CN115208731B - A signal peak-to-average power ratio PAPR suppression method and device - Google Patents

Abstract

The invention relates to the technical field of wireless communication, in particular to a signal peak-to-average power ratio PAPR suppression method and device. The method includes: inputting a modulated OFDM transmission signal; performing adaptive constellation mapping and demapping on the transmission signal through the DM-Net improved OFDM system to reduce the peak-to-average power ratio, the adaptability The constellation mapping means that according to the PAPR performance, DM‑Net will automatically adjust the position of the signal constellation point. The DM‑Net is a network structure based on deep learning, including two parts: the M‑Net at the sending end and the D‑Net at the receiving end. The M-Net at the sending end changes the constellation point distribution of the transmission signal by adjusting the constellation of the transmission signal in the frequency domain, so that the time-domain signal sent by the sending end of the system has a lower PAPR value, and the D-Net at the receiving end Net is responsible for restoring the signal after passing through the channel to the original frequency domain signal, so as to ensure the overall bit error rate performance of the system. By adopting the present invention, an effective balance between bit error rate performance and PAPR performance is realized and the complexity is low.

Description

A signal peak-to-average power ratio PAPR suppression method and device

technical field

The invention relates to the technical field of wireless communication, in particular to a signal peak-to-average power ratio PAPR suppression method and device.

Background technique

Orthogonal Frequency Division Multiplexing (OFDM) has the characteristics of high spectrum utilization, strong resistance to multipath fading and strong resistance to intersymbol interference, and plays an important role in 5G. Peak-to-average power ratio (Peak to Average Power Ratio, PAPR) is one of the main technical bottlenecks of the OFDM system. Too high a peak-to-average power ratio will make the RF power amplifier work in the nonlinear region, resulting in nonlinear distortion and power of the signal. sharp increase in consumption. The existing peak-to-average power ratio suppression methods for OFDM systems still have problems such as PAPR performance and bit error rate (Bit Error Rate, BER) performance cannot be effectively balanced, high complexity, and low spectrum utilization.

Contents of the invention

The main purpose of the present invention is to provide a signal peak-to-average power ratio PAPR suppression method and device, so as to solve the problems existing in the prior art.

In order to achieve the above object, the present invention provides a signal peak-to-average power ratio PAPR suppression method, including:

Input the modulated Orthogonal Frequency Division Multiplexing technology OFDM transmission signal;

Through the OFDM system improved by DM-Net, adaptive constellation mapping and demapping are performed on the transmission signal to reduce the peak-to-average power ratio. The adaptive constellation mapping refers to that DM-Net will automatically adjust the signal constellation according to the PAPR performance. The position of the point, the DM-Net is a network structure based on deep learning, including two parts: the M-Net at the sending end and the D-Net at the receiving end, where the M-Net at the sending end adjusts the constellation of the transmission signal in the frequency domain , changing the constellation point distribution of the transmission signal, so that the time-domain signal sent by the system sending end has a lower PAPR value, and the D-Net at the receiving end is responsible for restoring the signal after passing through the channel to the original frequency-domain signal sent to achieve The purpose of ensuring the overall bit error rate performance of the system.

Optionally, performing adaptive constellation mapping and demapping on the transmission signal through the DM-Net improved OFDM system to reduce the peak-to-average power ratio specifically includes:

Assuming that the number of subcarriers in the system is N, the transmitted modulation signal is X=[X ₀ ,X ₁ ,...,X _N-2 ,X _N-1 ], and X is expressed as X _K ＝[X _R (K )X _I (K)], where X _R (K) represents the real part of X _K , X _I (K) represents the imaginary part of X _K , 0≤K≤N-1, at the sending end through M-Net pair input Signal X _K is subjected to constellation mapping, and the mapping relationship is T _K =f(X _K ), where f(·) represents the mapping relationship function, T _K represents the signal after constellation mapping, 0≤K≤N-1,

The expression of the mapping relationship function is:

f(X)＝tanh(tanh(conv (tanh(conv(X,W _conv1 )+b _conv1 ),W _conv2 )+b _conv2 )*W _f +b _f )(1)

Where W _conv1 and b _conv1 represent the weight matrix and bias matrix of the first convolutional layer respectively, W _conv2 and b _conv2 represent the weight matrix and bias matrix of the second convolutional layer respectively, W _f and b _f represent The weight matrix and bias matrix of the fully connected layer, the tanh function is the activation function used by the two convolutional layers and the fully connected layer;

Combine the real part and imaginary part of T into a complex vector form, then oversample T, the oversampling multiple is L, and the oversampled signal is T _L , and then get the time domain transmission through the inverse fast Fourier transform IFFT Signal:

Where 0≤n≤LN-1, 0≤K≤LN-1, the PAPR calculation formula of the time domain signal is:

Where max[|x(n)| ² ] represents the maximum power of the signal, E[|x(n)| ² ] represents the mean value of the signal power, 0≤n≤LN-1;

After the transmission signal has undergone M-Net adaptive constellation mapping, the distribution of signal constellation points ensures that the time domain signal sent by the system transmitter has a low PAPR value;

The time-domain transmission signal arrives at the receiving end after passing through the channel, and the signal received by the receiving end is set to be r(n), 0≤n≤LN-1, and the fast Fourier transform FFT is performed to obtain the frequency domain signal, and the frequency domain signal is obtained for the frequency domain Domain signal is down-sampled to obtain R _K , 0≤K≤N-1, and R _K is split, R _K =[R _R (K)R _I (K)], where R _R (K) represents the value of R _K The real part, R _I (K) represents the imaginary part of _RK , and then the D-Net at the receiving end demaps the received signal, and the mapping relationship is where /> Represents the demapping relational function to get the final signal />

The expression of the demapping relational function is:

Among them, Q _conv1 and U _conv1 represent the weight matrix and bias matrix of the first convolution layer respectively, Q _conv2 and U _conv2 represent the weight matrix and bias matrix of the second convolution layer respectively, Q _f and U _f represent The weight matrix and bias matrix of the fully connected layer, the tanh function is the activation function used by the two convolutional layers and the fully connected layer.

Optionally, the method further includes: adjusting the network architecture of the M-Net and D-Net and training related parameters to achieve a balance between the BER performance and the PAPR performance of the system. The specific training process includes:

First of all, regardless of the PAPR performance of the time-domain signal at the sending end, the system loss function only includes the bit error rate, which is the signal X sent by the sending end and the signal recovered by the receiving end co-determined, while the receiver recovers the signal /> It is determined jointly by M-Net and D-Net, that is, the value of the system loss function is related to M-Net and D-Net. During the training process of the OFDM system improved by DM-Net, by adjusting M-Net and D- Net architecture, and use the optimization algorithm to continuously iterate the parameters of the M-Net and D-Net models to reduce the value of the system loss function, that is, through the joint optimization of the M-Net at the sending end and the D-Net at the receiving end, the system is in Gaussian The optimal bit error rate performance is obtained under the condition of white noise channel or Rayleigh fading channel;

Secondly, based on the structure and parameters of the DM-Net trained in the first step, the PAPR item is added to the system loss function, and the M-Net at the sending end and the D-Net at the receiving end are jointly trained again, and at the same time, after an appropriate selection of weight factors , so as to ensure the bit error rate performance and the required PAPR performance of the system at the same time under the condition of Gaussian white noise channel or Rayleigh fading channel.

Optionally, the network architecture and network parameters of the optimal M-Net and D-Det are finally trained. Both M-Net and D-Det include a two-layer convolutional layer and a fully connected layer. The first and The size of the convolution kernel of the second convolutional layer is 3×3, the number of convolution kernels is 5, the number of nodes in the fully connected layer is 512, and the activation functions of the convolutional layer and the fully connected layer are both is the tanh function.

Optionally, during the joint training process of the sending end M-Det and the receiving end D-Net, an optimization algorithm is used to continuously iterate the model parameters to reduce the value of the model loss function, and the optimization algorithm includes but is not limited to the gradient descent method , Adam algorithm, RMSProp algorithm.

The present invention also provides a signal peak-to-average power ratio PAPR suppression device, comprising:

The input unit is used to input the modulated Orthogonal Frequency Division Multiplexing (OFDM) transmission signal;

The processing unit is used to perform adaptive constellation mapping and demapping on the transmission signal through the DM-Net improved OFDM system to reduce the peak-to-average power ratio. The adaptive constellation mapping refers to the PAPR performance according to the DM-Net The position of the signal constellation point will be automatically adjusted. The DM-Net is a network structure based on deep learning, including two parts: the M-Net at the sending end and the D-Net at the receiving end. The adjustment of the frequency domain constellation changes the constellation point distribution of the transmission signal, so that the time domain signal sent by the system sending end has a lower PAPR value, and the D-Net at the receiving end is responsible for restoring the signal after passing through the channel to the original sent one The frequency domain signal achieves the purpose of ensuring the overall bit error rate performance of the system.

Optionally, the processing unit is specifically configured to:

Assuming that the number of subcarriers in the system is N, the transmitted modulation signal is X=[X ₀ ,X ₁ ,...,X _N-2 ,X _N-1 ], and X is expressed as X _K ＝[X _R (K )X _I (K)], where X _R (K) represents the real part of X _K , X _I (K) represents the imaginary part of X _K , 0≤K≤N-1, at the sending end through M-Net pair input Signal X _K is subjected to constellation mapping, and the mapping relationship is T _K =f(X _K ), where f(·) represents the mapping relationship function, T _K represents the signal after constellation mapping, 0≤K≤N-1,

The expression of the mapping relationship function is:

f(X)＝tanh(tanh(conv (tanh(conv(X,W _conv1 )+b _conv1 ),W _conv2 )+b _conv2 )*W _f +b _f )(1)

Where W _conv1 and b _conv1 represent the weight matrix and bias matrix of the first convolutional layer respectively, W _conv2 and b _conv2 represent the weight matrix and bias matrix of the second convolutional layer respectively, W _f and b _f represent The weight matrix and bias matrix of the fully connected layer, the tanh function is the activation function used by the two convolutional layers and the fully connected layer;

Combine the real part and imaginary part of T into a complex vector form, then oversample T, the oversampling multiple is L, and the oversampled signal is T _L , and then get the time domain transmission through the inverse fast Fourier transform IFFT Signal:

Where 0≤n≤LN-1, 0≤K≤LN-1, the PAPR calculation formula of the time domain signal is:

Where max[|x(n)| ² ] represents the maximum power of the signal, E[|x(n)| ² ] represents the mean value of the signal power, 0≤n≤LN-1;

After the transmission signal has undergone M-Net adaptive constellation mapping, the distribution of signal constellation points ensures that the time domain signal sent by the system transmitter has a low PAPR value;

The time-domain transmission signal arrives at the receiving end after passing through the channel, and the signal received by the receiving end is set to be r(n), 0≤n≤LN-1, and the fast Fourier transform FFT is performed to obtain the frequency domain signal, and the frequency domain signal is obtained for the frequency domain Domain signal is down-sampled to obtain R _K , 0≤K≤N-1, and R _K is split, R _K =[R _R (K) R _I (K)], where R _R (K) represents the value of R _K The real part, R _I (K) represents the imaginary part of _RK , and then the D-Net at the receiving end demaps the received signal, and the mapping relationship is where /> Represents the demapping relational function to get the final signal />

The expression of the demapping relational function is:

Among them, Q _conv1 and U _conv1 represent the weight matrix and bias matrix of the first convolution layer respectively, Q _conv2 and U _conv2 represent the weight matrix and bias matrix of the second convolution layer respectively, Q _f and U _f represent The weight matrix and bias matrix of the fully connected layer, the tanh function is the activation function used by the two convolutional layers and the fully connected layer.

Optionally, the device further includes: a training unit, configured to adjust the network architecture of the M-Net and D-Net and train related parameters to achieve both system bit error rate BER performance and PAPR performance,

The training unit is specifically used for:

First of all, regardless of the PAPR performance of the time-domain signal at the sending end, the system loss function only includes the bit error rate, which is the signal X sent by the sending end and the signal recovered by the receiving end co-determined, while the receiver recovers the signal /> It is determined jointly by M-Net and D-Net, that is, the value of the system loss function is related to M-Net and D-Net. During the training process of the OFDM system improved by DM-Net, by adjusting M-Net and D- Net architecture, and use the optimization algorithm to continuously iterate the parameters of the M-Net and D-Net models to reduce the value of the system loss function, that is, through the joint optimization of the M-Net at the sending end and the D-Net at the receiving end, the system is in Gaussian The optimal bit error rate performance is obtained under the condition of white noise channel or Rayleigh fading channel;

Secondly, based on the structure and parameters of the DM-Net trained in the first step, the PAPR item is added to the system loss function, and the M-Net at the sending end and the D-Net at the receiving end are jointly trained again, and at the same time, after an appropriate selection of weight factors , so as to ensure the bit error rate performance and the required PAPR performance of the system at the same time under the condition of Gaussian white noise channel or Rayleigh fading channel.

Optionally, the network architecture and network parameters of the optimal M-Net and D-Det are finally trained. Both M-Net and D-Det include a two-layer convolutional layer and a fully connected layer. The first and The size of the convolution kernel of the second convolutional layer is 3×3, the number of convolution kernels is 5, the number of nodes in the fully connected layer is 512, and the activation functions of the convolutional layer and the fully connected layer are both is the tanh function.

Optionally, the training unit is further configured to use an optimization algorithm to continuously iterate model parameters to reduce the value of the model loss function during the joint training process of the sending end M-Det and the receiving end D-Net, and the optimization Algorithms include but are not limited to gradient descent method, Adam algorithm, RMSProp algorithm.

The beneficial effects of the present invention at least include: achieving an effective balance between bit error rate performance and PAPR performance and having lower complexity.

Other features and advantages of the present invention will be described in detail in the following detailed description.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained based on these drawings without creative effort.

FIG. 1 is a schematic flow diagram of a signal peak-to-average power ratio PAPR suppression method according to an embodiment of the present invention;

Fig. 2 shows the schematic diagram of the DM-Net improved OFDM system of the embodiment of the present invention;

Fig. 3 shows the DM-Net neural network schematic diagram of the embodiment of the present invention;

Fig. 4 shows the PAPR performance schematic diagram of the system of the embodiment of the present invention;

FIG. 5 shows a schematic diagram of a system bit error rate according to an embodiment of the present invention;

6 is a schematic structural diagram of a signal peak-to-average power ratio PAPR suppression device according to an embodiment of the present invention;

Fig. 7 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention.

Detailed ways

Specific embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described here are only used to illustrate and explain the present invention, and are not intended to limit the present invention.

As shown in FIG. 1, a signal peak-to-average power ratio PAPR suppression method according to an embodiment of the present invention is shown, including:

S1. Input the modulated Orthogonal Frequency Division Multiplexing (OFDM) transmission signal;

S2. Perform adaptive constellation mapping and demapping on the transmission signal through the DM-Net improved OFDM system to reduce the peak-to-average power ratio. The adaptive constellation mapping refers to that DM-Net will automatically adjust according to the PAPR performance The position of the signal constellation point, the DM-Net is a network structure based on deep learning, including two parts: the M-Net at the sending end and the D-Net at the receiving end, where the M-Net at the sending end performs the frequency domain constellation analysis on the transmission signal The adjustment of the constellation point distribution of the transmission signal is changed, so that the time domain signal sent by the system sending end has a lower PAPR value, and the D-Net at the receiving end is responsible for restoring the signal after passing through the channel to the frequency domain signal originally sent , to achieve the purpose of ensuring the performance of the overall bit error rate of the system.

Optionally, performing adaptive constellation mapping and demapping on the transmission signal through the DM-Net improved OFDM system to reduce the peak-to-average power ratio specifically includes:

Assuming that the number of subcarriers in the system is N, the transmitted modulation signal is X=[X ₀ ,X ₁ ,...,X _N-2 ,X _N-1 ], and X is expressed as X _K ＝[X _R (K )X _I (K)], where X _R (K) represents the real part of X _K , X _I (K) represents the imaginary part of X _K , 0≤K≤N-1, at the sending end through M-Net pair input Signal X _K is subjected to constellation mapping, and the mapping relationship is T _K =f(X _K ), where f(·) represents the mapping relationship function, T _K represents the signal after constellation mapping, 0≤K≤N-1,

The expression of the mapping relationship function is:

f(X)＝tanh(tanh(conv (tanh(conv(X,W _conv1 )+b _conv1 ),W _conv2 )+b _conv2 )*W _f +b _f )(1)

Where W _conv1 and b _conv1 represent the weight matrix and bias matrix of the first convolutional layer respectively, W _conv2 and b _conv2 represent the weight matrix and bias matrix of the second convolutional layer respectively, W _f and b _f represent The weight matrix and bias matrix of the fully connected layer, the tanh function is the activation function used by the two convolutional layers and the fully connected layer;

Combine the real part and imaginary part of T into a complex vector form, then oversample T, the oversampling multiple is L, and the oversampled signal is T _L , and then get the time domain transmission through the inverse fast Fourier transform IFFT Signal:

Where 0≤n≤LN-1, 0≤K≤LN-1, the PAPR calculation formula of the time domain signal is:

Where max[|x(n)| ² ] represents the maximum power of the signal, E[|x(n)| ² ] represents the mean value of the signal power, 0≤n≤LN-1;

After the transmission signal has undergone M-Net adaptive constellation mapping, the distribution of signal constellation points ensures that the time domain signal sent by the system transmitter has a low PAPR value;

The time-domain transmission signal arrives at the receiving end after passing through the channel, and the signal received by the receiving end is set to be r(n), 0≤n≤LN-1, and the fast Fourier transform FFT is performed to obtain the frequency domain signal, and the frequency domain signal is obtained for the frequency domain Domain signal is down-sampled to obtain R _K , 0≤K≤N-1, and R _K is split, R _K =[R _R (K)R _I (K)], where R _R (K) represents the value of R _K The real part, R _I (K) represents the imaginary part of _RK , and then the D-Net at the receiving end demaps the received signal, and the mapping relationship is where /> Represents the demapping relational function to get the final signal />

The expression of the demapping relational function is:

Among them, Q _conv1 and U _conv1 represent the weight matrix and bias matrix of the first convolution layer respectively, Q _conv2 and U _conv2 represent the weight matrix and bias matrix of the second convolution layer respectively, Q _f and U _f represent The weight matrix and bias matrix of the fully connected layer, the tanh function is the activation function used by the two convolutional layers and the fully connected layer.

Optionally, the method further includes: adjusting the network architecture of the M-Net and D-Net and training related parameters to achieve a balance between the BER performance and the PAPR performance of the system. The specific training process includes:

First of all, regardless of the PAPR performance of the time-domain signal at the sending end, the system loss function only includes the bit error rate, which is the signal X sent by the sending end and the signal recovered by the receiving end co-determined, while the receiver recovers the signal /> It is determined jointly by M-Net and D-Net, that is, the value of the system loss function is related to M-Net and D-Net. During the training process of the OFDM system improved by DM-Net, by adjusting M-Net and D- Net architecture, and use the optimization algorithm to continuously iterate the parameters of the M-Net and D-Net models to reduce the value of the system loss function, that is, through the joint optimization of the M-Net at the sending end and the D-Net at the receiving end, the system is in Gaussian The optimal bit error rate performance is obtained under the condition of white noise channel or Rayleigh fading channel;

Secondly, based on the structure and parameters of the DM-Net trained in the first step, the PAPR item is added to the system loss function, and the M-Net at the sending end and the D-Net at the receiving end are jointly trained again, and at the same time, after an appropriate selection of weight factors , so as to ensure the bit error rate performance and the required PAPR performance of the system at the same time under the condition of Gaussian white noise channel or Rayleigh fading channel.

Optionally, the network architecture and network parameters of the optimal M-Net and D-Det are finally trained. Both M-Net and D-Det include a two-layer convolutional layer and a fully connected layer. The first and The size of the convolution kernel of the second convolutional layer is 3×3, the number of convolution kernels is 5, the number of nodes in the fully connected layer is 512, and the activation functions of the convolutional layer and the fully connected layer are both is the tanh function.

Optionally, during the joint training process of the sending end M-Det and the receiving end D-Net, an optimization algorithm is used to continuously iterate the model parameters to reduce the value of the model loss function, and the optimization algorithm includes but is not limited to the gradient descent method , Adam algorithm, RMSProp algorithm.

The following is a detailed description of a signal peak-to-average power ratio PAPR suppression method in the embodiment of the present invention: the OFDM system provided by the embodiment of the present invention performs adaptive constellation mapping and demapping on the transmission signal to reduce Peak-to-average power ratio, the adaptive constellation mapping means that according to PAPR performance, DM-Net will automatically adjust the position of signal constellation points, and the DM-Net is a network structure based on deep learning, including the M-Net at the sending end and the D-Net at the receiving end, where the M-Net at the sending end changes the constellation point distribution of the transmission signal by adjusting the constellation of the transmission signal in the frequency domain, so that the time domain signal sent by the system sending end has a lower For the purpose of the PAPR value, the D-Net at the receiving end is responsible for restoring the signal after passing through the channel to the original frequency domain signal, so as to ensure the overall bit error rate performance of the system.

The improved OFDM system architecture is shown in Figure 2, and DM-Net is shown in Figure 3. The DM-Net model includes an input layer, a convolutional layer, a fully connected layer and an output layer, but the difference from the traditional convolutional neural network CNN Yes: do not consider pooling, normalization and other processes;

Assuming that the number of subcarriers in the system is N, the transmitted modulation signal is X=[X ₀ ,X ₁ ,...,X _N-2 ,X _N-1 ], and X is expressed as X _K ＝[X _R (K )X _I (K)], where X _R (K) represents the real part of X _K , X _I (K) represents the imaginary part of X _K , 0≤K≤N-1, at the sending end through M-Net pair input Signal X _K is subjected to constellation mapping, and the mapping relationship is T _K =f(X _K ), where f(·) represents the mapping relationship function, T _K represents the signal after constellation mapping, 0≤K≤N-1,

The expression of the mapping relationship function is:

f(X)＝tanh(tanh(conv (tanh(conv(X,W _conv1 )+b _conv1 ),W _conv2 )+b _conv2 )*W _f +b _f )(1)

Where W _conv1 and b _conv1 represent the weight matrix and bias matrix of the first convolutional layer respectively, W _conv2 and b _conv2 represent the weight matrix and bias matrix of the second convolutional layer respectively, W _f and b _f represent The weight matrix and bias matrix of the fully connected layer, the tanh function is the activation function used by the two convolutional layers and the fully connected layer;

The constellation points of the signal T are different from the constellation points evenly distributed by modulation methods such as multi-ary quadrature amplitude modulation. The main purpose of using M-Net for constellation mapping is to adjust the position of the original signal constellation points according to the PAPR performance requirements. After M-Net The distribution of signal constellation points after Net constellation mapping is relatively messy and irregular, which is determined by M-Net according to the required PAPR performance requirements. When the system PAPR performance requirements are different, the signal after M-Net constellation mapping The distribution of constellation points is also different;

Combining the real part and imaginary part of T into a complex vector form, then oversampling T, the oversampling multiple is L, the oversampled signal is T _L , and then undergoes Inverse Fast Fourier Transform (Inverse Fast Fourier Transform ,IFFT) to get the time-domain transmission signal:

Where 0≤n≤LN-1, 0≤K≤LN-1, the PAPR calculation formula of the time domain signal is:

Where max[|x(n)| ² ] represents the maximum power of the signal, E[|x(n)| ² ] represents the mean value of the signal power, 0≤n≤LN-1;

After the transmission signal has undergone M-Net adaptive constellation mapping, the distribution of signal constellation points ensures that the time domain signal sent by the system transmitter has a low PAPR value;

Complementary Cumulative Distribution Function (CCDF) is usually used to quantitatively describe the PAPR suppression performance of different algorithms, which indicates the probability that the actual peak-to-average power ratio of the signal exceeds a given peak-to-average power ratio PAPR0;

The time-domain transmission signal arrives at the receiving end after passing through the channel, and the signal received by the receiving end is set to be r(n), 0≤n≤LN-1, and the fast Fourier transform FFT is performed to obtain the frequency domain signal, and the frequency domain signal is obtained for the frequency domain Domain signal is down-sampled to obtain R _K , 0≤K≤N-1, and R _K is split, R _K =[R _R (K)R _I (K)], where R _R (K) represents the value of R _K The real part, R _I (K) represents the imaginary part of _RK , and then the D-Net at the receiving end demaps the received signal, and the mapping relationship is where /> Represents the demapping relational function to get the final signal />

The expression of the demapping relational function is:

Among them, Q _conv1 and U _conv1 represent the weight matrix and bias matrix of the first convolution layer respectively, Q _conv2 and U _conv2 represent the weight matrix and bias matrix of the second convolution layer respectively, Q _f and U _f represent The weight matrix and bias matrix of the fully connected layer, the tanh function is the activation function used by the two convolutional layers and the fully connected layer.

The suppression method in the embodiment of the present invention also includes: adjusting the network architecture of the M-Net and D-Net and training related parameters to achieve both BER performance and PAPR performance of the system. The specific training process includes:

First of all, regardless of the PAPR performance of the time-domain signal at the sending end, the system loss function only includes the bit error rate, which is the signal X sent by the sending end and the signal recovered by the receiving end co-determined, while the receiver recovers the signal /> It is determined jointly by M-Net and D-Net, that is, the value of the system loss function is related to M-Net and D-Net. During the training process of the OFDM system improved by DM-Net, by adjusting M-Net and D- Net architecture, and use the optimization algorithm to continuously iterate the parameters of the M-Net and D-Net models to reduce the value of the system loss function, that is, through the joint optimization of the M-Net at the sending end and the D-Net at the receiving end, the system is in Gaussian The optimal bit error rate performance is obtained under the condition of white noise channel or Rayleigh fading channel;

Secondly, based on the structure and parameters of the DM-Net trained in the first step, the PAPR item is added to the system loss function, and the M-Net at the sending end and the D-Net at the receiving end are jointly trained again, and at the same time, after an appropriate selection of weight factors , so as to ensure the bit error rate performance and the required PAPR performance of the system at the same time under the condition of Gaussian white noise channel or Rayleigh fading channel.

In the joint training process of M-Net and D-Net, the optimization algorithm can be used to continuously iterate the model parameters to reduce the value of the model loss function. The optimization algorithm includes but is not limited to the gradient descent method, Adam algorithm, RMSProp algorithm, so as to realize The purpose of taking into account both system bit error rate performance and PAPR performance.

The overall loss function of the system can be expressed as:

Loss＝Φ ₁ +ηΦ ₂ (5)

Φ ₂ ＝PAPR(x) (7)

Among them, Φ ₁ represents the system BER, Φ ₂ represents the system PAPR value, and η represents the weight factor of the loss function, which is used to adjust the weight of BER and PAPR in the loss function. When η is large, more attention is paid to the PAPR performance of the system. Instead, pay more attention to the BER performance of the system.

The embodiment of the present invention finally trains to obtain the optimal network architecture and network parameters of M-Net and D-Det. Both M-Net and D-Det include two layers of convolutional layers and one layer of fully connected layers. The first and The size of the convolution kernel of the second convolutional layer is 3×3, the number of convolution kernels is 5, the number of nodes in the fully connected layer is 512, and the activation functions of the convolutional layer and the fully connected layer are both is the tanh function.

Simulation environment and parameter settings:

The overall performance of the system is simulated based on the tensorflow framework of python. The number of subcarriers in the system is set to 64, the modulation method is 4QAM modulation, and four times oversampling. The bit error rate and PAPR performance simulation of the system is performed under the Rayleigh fading channel model. The PAPR performance of the obtained system is shown in Fig. 4, and the bit error rate of the obtained system is shown in Fig. 5.

As shown in FIG. 6, an embodiment of the present invention also provides a signal peak-to-average power ratio PAPR suppression device, including:

The input unit 610 is used to input the modulated Orthogonal Frequency Division Multiplexing (OFDM) transmission signal;

The processing unit 620 is configured to perform adaptive constellation mapping and demapping on the transmission signal through the DM-Net improved OFDM system to reduce the peak-to-average power ratio. The adaptive constellation mapping refers to the performance according to PAPR, DM- Net will automatically adjust the position of the signal constellation point. The DM-Net is a network structure based on deep learning, including two parts: the M-Net at the sending end and the D-Net at the receiving end. The M-Net at the sending end passes through the transmission signal In the frequency domain constellation adjustment, the distribution of the constellation points of the transmission signal is changed, so that the time domain signal sent by the system sending end has a lower PAPR value, and the D-Net at the receiving end is responsible for restoring the signal after passing through the channel to the original transmission frequency domain signal to achieve the purpose of ensuring the overall bit error rate performance of the system.

Optionally, the processing unit is specifically configured to:

Assuming that the number of subcarriers in the system is N, the transmitted modulation signal is X=[X ₀ ,X ₁ ,...,X _N-2 ,X _N-1 ], and X is expressed as X _K ＝[X _R (K )X _I (K)], where X _R (K) represents the real part of X _K , X _I (K) represents the imaginary part of X _K , 0≤K≤N-1, at the sending end through M-Net pair input Signal X _K is subjected to constellation mapping, and the mapping relationship is T _K =f(X _K ), where f(·) represents the mapping relationship function, T _K represents the signal after constellation mapping, 0≤K≤N-1,

The expression of the mapping relationship function is:

f(X)＝tanh(tanh(conv(tanh(conv(X,W _conv1 )+b _conv1 ),W _conv2 )+b _conv2 )*W _f +b _f )(1)

Where W _conv1 and b _conv1 represent the weight matrix and bias matrix of the first convolutional layer respectively, W _conv2 and b _conv2 represent the weight matrix and bias matrix of the second convolutional layer respectively, W _f and b _f represent The weight matrix and bias matrix of the fully connected layer, the tanh function is the activation function used by the two convolutional layers and the fully connected layer;

Combine the real part and imaginary part of T into a complex vector form, then oversample T, the oversampling multiple is L, and the oversampled signal is T _L , and then get the time domain transmission through the inverse fast Fourier transform IFFT Signal:

Where 0≤n≤LN-1, 0≤K≤LN-1, the PAPR calculation formula of the time domain signal is:

Where max[|x(n)| ² ] represents the maximum power of the signal, E[|x(n)| ² ] represents the mean value of the signal power, 0≤n≤LN-1;

After the transmission signal has undergone M-Net adaptive constellation mapping, the distribution of signal constellation points ensures that the time domain signal sent by the system transmitter has a low PAPR value;

The time-domain transmission signal arrives at the receiving end after passing through the channel, and the signal received by the receiving end is set to be r(n), 0≤n≤LN-1, and the fast Fourier transform FFT is performed to obtain the frequency domain signal, and the frequency domain signal is obtained for the frequency domain Domain signal is down-sampled to obtain R _K , 0≤K≤N-1, and R _K is split, R _K =[R _R (K) R _I (K)], where R _R (K) represents the value of R _K The real part, R _I (K) represents the imaginary part of _RK , and then the D-Net at the receiving end demaps the received signal, and the mapping relationship is where /> Represents the demapping relational function to get the final signal />

The expression of the demapping relational function is:

Among them, Q _conv1 and U _conv1 represent the weight matrix and bias matrix of the first convolution layer respectively, Q _conv2 and U _conv2 represent the weight matrix and bias matrix of the second convolution layer respectively, Q _f and U _f represent The weight matrix and bias matrix of the fully connected layer, the tanh function is the activation function used by the two convolutional layers and the fully connected layer.

Optionally, the device further includes: a training unit, configured to adjust the network architecture of the M-Net and D-Net and train related parameters to achieve both system bit error rate BER performance and PAPR performance,

The training unit is specifically used for:

First of all, regardless of the PAPR performance of the time-domain signal at the sending end, the system loss function only includes the bit error rate, which is the signal X sent by the sending end and the signal recovered by the receiving end co-determined, while the receiver recovers the signal /> It is determined jointly by M-Net and D-Net, that is, the value of the system loss function is related to M-Net and D-Net. During the training process of the OFDM system improved by DM-Net, by adjusting M-Net and D- Net architecture, and use the optimization algorithm to continuously iterate the parameters of the M-Net and D-Net models to reduce the value of the system loss function, that is, through the joint optimization of the M-Net at the sending end and the D-Net at the receiving end, the system is in Gaussian The optimal bit error rate performance is obtained under the condition of white noise channel or Rayleigh fading channel;

Secondly, based on the structure and parameters of the DM-Net trained in the first step, the PAPR item is added to the system loss function, and the M-Net at the sending end and the D-Net at the receiving end are jointly trained again, and at the same time, after an appropriate selection of weight factors , so as to ensure the bit error rate performance and the required PAPR performance of the system at the same time under the condition of Gaussian white noise channel or Rayleigh fading channel.

Optionally, the network architecture and network parameters of the optimal M-Net and D-Det are finally trained. Both M-Net and D-Det include a two-layer convolutional layer and a fully connected layer. The first and The size of the convolution kernel of the second convolutional layer is 3×3, the number of convolution kernels is 5, the number of nodes in the fully connected layer is 512, and the activation functions of the convolutional layer and the fully connected layer are both is the tanh function.

Optionally, the training unit is further configured to use an optimization algorithm to continuously iterate model parameters to reduce the value of the model loss function during the joint training process of the sending end M-Det and the receiving end D-Net, and the optimization Algorithms include but are not limited to gradient descent method, Adam algorithm, RMSProp algorithm.

7 is a schematic structural diagram of an electronic device 700 provided by an embodiment of the present invention. The electronic device 700 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 701 and one or more memories 702, wherein at least one instruction is stored in the memory 702, and the at least one instruction is loaded and executed by the processor 701 to realize the above-mentioned signal peak-to-average power ratio PAPR suppression method step.

In an exemplary embodiment, there is also provided a computer-readable storage medium, for example, a memory including instructions, and the above-mentioned instructions can be executed by a processor in the terminal to implement the above-mentioned method for suppressing a signal peak-to-average power ratio (PAPR). For example, the computer readable storage medium may be ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, and the like.

Those of ordinary skill in the art can understand that all or part of the steps for implementing the above embodiments can be completed by hardware, and can also be completed by instructing related hardware through a program. The program can be stored in a computer-readable storage medium. The above-mentioned The storage medium mentioned may be a read-only memory, a magnetic disk or an optical disk, and the like.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within range.

Claims (8)

1. The method for suppressing the PAPR of the signal is characterized by comprising the following steps:

inputting modulated OFDM transmission signals;

the method comprises the steps that an OFDM system improved by DM-Net carries out adaptive constellation mapping and demapping on a transmission signal to reduce the peak-to-average power ratio, wherein the adaptive constellation mapping means that according to PAPR performance, the DM-Net can automatically adjust the position of a signal constellation point, the DM-Net is a network structure based on deep learning and comprises an M-Net part of a transmitting end and a D-Net part of a receiving end, the M-Net part of the transmitting end changes constellation point distribution of the transmission signal through adjustment of the transmission signal on a frequency domain constellation, the purpose that a time domain signal sent by the transmitting end of the system has a lower PAPR value is achieved, and the D-Net part of the receiving end is responsible for recovering the signal after the channel into the original transmitted frequency domain signal, so that the purpose of guaranteeing the overall error rate performance of the system is achieved;

the method for reducing peak-to-average power ratio by adaptive constellation mapping and demapping of the transmission signal by the DM-Net improved OFDM system specifically comprises the following steps:

let the number of sub-carriers of the system be N, the transmitted modulation signal be X= [ X ] ₀ ,X ₁ ,...,X _N-2 ,X _N-1 ]X is represented as X _K ＝[X _R (K) X _I (K)]Wherein X is _R (K) Represents X _K The real part of X _I (K) Represents X _K K is more than or equal to 0 and less than or equal to N-1, and the input signal X is input at the transmitting end through M-Net _K Constellation mapping is carried out, and the mapping relation is T _K ＝f(X _K ) Wherein f (·) represents a mapping function, T _K Represents the signal after constellation mapping, K is more than or equal to 0 and less than or equal to N-1,

the expression of the mapping relation function is:

f(X)＝tanh(tanh(conv(tanh(conv(X,W _conv1 )+b _conv1 ),W _conv2 )+b _conv2 )*W _f +b _f ) (1)

wherein W is _conv1 ，b _conv1 Weight matrix and bias matrix respectively representing the first convolution layer, W _conv2 ，b _conv2 Weight matrix and bias matrix respectively representing the second convolution layer, W _f And b _f Respectively representing a weight matrix and a bias matrix of the full connection layer, wherein the tanh function is an activation function utilized by the two convolution layers and the full connection layer;

combining the real part and the imaginary part in T into a complex vector form, and then oversampling the T, wherein the oversampling multiple is L, and the oversampled signal is T _L And then obtaining a time domain transmission signal through Inverse Fast Fourier Transform (IFFT):

wherein n is more than or equal to 0 and less than or equal to LN-1, K is more than or equal to 0 and less than or equal to LN _- 1, the PAPR calculation formula of the time domain signal is as follows:

wherein the method comprises the steps ofRepresenting the maximum power of the signal, E [ |x (n) | ² ]Represents the average value of signal power, n is more than or equal to 0 and less than or equal to LN-1;

after the transmission signal is subjected to M-Net adaptive constellation mapping, the distribution of signal constellation points ensures that the time domain signal transmitted by a system transmitting end has a lower PAPR value;

the time domain transmission signal reaches the receiving end after passing through the channel, the signal received by the receiving end is set as R (n), n is more than or equal to 0 and less than or equal to LN-1, fast Fourier transform FFT is carried out to obtain a frequency domain signal, and downsampling is carried out on the frequency domain signal to obtain R _K K is more than or equal to 0 and less than or equal to N-1, R is _K Splitting, R _K ＝[R _R (K)R _I (K)]Wherein R is _R (K) Represents R _K The real part of R _I (K) Represents R _K Then the received signal is demapped through the D-Net of the receiving end, the mapping relation is thatWherein->Representing the demapping relation function to obtain the final signal +.>

The expression of the demapping relationship function is:

wherein Q is _conv1 ，U _conv1 Weights respectively representing the first convolution layerRe-matrix and bias matrix, Q _conv2 ，U _conv2 Respectively representing a weight matrix and a bias matrix of a second convolution layer, Q _f And U _f Representing the weight matrix and the bias matrix of the fully connected layer, respectively, the tanh function is the activation function utilized by the two convolutional layers and the fully connected layer.

2. The method of claim 1, wherein the method further comprises: the system bit error rate BER performance and PAPR performance are considered by adjusting the network architecture of the M-Net and the D-Net and training related parameters, and the specific training process comprises the following steps:

firstly, the PAPR performance of the time domain signal of the transmitting end is not considered, the system loss function only comprises the error rate, and the error rate is the signal X transmitted by the transmitting end and the signal recovered by the receiving endCo-determination, and signal recovered at the receiving end +.>The method is jointly determined by M-Net and D-Net, namely, the value of a system loss function is related to the M-Net and the D-Net, in the training process of an OFDM system improved by DM-Net, the system obtains the optimal error rate performance under the condition of a Gaussian white noise channel or a Rayleigh fading channel by adjusting the architecture of the M-Net and the D-Net and continuously iterating parameters of an M-Net model and a D-Net model by utilizing an optimization algorithm so as to reduce the value of the system loss function, namely, the system is subjected to joint optimization of the M-Net of a transmitting end and the D-Net of a receiving end;

secondly, based on the structure and parameters of the DM-Net trained in the first step, PAPR items are added in a system loss function, the M-Net of a transmitting end and the D-Net of a receiving end are jointly trained again, and meanwhile, through proper selection of weight factors, the error rate performance and the required PAPR performance of the system can be ensured under the condition of Gaussian white noise channels or Rayleigh fading channels.

3. The method of claim 2, wherein the network architecture and network parameters of the M-Net and D-Det are optimized by final training, the M-Net and D-Det each comprise two convolution layers and a full connection layer, the convolution kernels of the first and second convolution layers are 3 x 3, the number of convolution kernels is 5, the number of nodes of the full connection layer is 512, and the activation functions of the convolution layers and the full connection layer are tanh functions.

4. The method of claim 2, wherein during the joint training of the transmitting end M-Det and the receiving end D-Net, an optimization algorithm is used to iterate model parameters continuously to reduce the value of the model loss function, wherein the optimization algorithm includes, but is not limited to, a gradient descent method, an Adam algorithm, and an RMSProp algorithm.

5. A signal peak-to-average power ratio, PAPR, suppression apparatus comprising:

an input unit for inputting the modulated OFDM transmission signal;

the processing unit is used for reducing peak-to-average power ratio by carrying out adaptive constellation mapping and demapping on the transmission signal through an OFDM system with improved DM-Net, wherein the adaptive constellation mapping refers to that according to PAPR performance, DM-Net can automatically adjust the position of signal constellation points, the DM-Net is a network structure based on deep learning and comprises an M-Net part of a transmitting end and a D-Net part of a receiving end, the M-Net part of the transmitting end changes constellation point distribution of the transmission signal through adjustment of the transmission signal on a frequency domain constellation, the purpose that a time domain signal transmitted by the transmitting end of the system has a lower PAPR value is achieved, and the D-Net part of the receiving end is responsible for recovering the signal after the signal is transmitted into the original frequency domain signal, so that the purpose of ensuring the overall error rate performance of the system is achieved;

the processing unit is specifically configured to:

let the number of sub-carriers of the system be N, the transmitted modulation signal be X= [ X ] ₀ ,X ₁ ,...,X _N-2 ,X _N-1 ]X is represented as X _K ＝[X _R (K) X _I (K)]Wherein X is _R (K) Represents X _K The real part of X _I (K) Represents X _K K is more than or equal to 0 and less than or equal to N-1, and the input signal X is input at the transmitting end through M-Net _K Constellation mapping is carried out, and the mapping relation is T _K ＝f(X _K ) Wherein f (·) represents a mapping function, T _K Represents the signal after constellation mapping, K is more than or equal to 0 and less than or equal to N-1,

the expression of the mapping relation function is:

f(X)＝tanh(tanh(conv(tanh(conv(X,W _conv1 )+b _conv1 ),W _conv2 )+b _conv2 )*W _f +b _f ) (1)

wherein W is _conv1 ，b _conv1 Weight matrix and bias matrix respectively representing the first convolution layer, W _conv2 ，b _conv2 Weight matrix and bias matrix respectively representing the second convolution layer, W _f And b _f Respectively representing a weight matrix and a bias matrix of the full connection layer, wherein the tanh function is an activation function utilized by the two convolution layers and the full connection layer;

combining the real part and the imaginary part in T into a complex vector form, and then oversampling the T, wherein the oversampling multiple is L, and the oversampled signal is T _L And then obtaining a time domain transmission signal through Inverse Fast Fourier Transform (IFFT):

wherein n is more than or equal to 0 and less than or equal to LN-1, K is more than or equal to 0 and less than or equal to LN-1, and the PAPR calculation formula of the time domain signal is as follows:

wherein the method comprises the steps ofRepresenting the maximum power of the signal, E [ |x (n) ² The I represents the average value of signal power, and n is more than or equal to 0 and less than or equal to LN-1;

after the transmission signal is subjected to M-Net adaptive constellation mapping, the distribution of signal constellation points ensures that the time domain signal transmitted by a system transmitting end has a lower PAPR value;

the time domain transmission signal reaches the receiving end after passing through the channel, the signal received by the receiving end is set as R (n), n is more than or equal to 0 and less than or equal to LN-1, fast Fourier transform FFT is carried out to obtain a frequency domain signal, and downsampling is carried out on the frequency domain signal to obtain R _K K is more than or equal to 0 and less than or equal to N-1, R is _K Splitting, R _K ＝[R _R (K)R _I (K)]Wherein R is _R (K) Represents R _K The real part of R _I (K) Represents R _K Then the received signal is demapped through the D-Net of the receiving end, the mapping relation is thatWherein->Representing the demapping relation function to obtain the final signal +.>

The expression of the demapping relationship function is:

wherein Q is _conv1 ，U _conv1 A weight matrix and a bias matrix, Q, representing the first convolution layer, respectively _conv2 ，U _conv2 Respectively representing a weight matrix and a bias matrix of a second convolution layer, Q _f And U _f Representing the weight matrix and the bias matrix of the fully connected layer, respectively, the tanh function is the activation function utilized by the two convolutional layers and the fully connected layer.

6. The apparatus of claim 5, wherein the apparatus further comprises: a training unit for realizing the combination of the BER performance and the PAPR performance of the system through the adjustment of the network architecture of the M-Net and the D-Net and the training of related parameters,

the training unit is specifically configured to:

firstly, the PAPR performance of the time domain signal of the transmitting end is not considered, the system loss function only comprises the error rate, and the error rate is the signal X transmitted by the transmitting end and the signal recovered by the receiving endCo-determination, and signal recovered at the receiving end +.>The method is jointly determined by M-Net and D-Net, namely, the value of a system loss function is related to the M-Net and the D-Net, in the training process of an OFDM system improved by DM-Net, the system obtains the optimal error rate performance under the condition of a Gaussian white noise channel or a Rayleigh fading channel by adjusting the architecture of the M-Net and the D-Net and continuously iterating parameters of an M-Net model and a D-Net model by utilizing an optimization algorithm so as to reduce the value of the system loss function, namely, the system is subjected to joint optimization of the M-Net of a transmitting end and the D-Net of a receiving end;

secondly, based on the structure and parameters of the DM-Net trained in the first step, PAPR items are added in a system loss function, the M-Net of a transmitting end and the D-Net of a receiving end are jointly trained again, and meanwhile, through proper selection of weight factors, the error rate performance and the required PAPR performance of the system can be ensured under the condition of Gaussian white noise channels or Rayleigh fading channels.

7. The apparatus of claim 6, wherein the network architecture and its network parameters are optimized for final training, the M-Net and the D-Det each comprise two convolution layers and one full connection layer, the convolution kernels of the first and second convolution layers are 3 x 3, the number of convolution kernels is 5, the number of nodes of the full connection layer is 512, and the activation functions of the convolution layers and the full connection layer are tanh functions.

8. The apparatus of claim 6, wherein the training unit is further configured to use an optimization algorithm to iterate model parameters continuously to reduce a value of a model loss function during joint training of the transmitting end M-Det and the receiving end D-Net, the optimization algorithm including, but not limited to, a gradient descent method, an Adam algorithm, an RMSProp algorithm.