value of C is 201 .

In step CD-2, the transmitting end applies the Rapp model power amplifier to the signals U{u ₁ ,...,u _k ,...,u _K }, specifically through the following t _k =PA(u _k ) to each QAM The signals are processed and updated respectively to obtain the signals T{t ₁ ,...,t _k ,...,t _K }.

Among them, PA(·) represents the transformation function of the power amplifier to the input signal, A _in represents the amplitude of the input signal _{uk, t k represents} the output signal of the input signal _uk after the Rapp model of the power amplifier, v, A ₀ , p The values are 1, 1, and 2, respectively.

In step D1, the receiving end receives the signal from the transmitting end, and applies a matched filter to the obtained signal R{r ₁ ,..., _rk ,...,r _K }, through y _k =1( _rk ) respectively process and update each QAM signal to obtain the signal Y{y ₁ ,...,y _k ,...,y _K }, and then enter step D2; wherein, the transformation function in the matched filter and the square root The transform function in the raised cosine filter is the same.

In step D2, Y{y ₁ , y ₂ ,...,y _K } and the corresponding power backoff coefficient G are used as the input of the neural network, and the number of nodes in the output layer of the neural network is equal to the bits carried on the QAM signal _sk The number m=10, the output result of the neural network represents the probability value P{d _i =1|y _k } that each bit in the QAM signal _sk is 1.

In step E, when the probability value P{d _i =1|y _k } satisfies the condition of P{d _i =1|y _k }≥0.5, the value of the bit d _i is judged to be 1; when the probability value P{ When d _i =1|y _k } satisfies the condition of P{d _i =1|y _k }<0.5, the value of the bit d _i is judged to be 0; after the above judgment of each bit in each QAM signal is completed, According to the order of each QAM signal and the order of each bit in the QAM signal, the values of all the bits are sorted, that is, the resultant binary bit stream received by the receiving end is obtained.

It should be noted in the execution of the above-mentioned steps A to E that when training the neural network, the training samples Ψ are obtained through the steps A to D, including the signals Y'{y ₁ ',..., y' _k ,...,y' _K } and power backoff coefficient G, where G is set to 0dB, -2dB, -4dB, -6dB, and different power backoff coefficient G conditions can be obtained through steps A to E Under the signal Y', take the signal Y' and the corresponding power backoff coefficient G as the training sample Ψ, determine the number of hidden layers of the neural network L = 6, the number of neurons in each hidden layer Z = 64, and then initialize the neural network. The weight and bias of the network, the initialization method is selected xavier initialization, the neural network is trained through steps A to E, the number of training is set to 600,000 times, and the gradient descent method of the neural network backpropagation algorithm selects the Adam algorithm.

Compared with the traditional demodulation algorithm in the existing nonlinear scenario, the superiority of the neural network demodulation method is evaluated from the perspective of the system bit error rate (BER). As shown in Figure 3, the abscissa represents the normalized output power, The performance of the neural network demodulation method is basically the same as that of the existing best traditional demodulation algorithm, but the complexity of the algorithm is reduced. Now compare the complexity of the neural network demodulation method in this embodiment with the traditional demodulation algorithm, As shown in Table 2 below.

Table 2

The control method for high-order modulation signal transmission designed by the present invention and suitable for nonlinear interference is applied in practice. In the second embodiment, it is necessary to build a communication system simulation platform with nonlinear interference in the transmission process. The communication system simulation platform The main physical layer parameters of the platform are shown in Table 3 below.

table 3

As shown in FIG. 1 , applying the neural network demodulation method designed by the present invention and suitable for high-order modulated signals subject to nonlinear interference, the length of the binary bit stream X{x ₁ ,x ₂ ,...,x _N } is 9220, that is, N=9220; the execution steps are as follows.

In step A, the modulation order M is set to 1024, and the number of bits m=10 and the number of groups K=922 in a single QAM signal are obtained.

In step B, the transmitting end applies the constellation mapping method to sequentially group each bit in the binary bit stream X to obtain 922 QAM signals to form 1024-QAM signals S{s ₁ ,...,s _k ,... ,s _K };

In step C, the transmitter applies a multiplier to the 1024-QAM signal S{s ₁ ,...,s _k ,...,s _K }, and processes each QAM signal through v _k =G×s _k Update, obtain the signal V{v ₁ ,...,v _k ,...,v _K }, where the power backoff coefficient G is selected according to different situations, when training the neural network, the power backoff coefficient G needs to be set to 0dB , -3dB, -6dB, -9dB, -12dB, -14dB, -16dB.

In step CD-1, the transmitter applies a square root raised cosine filter to the signals V{v ₁ ,...,v _k ,...,v _K }, and applies u _k =Ι(v _k ) to each QAM signal. Process and update separately to obtain the signal U{u ₁ ,...,u _k ,...,u _K }. In practical applications, the roll-off coefficient of the square root raised cosine filter is 0.25, and the number of taps is 201 , the expression for the square root raised cosine filter is:

Wherein, v _k represents the kth QAM signal in the signal V, uk represents the _{kth QAM signal in the signal U, h k represents} the filter coefficient, C represents the length of the filter, and in this embodiment, the value of C is 201 .

In step CD-2, the transmitting end applies the Saleh model power amplifier to the signals U{u ₁ ,...,u _k ,...,u _K }, specifically through the following t _k =PA(u _k ) to each QAM The signals are processed and updated respectively to obtain the signals T{t ₁ ,...,t _k ,...,t _K }.

Among them, PA(·) represents the transformation function of the power amplifier to the input signal, A _in represents the amplitude of the input signal _{uk, t k represents} the output signal of the input signal _uk after passing through the Saleh model of the power amplifier, g ₀ , A ₀ , The values of α and β are 2, 1, 2, and 1, respectively.

In step D1, the receiving end receives the signal from the transmitting end, and applies a matched filter to the obtained signal R{r ₁ ,..., _rk ,...,r _K }, through y _k =1( _rk ) respectively process and update each QAM signal to obtain the signal Y{y ₁ ,...,y _k ,...,y _K }, and then enter step D2; wherein, the transformation function in the matched filter and the square root The transform function in the raised cosine filter is the same.

In step D2, Y{y ₁ , y ₂ ,...,y _K } and the corresponding power backoff coefficient G are used as the input of the neural network, and the number of nodes in the output layer of the neural network is equal to the bits carried on the QAM signal _sk The number m=10, the output result of the neural network represents the probability value P{d _i =1|y _k } that each bit in the QAM signal _sk is 1.

In step E, when the probability value P{d _i =1|y _k } satisfies the condition of P{d _i =1|y _k }≥0.5, the value of the bit d _i is judged to be 1; when the probability value P{ When d _i =1|y _k } satisfies the condition of P{d _i =1|y _k }<0.5, the value of the bit d _i is judged to be 0; after the above judgment of each bit in each QAM signal is completed, According to the order of each QAM signal and the order of each bit in the QAM signal, the values of all the bits are sorted, that is, the resultant binary bit stream received by the receiving end is obtained.

It should be noted in the execution of the above steps A to E that when training the neural network, the training samples Ψ are obtained through steps A to D, including the signals Y'{ _y'1 ,..., y' _k ,...,y' _K } and the power backoff coefficient G, where G is set to 0dB, -3dB, -6dB, -9dB, -12dB, -14dB, -16dB, through steps A to E The signal Y' under different power back-off coefficients G can be obtained, and the signal Y' and the corresponding power back-off coefficient G are used as training samples Ψ, and the number of hidden layers of the neural network L=6 is determined, and the neural network of each hidden layer is The element number Z=64, then initialize the weight and bias of the neural network, select xavier initialization as the initialization method, train the neural network through steps A to E, and set the number of training to 600,000 times, and the gradient of the back-propagation algorithm of the neural network The descent method selects Adam's algorithm.

Compared with the traditional demodulation algorithm in the existing nonlinear scenario, the superiority of the neural network demodulation method is evaluated from the perspective of the system bit error rate (BER). As shown in Figure 4, the abscissa represents the normalized output power, The performance of the neural network demodulation method is basically the same as that of the existing best traditional demodulation algorithm, but the complexity of the algorithm is reduced.

The above technical solution is designed to be suitable for the transmission control method of high-order modulated signals subject to nonlinear interference. The demodulation function is realized through the neural network, so as to reduce the nonlinear interference caused by the power amplifier to the signal, so as to restore the signal at the transmitting end. The whole design Compared with the existing demodulation algorithms for high-order modulated signals in nonlinear scenarios, the scheme not only improves the performance, but also reduces the complexity of the algorithm.

The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned embodiments, and can also be made within the scope of knowledge possessed by those of ordinary skill in the art without departing from the purpose of the present invention. Various changes.

Claims (9)

1. A high-order modulation signal transmission control method suitable for nonlinear interference, for realizing transmission control from a transmitter to a receiver for a binary bit stream X{x ₁ ,x ₂ ,...,x _N }; It is characterized in that, comprises the following steps: Step A. According to the modulation order M, the transmitter obtains the number of bits m in a single QAM signal according to m=log ₂ M, and presses

Obtain the number of groups K, and then enter step B; wherein,

Represents rounded up, N represents the number of bits in the binary bit stream X; Step B. The transmitting end is m with the number of bits in a single QAM signal, and is grouped sequentially for each bit in the binary bit stream X, and obtaining K modulation orders is the QAM signal of M, forming the M-QAM signal S{s ₁ ,...,s _k ,...,s _K }, s _k represents the k-th QAM signal in the M-QAM signal S; among them, if the number of bits in the last QAM signal in the sequence is less than m, then at the end Fill 0 to satisfy the number of bits m, and then enter step C; Step C. The transmitter applies a multiplier to the M-QAM signal S{s ₁ ,...,s _k ,...,s _K }, and processes and updates each QAM signal through v _k =G×s _k , obtain the signal V{v ₁ ,...,v _k ,...,v _K }, where v _k represents the kth QAM signal in the signal V, and the transmitter sends this signal to the receiver, where, G represents the preset power back-off coefficient, and then enter step D; Step D. The receiving end receives the signal from the transmitting end, applies the real part, imaginary part, and power backoff coefficient G of each QAM signal in the received signal, as the input of the trained neural network, and each QAM signal is processed by the neural network , respectively obtain the probability that each bit in each QAM signal is 1, and then enter step E; Step E. The processing result of the neural network at the receiving end, respectively, for each bit in each QAM signal, determine whether the probability that the bit is 1 is not less than the preset probability threshold, and if yes, then determine that the value of the bit is equal to 1; otherwise, determine that the bit is is 0; after completing the above judgment on each bit in each QAM signal, according to the order of each QAM signal and the order of each bit in the QAM signal, sort the values of all the bits, that is, obtain the value received by the receiving end. The resulting binary bit stream; wherein, if there is a 0-fill operation for the last QAM signal in the sequence in step B, delete the last corresponding bit in the resulting binary bit stream. 2. a kind of high-order modulated signal transmission control method suitable for being subjected to nonlinear interference according to claim 1, is characterized in that: in the described step B, the transmitting end uses the bit number m in a single QAM signal to apply constellation mapping method, sequentially grouping each bit in the binary bit stream X. 3. A high-order modulation signal transmission control method suitable for being subjected to nonlinear interference according to claim 1 or 2, characterized in that: further comprising the steps CD-1 as follows, in the step C, the signal V{v ₁ is obtained ,...,v _k ,...,v _K }, enter step CD-1; Step CD-1. The transmitting end applies a square root raised cosine filter to the signals V{v ₁ ,...,v _k ,...,v _K }, and applies u _k =Ι(v _k ) to each QAM signal, respectively. Perform processing and update to obtain the signal U{u ₁ ,...,u _k ,...,u _K }, where u _k represents the kth QAM signal in the signal U, and the transmitter sends this signal to the receiver , where Ι(·) represents the transformation function of the square root raised cosine filter to the input signal, and then enters step D. 4. A kind of high-order modulation signal transmission control method suitable for being subjected to nonlinear interference according to claim 3, it is characterized in that: also comprise step CD-2 as follows, in described step CD- ₁ , obtain signal U{u1 ,...,u _k ,...,u _K }, enter step CD-2; Step CD-2. The transmitter applies a power amplifier to the signals U{u ₁ ,...,u _k ,...,u _K }, and processes and updates each QAM signal through t _k =PA(u _k ). , obtain the signal T{t ₁ ,...,t _k ,...,t _K }, where t _k represents the kth QAM signal in the signal T, and the transmitter sends this signal to the receiver, where, PA(·) represents the transformation function of the power amplifier to the input signal, and then proceeds to step D. 5. A kind of high-order modulation signal transmission control method suitable for being interfered by nonlinear according to claim 4, is characterized in that, described step D comprises the steps: Step D1. The receiving end receives the signal from the transmitting end, and applies a matched filter to the obtained signal R{r ₁ ,..., _rk ,...,r _K }, through y _k =1( _rk ) Each QAM signal is processed and updated to obtain the signal Y{y ₁ ,...,y _k ,...,y _K }, and then enter step D2; wherein, the transformation function in the matched filter and the square root rise The transformation function in the cosine filter is the same, rk represents the kth QAM signal in the signal R, and _yk represents the _kth QAM signal in the signal Y; Step D2. For the signal Y{y ₁ ,...,y _k ,...,y _K }, the receiving end applies the real part, imaginary part, and power backoff coefficient G of each QAM signal y _k in the signal, as For the input of the trained neural network, each QAM signal is processed by the neural network to obtain the probability that each bit in each QAM signal _yk is 1, and then step E is entered. 6. A high-order modulated signal transmission control method suitable for nonlinear interference according to claim 5, wherein the neural network is a fully connected neural network with L hidden layers, wherein each hidden layer is a fully connected neural network. The number of neurons in the layer is Z; in the neural network application, the relationship between the output a l of the lth hidden layer and the output a ^{l -1} of the l-1th hidden layer of the neural network is:

In the formula,

Represents the weight value of the ith neuron in the l-1th hidden layer in the neural network to the jth neuron in the lth hidden layer;

represents the bias of the ith neuron in the lth hidden layer in the neural network, and the biases of each neuron in the same hidden layer in the neural network are the same as each other; f( ) represents the activation function of the hidden layer in the neural network. 7. A kind of high-order modulation signal transmission control method suitable for being interfered by nonlinear according to claim 6, is characterized in that, described neural network training stage comprises the steps: Step 1. For the sample binary bit stream, through step A to step D, obtain the training sample Ψ, including the signal Y'{y' ₁ ,...,y' _k ,..., the receiving end passes through the matched filter y′ _K } and the power backoff coefficient G, thus determine that the number of nodes in the input layer of the neural network is equal to 3, which are used to receive the real part, the imaginary part, and the power backoff coefficient G of the QAM signal y′ _k respectively, and then enter the step II; Step II. Apply the xavier initialization method to initialize as described

and

And the bias of each neuron in the same hidden layer in the neural network is the same as each other, for each QAM signal _y'k in the signal _Y ', the input layer of the neural network receives the real part, imaginary part, and the power backoff coefficient G, and then according to the number of nodes in the output layer of the neural network equal to m, the probability that each bit in the QAM signal y'k is ₁ is obtained from the output layer of the neural network; training, determine the number of hidden layers in the neural network is L, the number of neurons in each hidden layer is Z, and optimize

and

8. A high-order modulation signal transmission control method suitable for nonlinear interference according to claim 7, characterized in that: a cost function in the form of cross-entropy is adopted in the neural network, and iteratively optimizes through a back-propagation algorithm in the neural network

and

9. A kind of high-order modulation signal transmission control method suitable for being subject to nonlinear interference according to claim 7, is characterized in that: the activation function of hidden layer in described neural network is tanh function, and the activation function of output layer is sigmoid function.

Images