CN119094299B

CN119094299B - A PAPR suppression method based on pulse shaping modulation

Info

Publication number: CN119094299B
Application number: CN202410979820.3A
Authority: CN
Inventors: 郑重; 张淅; 费泽松
Original assignee: Beijing Institute of Technology BIT
Current assignee: Beijing Institute of Technology BIT
Priority date: 2024-07-22
Filing date: 2024-07-22
Publication date: 2025-09-12
Anticipated expiration: 2044-07-22
Also published as: CN119094299A

Abstract

The invention provides a PAPR (peak to average power ratio) suppression method based on pulse shaping modulation, which belongs to the field of communication and comprises the steps of building an end-to-end OFDM (orthogonal frequency division multiplexing) communication system model, wherein the communication system model comprises a transmitter, a channel and a receiver, a transmitting end PS module is realized by a transmitting DNN, the transmitting DNN comprises a transmitting input layer and a transmitting output layer which are connected with an S/P (transmitter/receiver) module, the channel is a multipath fading channel, the receiver comprises an equalization module and a demodulation mapping module which are sequentially connected, and step S2 is used for carrying out joint training on a neural network formed by the transmitter, the channel and the receiver by a joint loss function L=alpha ₁L₁+α₂L₂. The invention can reduce PAPR of transmitting end without affecting transmission rate and error rate, and can realize flexible configuration of pulse shaping to flexibly support different communication scenes.

Description

PAPR (peak to average power ratio) suppression method based on pulse shaping modulation

Technical Field

The invention relates to the field of communication, in particular to a PAPR suppression method based on pulse shaping modulation.

Background

Orthogonal frequency division multiplexing (Orthogonal Frequency Division Multiplexing, OFDM) technology plays an important role in the fifth generation mobile communication system due to its robustness in multipath transmission and relatively low design complexity of the transceiver end. However, as a multi-carrier modulation technique, the OFDM technique, which is a multi-carrier modulation technique, superimposes a plurality of orthogonal sub-carriers, causes a large dynamic fluctuation in the envelope of the signal transmitted by the base station, and thus causes a high peak-to-average power ratio of the signal. The peak-to-average ratio signal typically needs to be transmitted through a linear power amplifier, which is however expensive and requires a higher process, which results in higher hardware costs. If the peak-to-average ratio signal passes through the nonlinear power amplifier, serious nonlinear distortion is generated, so that the signal distortion is caused, and the system performance is seriously deteriorated. The reserved subcarrier Technology (TR) proposed by Tellado can be used for reducing the peak-to-average ratio (PAPR) of each antenna, but the PAPR inhibition effect is not obvious and the data transmission rate is obviously reduced, and the PAPR inhibition capability of the selective mapping method proposed by Jeon H B is improved, but as each group of signals need to be subjected to phase rotation, IFFT operation, calculation and PAPR value comparison in the selective mapping method, the calculation complexity is closely related to the number of candidate phases, and the more the number of candidate phases is, the higher the calculation complexity is. In practical applications, it is necessary to comprehensively consider the suppression performance and the computational complexity of the PAPR of the OFDM communication system to achieve the balance between the two.

Therefore, a communication system is needed to reduce the PAPR and control the computation complexity without losing the transmission rate, and avoid the defect of single pulse shape.

Disclosure of Invention

The invention mainly aims to provide a PAPR suppression method based on pulse shaping modulation, which reduces the PAPR of a transmitting end on the premise of not influencing the transmission rate and the bit error rate, and can realize flexible configuration of pulse shaping so as to flexibly support different communication scenes.

The invention is realized by the following technical scheme:

A PAPR suppression method based on pulse shaping modulation comprises the following steps:

Step S1, building an end-to-end OFDM communication system model, wherein the communication system model comprises a transmitter, a channel and a receiver, the transmitter comprises a constellation modulation module, an S/P module and a transmitting end PS module which are sequentially connected, the transmitting end PS module is realized by a transmitting DNN, the transmitting DNN comprises a transmitting input layer connected with the S/P module and a transmitting output layer, the transmitting output layer is formed by N parallel PS filters, the transmitting end PS module outputs a time domain transmitting signal S _i＝(G_tx⊙F^H)x_i,s_i as an ith time domain transmitting signal main value sequence, x _i represents an ith frequency domain signal vector, For the stack of N PS filters corresponding to the N subcarriers, F represents the fourier transform matrix, (·) ^H represents the conjugate transpose, "-represents the hadamard product;

the channel is a multipath fading channel;

the receiver comprises an equalization module and a demodulation mapping module which are sequentially connected, wherein the equalization module performs FFT on a time domain received signal obtained through a channel and a time domain channel, and performs corresponding point division on two frequency domain information obtained by the FFT to obtain equalized frequency domain information;

Step S2, jointly training a neural network consisting of transmitters, channels and receivers with a joint loss function l=α ₁L₁+α₂L₂, where α ₁ denotes the transmitter loss function ratio, α ₂ denotes the receiver loss function ratio, S (n) is the nth element of s _i, Representing the probability of symbol demodulation on the kth subcarrier, anM is the modulation order, P _k,1,...,P_k,M represents the single thermal code adopted by the label, the value range is {0,1},As a desired function.

Further, the transmitting DNN further includes a transmitting hidden layer composed of one layer of N neurons, a first Dropout layer, one layer of 2N neurons, a second Dropout layer, one layer of 2N neurons, and a power normalization layer.

Further, the demodulation DNN further comprises a demodulation hidden layer consisting of a layer of 2N neurons, a layer of 4N neurons, a layer of 8N neurons and a layer of 2N neurons, wherein the first layer and the last layer of the demodulation hidden layer are connected in a jumping manner.

Further, the input data of the transmitting DNN and the demodulating DNN needs to be preprocessed, where the preprocessing includes splitting a real part and an imaginary part of each original data, and ordering the real part of each original data to form a real part data block, and ordering the imaginary part of each original data to form an imaginary part data block.

Further, in the step S1, after the signal S _i＝(G_tx⊙F^H)x_i is obtained, the transmitting end PS module adds a cyclic prefix to the signal to combat frequency selective fading.

Further, the neuron ratio parameter of the first Dropout layer and the second Dropout layer is set to 0.5.

Further, the step S2 specifically includes:

Step S21, acquiring 10 ⁶ groups of data as a dataset, dividing the dataset into a training set and a verification set, selecting a joint loss function L=alpha ₁L₁+α₂L₂, and setting super parameters;

s22, forward propagation training, calculating a joint loss function, selecting a proper gradient descent algorithm to finish backward propagation, and updating parameters until the training is finished;

Step S23, counting a loss curve and an accuracy rate fitting curve, paying attention to training time, if the training time is not converged, adjusting super parameters, returning to the step S21, and if the training time is converged, entering the step S24;

And step S24, acquiring 10 ⁵ groups of data as a test set, inputting the test set, and testing the performance by counting the PAPR and BER in the test set.

Further, in the joint loss function, α ₁:α₂ =0.01:3.

Further, the multipath fading channel is represented asWhere τ represents the maximum number of delay paths, δ (·) represents the discrete impulse response, a _p is the complex baseband channel coefficient of the p-th delay path, and τ _p represents the delay of the p-th delay path.

As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following advantages:

1. The transmitting DNN, the channel and the demodulation DNN of the receiving end PS module form an encoder, a delicate pulse forming mechanism is introduced into the traditional OFDM system framework, so that the PAPR of a transmission waveform is optimized, the transmitting DNN and the demodulation DNN are jointly trained by a joint loss function, the purposes of inhibiting the PAPR of the transmission signal and preventing the system error rate from deteriorating are achieved, the balance of reducing the PAPR and controlling the calculation complexity is achieved on the premise of not influencing the transmission rate and the error rate, the flexible configuration of pulse forming can be realized, the defect of single pulse shape is avoided, and different communication scenes are flexibly supported.

2. The invention preprocesses the input data of the transmitting DNN and the demodulating DNN, and can realize better performance on the premise of not influencing the transmission rate.

Drawings

The invention is further described in detail below with reference to the drawings and the specific examples.

Fig. 1 is a functional block diagram of an end-to-end OFDM communication system model of the present invention.

Fig. 2 is a schematic diagram of a PS filter according to the present invention.

Fig. 3 is a schematic diagram of the structure of the emitting DNN of the present invention.

FIG. 4 is a schematic diagram of data preprocessing according to the present invention.

FIG. 5 is a schematic diagram of the present invention before and after using the first dropout layer or the second dropout layer.

FIG. 6 is a graph of the loss function of a training set and a validation set during training of the present invention.

Fig. 7 is a graph comparing the PAPR performance of the present invention with the PAPR performance of the CP-OFDM system under different training super parameters when 16QAM modulation is used.

Fig. 8 is a graph comparing BER performance of the present invention with that of the CP-OFDM system under different training super parameters when 16QAM modulation is adopted.

Detailed Description

The invention is further described below by means of specific embodiments.

The PAPR suppression method based on pulse shaping modulation comprises the following steps:

Step S1, an end-to-end OFDM communication system model shown in fig. 1 is built, wherein the communication system model comprises a transmitter, a channel and a receiver. The transmitter has the function of converting the bit stream into an actual time domain transmitting signal and comprises a constellation modulation module, an S/P module, a transmitting end PS (pulse shaping) module and a CP module which are connected in sequence. The constellation modulation module maps the bit stream into frequency domain constellation symbols according to the selected modulation mode, when the QPSK modulation mode is selected, the constellation modulation module converts continuous 2-bit information into 1 frequency domain symbol, and the value range of the frequency domain symbol is {0.71+0.71j,0.71-0.71j, -0.71+0.71-0.71 j }. The S/P module is used for converting the serial frequency domain signal into a parallel frequency domain signal.

The transmitting end PS module is used for adjusting the power distribution of a transmitting signal on a time-frequency domain so as to finish the optimization of the PAPR, and is realized by transmitting DNN, wherein the transmitting DNN comprises a transmitting input layer, a transmitting hidden layer and a transmitting output layer which are connected with the output end of the S/P module, the input of the transmitting input layer is an OFDM time domain signal, the transmitting output layer is formed by N parallel PS filters, and the transmitting end PS module outputs a time domain transmitting signalWherein N is the number of subcarriers, X (k, i) is the frequency domain symbol corresponding to the ith OFDM signal of the kth subcarrier, and the PS filter corresponding to the ith subcarrier at the transmitting end isFor the transmitting-side pulse shape, T is the duration of the discrete-time-domain OFDM signal, N represents the nth sampling point of the discrete time domain, in this embodiment, the length of the PS filter is defined as N, i.e., t=n, the time-domain transmit signal may be rewritten into the matrix form s _i＝(G_tx⊙F^H)x_i,s_i as the main value sequence of the ith time-domain transmit signal, x _i represents the ith frequency-domain signal vector,For a stack of N PS filters corresponding to the N subcarriers, F represents the fourier transform matrix, (·) ^H represents the conjugate transpose, (·) ^H represents the hadamard product, as shown in fig. 2 which is a schematic diagram of the PS filters, in fig. 2,Is the filter on the kth subcarrierIs the nth value of (2).

After the signal s _i＝(G_tx⊙F^H)x_i is obtained, the transmitting end PS module implements adding a Cyclic Prefix (CP) with a corresponding length to the signal at the CP module to combat frequency selective fading.

The structure of the emitting DNN is shown in fig. 3, and the emitting hidden layer is composed of one layer of N neurons, a first Dropout layer, one layer of 2N neurons, a second Dropout layer, one layer of 2N neurons and a power normalization layer, where the neuron proportion parameters of the first Dropout layer and the second Dropout layer are set to 0.5, which indicates that 50% of the neurons will be deactivated in the layer, and there is no trainable parameter in the power normalization layer. A schematic of the comparison before and after the Dropout layer is used is shown in fig. 5.

The actual design of the PS module at the transmitting end is the Phase value on the PPN (Poly-Phase Network), the data is input into the neural Network to obtain the specific required Phase value, and then the PPN realizes the specific function, thereby providing a method for realizing the digital design and the analog device.

The channel being a multipath fading channel, expressed asWhere τ represents the maximum number of delay paths, δ (·) represents the discrete impulse response, a _p is the complex baseband channel coefficient of the p-th delay path, and τ _p represents the delay of the p-th delay path. Multipath fading channels can be seen as a channel layer without optimizable parameters, with no trainable parameters inside, only enabling the conversion of the transmitted signal to the received signal.

For this channel, the conversion between the transmitted signal x to the received signal y can be expressed as y _u＝ΔHx_u++w_u

Where x _u represents the time-domain transmit signal at the current time, i.e., for the received signal y _u at the current time, the received signal is a cyclic convolution of the channel vector with the transmit signal due to multipath effects and CP.

The receiver does not perform pulse shaping optimization, so the receiver processes the received signal after the CP is removed, and the receiver comprises an equalization module and a demodulation mapping module which are connected in sequence. The equalization module performs N-point FFT on a time domain receiving signal obtained through a channel and the time domain channel, performs corresponding point division on two N-point frequency domain information sequences obtained through the FFT to obtain equalized frequency domain information, and inputs the equalized frequency domain information and the frequency domain channel information into the demodulation mapping module. The demodulation mapping module is realized by demodulation DNN, the demodulation DNN comprises a demodulation input layer, a demodulation hidden layer and a demodulation output layer which are connected with the equalization module, the demodulation output layer is formed by Softmax output splicing on N subcarriers, the dimension and the information bit number are kept consistent to output a final demodulated bit value, the demodulation hidden layer is composed of a layer of 2N neurons, a layer of 4N neurons, a layer of 8N neurons and a layer of 2N neurons, and a first layer and a last layer of the demodulation hidden layer are connected in a jumping way to prevent gradient disappearance.

Step S2, jointly training a neural network consisting of transmitters, channels and receivers with a joint loss function l=α ₁L₁+α₂L₂, where α ₁ denotes the transmitter loss function ratio, α ₂ denotes the receiver loss function ratio,X (n) is the nth element of s _i, The output of a Softmax classifier, representing the probability of symbol demodulation on the kth subcarrier, andM is the modulation order, P _k,1,...,P_k,M represents the single thermal coding adopted by the tag, and because the output purpose is to directly recover the transmission bit quality, the value range of P _k,1,...,P_k,M is {0,1},As a desired function;

The method specifically comprises the following steps:

Step S21, acquiring 10 ⁶ groups of data as data sets, dividing the data sets into training sets and verification sets (dividing the training sets and the verification sets according to 8:2), inputting labels and the data sets, selecting a joint loss function l=α ₁L₁+α₂L₂, and setting super parameters, wherein the super parameters are shown in table 1:

TABLE 1

Super parameter	Description of the invention
		N	Number of subcarriers
Drop_rate	Dropout neuron ratio
		α₁	Transmitter loss function scaling
α₂	Receiver loss function scaling
		lr	Initial learning rate
batch_size	Small lot number
		dl	Learning rate decay index

The neural network is built based on Tensorflow 2.1.1 and Kreas.2.3.1 environments. For the joint loss function l=α ₁L₁+α₂L₂, if α ₁ occupies a relatively large area, the neural network will pay more attention to improving the PAPR suppression capability of the OFDM system in training, and if α ₁ occupies a relatively small area, the neural network will pay more attention to reducing the error rate of the system in training.

Specifically, training of an automatic encoder is completed by adopting an Adam optimization algorithm and a small batch data processing mode, the batch size=1500, and the initial learning rate lr=0.001;

specifically, a learning rate adjustment strategy is adopted in which the learning rate is attenuated every 5 epochs, and the learning rate attenuation rate dl=0.8;

In order to ensure the quality of the data and prevent the loss of the information of the imaginary part, the performance of the neural network is further improved, the input data of the transmitting DNN and the demodulating DNN is required to be preprocessed, as shown in fig. 4, the preprocessing includes splitting the real part and the imaginary part of each original data, sorting and combining the real parts of each original data to form a real part data block, and sorting and combining the imaginary parts of each original data to form an imaginary part data block.

Fig. 6 is a graph of the loss function at a ₁＝0.1,α₂ =10. The trend of the loss function change for the training set and the validation set during training is depicted visually in fig. 5. By observing the curve, the performance of the neural network on the training set and the verification set is almost consistent, the neural network shows good stability and reliability when predicting unknown data, which means that the knowledge learned by the neural network on the training set is also suitable for the verification set, thereby proving that the neural network has good generalization capability.

Fig. 7 and 8 show the performance change of the neural network at different loss function weights of α ₁:α₂, where α ₁ is more important to improve the error rate performance of the system than the smaller one, and α ₁ is more important to improve the PAPR suppression capability of the system than the larger one. By combining the data, the invention finally selects the training of the neural network when the weight ratio of the loss function is alpha ₁:α₂ =0.01:3, and obtains the phase value required to be realized by the simulation device PPN. It can be seen that the PAPR suppression capability of the system is significantly improved compared with the CP-OFDM system (corresponding to the "original signal" in fig. 7 and 8), and the BER performance is not significantly lost, and the transmission rate of the system is not affected. Where the ordinate of fig. 7 is a complementary cumulative distribution function (Complementary Cumulative Distribution Function, CCDF) that is used to represent the statistical properties of the peak-to-average PAPR in an OFDM system.

In the present application, the terms "first," "second," "third," and the like are used merely to distinguish between similar objects and not necessarily to describe a particular sequence or order, nor are they to be construed as indicating or implying a relative importance. In the description, the directions or positional relationships indicated by "upper", "lower", "left", "right", "front" and "rear", etc. are used for convenience of description of the present application based on the directions or positional relationships shown in the drawings, and are not intended to indicate or imply that the apparatus must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the scope of protection of the present application. The specific meaning of the above terms in the present application can be understood by those of ordinary skill in the art according to the specific circumstances.

In addition, in the description of the present application, unless otherwise indicated, "a plurality" means two or more. "and/or" describes an association relationship of an association object, and indicates that there may be three relationships, for example, a and/or B, and may indicate that there are three cases of a alone, a and B together, and B alone. The character "/" generally indicates that the context-dependent object is an "or" relationship.

The foregoing is merely illustrative of specific embodiments of the present invention, but the design concept of the present invention is not limited thereto, and any insubstantial modification of the present invention by using the design concept shall fall within the scope of the present invention.

Claims

1. A PAPR suppression method based on pulse shaping modulation, characterized in that it comprises the following steps:

Step S1: Build an end-to-end OFDM communication system model, which includes a transmitter, a channel, and a receiver. The transmitter includes a constellation modulation module, an S/P module, and a transmitter PS module connected in sequence. The transmitter PS module is implemented by a transmitting DNN. The transmitting DNN includes a transmitting input layer connected to the S/P module, and a transmitting output layer. The transmitting output layer is formed by N parallel PS filters. The transmitting PS module outputs a time domain transmission signal s _i =(G _tx ⊙F ^H ) _xi , where _si is the i-th time domain transmission signal main value sequence, and _xi represents the i-th frequency domain signal vector. is a stack of N PS filters corresponding to N subcarriers, F represents the Fourier transform matrix, (·) ^H represents the conjugate transpose, and ⊙ represents the Hadamard product;

The channel is a multipath fading channel;

The receiver includes an equalization module and a demodulation mapping module connected in sequence. The equalization module performs FFT on the time domain received signal and the time domain channel obtained through the channel, and divides the two frequency domain information obtained by the FFT by corresponding points to obtain the equalized frequency domain information. The demodulation mapping module is implemented by a demodulation DNN. The demodulation DNN includes a demodulation input layer connected to the equalization module and a demodulation output layer. The demodulation output layer is composed of the Softmax outputs on N subcarriers to output the final demodulated bit value.

Step S2: Jointly train the neural network composed of the transmitter, channel, and receiver using the joint loss function L = α ₁ L ₁ + α ₂ L ₂ , where α ₁ represents the ratio of the transmitter loss function, α ₂ represents the ratio of the receiver loss function, s(n) is the nth element of _si , represents the probability of symbol demodulation on the kth subcarrier, and M is the modulation order, P _k,1 ,...,P _k,M represents the one-hot encoding used by the label, and its value range is {0, 1}, is the expected function.

2. The PAPR suppression method based on pulse shaping modulation according to claim 1 is characterized in that the transmitting DNN also includes a transmitting hidden layer consisting of a layer of N neurons, a first dropout layer, a layer of 2N neurons, a second dropout layer, a layer of 2N neurons, and a power normalization layer.

3. A PAPR suppression method based on pulse shaping modulation according to claim 2, characterized in that: the demodulation DNN also includes a demodulation hidden layer consisting of a layer of 2N neurons, a layer of 4N neurons, a layer of 8N neurons, and a layer of 2N neurons, and the first and last layers of the demodulation hidden layer use skip connections.

4. A PAPR suppression method based on pulse shaping modulation according to claim 1, 2 or 3, characterized in that: the input data of the transmitting DNN and the demodulating DNN need to be preprocessed, and the preprocessing includes splitting the real part and the imaginary part of each original data, and sorting the real part of each original data to form a real data block, and sorting the imaginary part of each original data to form an imaginary data block.

5. A PAPR suppression method based on pulse shaping modulation according to claim 1, 2 or 3, characterized in that: in said step S1, after obtaining the signal _si = ( _Gtx ^⊙FH ) _xi , the transmitting end PS module adds a cyclic prefix to the signal to combat frequency selective fading.

6. A PAPR suppression method based on pulse shaping modulation according to claim 2 or 3, characterized in that: the neuron ratio parameter of the first Dropout layer and the second Dropout layer is set to 0.5.

7. The PAPR suppression method based on pulse shaping modulation according to claim 1, 2 or 3, wherein step S2 specifically comprises:

Step S21: Obtain 10 ⁶ sets of data as a data set, divide the data set into a training set and a validation set, select a joint loss function L = α ₁ L ₁ + α ₂ L ₂ , and set hyperparameters;

Step S22: forward propagation training, calculating the joint loss function, and selecting an appropriate gradient descent algorithm to complete back propagation and parameter update until the training is completed;

Step S23: Calculate the loss curve and the accuracy fitting curve, and pay attention to the training time. If it does not converge, adjust the hyperparameters and return to step S21. If it converges, proceed to step S24.

Step S24: Obtain 10 ⁵ groups of data as a test set, input the test set, and test the performance by statistically analyzing the PAPR and BER in the test set.

8 . The PAPR suppression method based on pulse shaping modulation according to claim 6 , wherein in the joint loss function, α ₁ :α ₂ =0.01:3.

9. A PAPR suppression method based on pulse shaping modulation according to claim 1, 2 or 3, characterized in that: the multipath fading channel is represented by Where τ represents the maximum number of delay paths, δ(·) represents the discrete impulse response, a _p is the complex baseband channel coefficient of the p-th delay path, and τ _p represents the delay of the p-th delay path.