CN111368979A - PAPR suppression method for MIMO-OFDM system - Google Patents

PAPR suppression method for MIMO-OFDM system Download PDF

Info

Publication number
CN111368979A
CN111368979A CN202010149663.5A CN202010149663A CN111368979A CN 111368979 A CN111368979 A CN 111368979A CN 202010149663 A CN202010149663 A CN 202010149663A CN 111368979 A CN111368979 A CN 111368979A
Authority
CN
China
Prior art keywords
encoder
decoder
signal
mimo
papr
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010149663.5A
Other languages
Chinese (zh)
Other versions
CN111368979B (en
Inventor
高明
潘毅恒
李靖
廖覃明
黄凤杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Original Assignee
Xidian University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xidian University filed Critical Xidian University
Priority to CN202010149663.5A priority Critical patent/CN111368979B/en
Publication of CN111368979A publication Critical patent/CN111368979A/en
Application granted granted Critical
Publication of CN111368979B publication Critical patent/CN111368979B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Neurology (AREA)
  • Radio Transmission System (AREA)

Abstract

The invention relates to a PAPR suppression method of an MIMO-OFDM system, which comprises the following steps: 1. constructing a self-encoder model by using a convolutional neural network and a fully-connected feedforward neural network; 2. constructing a cost function and performing offline combined training on an encoder and a decoder of the model by using a sample data set to minimize the cost function and respectively obtain feedback model parameters of the encoder and the decoder; inputting the feedback model parameters into an encoder and a decoder respectively to obtain a trained self-encoder model; 3. applying the trained self-encoder model to an MIMO-OFDM system to complete the suppression of PAPR; the invention reduces the delay of the MIMO-OFDM system caused by the PAPR suppression algorithm, obviously improves the PAPR performance on the premise of ensuring the error rate performance without transmitting any extra sideband information; the system has low calculation complexity and high efficiency.

Description

PAPR suppression method for MIMO-OFDM system
The technical field is as follows:
the invention relates to a PAPR suppression method of a wireless communication system, in particular to a PAPR suppression method of an MIMO-OFDM system.
(II) background art:
an mimo (Multiple Input Multiple output) -ofdm (orthogonal frequency division multiplexing) system has the characteristics of high spectrum utilization rate, high transmission rate, strong multipath interference resistance and the like, and is a key technology in 4G and 5G mobile communication technologies. Compared with the siso (Single Input Single output) system, the MIMO-OFDM system also faces the problem caused by high papr (peak to Average Power ratio), that is, the OFDM symbol has large amplitude fluctuation in the time domain, which is easily beyond the dynamic range of the transmission Power amplifier, causing signal nonlinear distortion, resulting in severe performance degradation of the whole system.
The mainstream method for reducing PAPR in MIMO-OFDM system at present is still the traditional method used in SISO and its improved version. Clipping is one of typical conventional PAPR reduction algorithms, and the amplitude envelope of the signal is limited to a predetermined value. This approach can significantly reduce PAPR but at the same time introduces severe signal distortion. PTS and SLM are two common probability techniques, the PTS technique randomly divides the modulated signal to be transmitted into a plurality of sub-blocks, each sub-block selects a corresponding phase factor and carries out weighted superposition on each sub-block, and then IDFT conversion is carried out on the superposed signal to obtain a final transmitting signal; the PTS technique can effectively reduce the PAPR of a transmission signal by selecting an appropriate phase factor. The SLM selects a plurality of groups of phase factors, multiplies the phase factors by modulated signals to be transmitted respectively, performs IDFT conversion, and selects a group of signals with lower PAPR as final transmitting signals. The probability-based technique has good PAPR performance, but requires high computational complexity support, and the receiving end also needs additional sideband information to restore the signal. The ACE is a technology with good performance on ber (bit Error ratio) performance and PAPR performance, and achieves the purpose of giving consideration to both the bit Error rate and PAPR performance by artificially limiting the moving direction of constellation points and moving the modulated signal to be transmitted towards the PAPR reduction direction on a constellation diagram, which has the defect of high computational complexity.
In recent years, artificial intelligence technologies represented by deep learning are widely applied to many fields such as image and audio processing, and great potential in the aspect of optimization algorithms is shown. In consideration of the above-mentioned various PAPR suppression algorithms having the disadvantages of too high computational complexity, limited PAPR Reduction amplitude, etc., a document "a Novel PAPR Reduction Scheme for ofdm system Based on Deep Learning" proposes a PAPR Reduction Scheme PRNet Based on Deep Learning, in which an encoder is placed at a transmitting end, a decoder is placed at a receiving end, and a binary signal without constellation modulation is used as input. Although simulation results show that the PAPR performance of the PRNet is improved compared with the conventional PAPR suppression algorithm, the scheme is directed to a single antenna system and cannot be applied to a MIMO system. Therefore, it is necessary to research a deep learning based PAPR reduction method applicable to the MIMO system.
(III) the invention content:
the technical problem to be solved by the invention is as follows: the PAPR suppression method of the MIMO-OFDM system can greatly reduce the delay caused by the adoption of the PAPR suppression algorithm of the MIMO-OFDM system, and obviously improve the PAPR performance on the premise of ensuring the error code rate performance to the maximum extent without transmitting any additional sideband information; the system has low calculation complexity and high efficiency.
The technical scheme of the invention is as follows:
a PAPR restraining method of a MIMO-OFDM system comprises the following steps:
step 1, constructing a self-encoder model CNN-AE by using a convolutional neural network and a fully-connected feedforward neural network; the self-encoder model CNN-AE comprises an encoder and a decoder, wherein the encoder is arranged at a signal transmitting end of the MIMO-OFDM system, and the decoder is arranged at a signal receiving end of the MIMO-OFDM system; after a signal transmitting end of the MIMO-OFDM system obtains data to be transmitted, constellation modulation is carried out on the data to be transmitted, such as 4-QAM modulation, so as to obtain a modulated signal x to be transmitted, because a neural network can only process real parts and can not directly process the modulated signal x to be transmitted, real parts and imaginary parts of the modulated signal x to be transmitted need to be separated, and the input signal x with the real parts and the imaginary parts separated is obtainedinWill input a signal xinInputting into an encoder, the output of which is the frequency domain signal to be transmittedThe signal X is a frequency domain signal X to be transmitted, is converted into a plurality of paths of signals in a serial/parallel mode and is used as a transmitting signal of the MIMO-OFDM system after OFDM modulation; the input of the decoder is a frequency domain signal obtained by detecting, OFDM demodulating and parallel/serial converting a received signal of the MIMO-OFDM system
Figure BDA0002401984620000021
The output of the decoder is recombined into a complex signal to obtain a reconstructed signal to be demodulated by a constellation
Figure BDA0002401984620000022
The encoder and the decoder respectively comprise 1 input layer, 4 one-dimensional convolutional layers, 4 normalization layers, 1 full-connection layer and 1 output layer;
step 2, training a self-encoder model CNN-AE by using a sample data set:
step 2.1, constructing a cost function: reconstructing the signal
Figure BDA0002401984620000031
The two-norm of the difference with the modulated signal x to be transmitted as a first cost function L of the self-encoder model CNN-AE1(ii) a In order to achieve the purpose of reducing PAPR of each path of OFDM symbols, the PAPR of a frequency domain signal X to be transmitted output by an encoder after OFDM modulation is taken as a second cost function L of a self-encoder model CNN-AE2；
Step 2.2, in order to reduce the cost function, adopting Adam optimization algorithm, using the sample data set to perform offline joint training on the encoder and decoder of the self-encoder model CNN-AE, i.e. traversing the self-encoder model CNN-AE through the training set, so that the first cost function L1Or a second cost function L2Minimizing to respectively obtain feedback model parameters of the encoder and feedback model parameters of the decoder; the feedback model parameters of the encoder comprise convolution kernel, full-connection layer weight, bias and normalization layer parameters of the encoder; the feedback model parameters of the decoder comprise convolution kernel, full link layer weight, bias and normalization layer parameters of the decoder;
step 2.3, respectively inputting the feedback model parameters of the encoder and the feedback model parameters of the decoder into the encoder and the decoder to obtain a trained self-encoder model CNN-AE;
and 3, applying the trained self-encoder model CNN-AE to an MIMO-OFDM system to complete the suppression of PAPR. The trained coder is positioned behind the constellation modulation at the transmitting end of the MIMO-OFDM system, and the trained decoder is positioned in front of the constellation modulation at the receiving end of the MIMO-OFDM system.
In step 1, the output of the decoder is the input signal xinReconstruction data for a target
Figure BDA0002401984620000032
Specifically, in the on-line test or use process, a transmitting end processes data to be transmitted, and inputs a modulated signal x to be transmitted into an encoder after real and imaginary parts are separated and connected, and the modulated signal x is transmitted into a channel after serial/parallel conversion and OFDM modulation; the receiving end carries out signal detection processing, OFDM demodulation and parallel/serial conversion on the received signal, and then inputs the signal into a decoder to obtain a reconstructed signal
Figure BDA0002401984620000033
In the step 1, each one-dimensional convolutional layer contains 1 convolutional kernel and 1 offset, and the one-dimensional convolutional layer performs convolution and offset on input data of the one-dimensional convolutional layer and outputs the input data through a hyperbolic tangent excitation function;
for each normalization layer, input data xnormPerforming special normalization processing to normalize the output of the layer
Figure BDA0002401984620000034
Where γ 'and β' are both parameters to be trained, γ 'is a one-dimensional constant, β' is a constant associated with the input data xnormVectors of the same dimension, v being a constant added to prevent the denominator from being zero, v having a value of 0.001; the existence of the normalization layer can lead the training of the whole neural network to reach convergence more quickly;
the full-link layer of the encoder has 1 full-link layer weight Wfc1And an offset b5, the full link layer of the decoder comprisingWith 1 full connection layer weight Wfc2The dimensions of the bias b10, the bias b5 and the bias b10 are the same, the full-link layer of the encoder and the full-link layer of the decoder both adopt hyperbolic tangent excitation functions, and the full-link layer mainly has the functions of sorting and reducing the dimensions of data processed by the convolutional layer and serves as a final output layer.
In step 2, a sample data set is formed by binary sequences to be transmitted randomly generated by a Matlab software simulation platform, constellation modulation, real-part separation and imaginary-part separation operations are sequentially carried out on the sample data set to obtain sample data, the sample data set is divided into a training set and a testing set according to the proportion of 25:1, and the training set and the testing set are used as the input of a self-encoder model CNN-AE.
In step 2, a first cost function L1Expressed as:
Figure BDA0002401984620000041
w 'in the formula'conv、WfcB, y, β respectively represent convolutional layer weight, full link layer weight, bias of each layer, weight of normalization layer, bias of normalization layer, |2Is the Euclidean norm, T is the number of samples in the sample data set, xtRepresenting the t-th sample sequence in the sample data set,
Figure BDA0002401984620000042
the t-th sample sequence is used as a reconstruction sequence corresponding to the input of the self-encoder model CNN-AE,
Figure BDA0002401984620000043
expressed as:
Figure BDA0002401984620000044
in the formula (f)enAnd fdeRespectively representing an encoder and a decoder, WenAnd benRespectively representing weights and offsets, gamma, in the encoderenAnd βenRepresenting encoder normalization parameters，WdeAnd bdeRepresenting weights and offsets in the decoder, gammadeAnd βdeRepresenting a decoder normalization parameter;
second cost function L2Expressed as:
Figure BDA0002401984620000045
in the formula, XiI is more than or equal to 1 and is more than or equal to N, representing the ith signal of the frequency domain signal X to be transmitted after serial/parallel conversiont，NtThe number of transmitting antennas used for a signal transmitting end of the MIMO-OFDM system, and the PAPR (-) represents the PAPR calculation.
In step 2.2, the offline joint training of the encoder and decoder of the self-encoder model CNN-AE using the sample data set is divided into two stages:
the first stage uses only the first cost function L1As a cost function, the signal-to-noise ratio in each iterative optimization process is randomly changed between 0dB and 30dB so as to achieve the purpose of optimizing the error rate performance;
in the second stage, on the basis of the model after the training in the first stage is finished, a cost function L is used as the cost function, and L is equal to L1+λ·L2And the signal-to-noise ratio is randomly changed between 5 dB and 30dB so as to achieve the purpose of jointly optimizing the error rate and the PAPR performance, wherein lambda is a constant value and is used for balancing the proportion of the error rate performance and the PAPR performance in training deviation.
The invention has the beneficial effects that:
1. the invention uses the trained self-encoder model CNN-AE to participate in the communication transmission process, can greatly reduce the delay of the MIMO-OFDM system caused by adopting the PAPR suppression algorithm, and obviously improves the PAPR performance on the premise of ensuring the error rate performance to the maximum extent without transmitting any additional sideband information.
2. The complexity of the algorithm is mainly concentrated in the on-line training process, and the calculation complexity is low in the actual use process of the trained self-encoder model CNN-AE, so that the efficiency of a communication system is obviously improved, and the practicability of the system is also improved.
(IV) description of the drawings:
FIG. 1 is a schematic structural diagram of a constructed self-encoder model CNN-AE;
FIG. 2 is a comparison of the error rate performance of the present invention in a simulation experiment with the prior art;
fig. 3 is a result of comparing PAPR performance of the present invention in a simulation experiment with that of the prior art.
(V) detailed embodiment:
the MIMO-OFDM system works in a single-user mode and under a multipath Rayleigh fading channel, and a transmitting terminal uses NtThe number of OFDM symbol sub-carriers is N, and N is used by a receiving endrAnd according to the receiving antenna, adopting a zero forcing signal detection scheme. The PAPR suppression method of the MIMO-OFDM system comprises the following steps:
step 1, constructing a self-encoder model CNN-AE (shown in figure 1) by using a convolutional neural network and a fully-connected feedforward neural network; the self-encoder model CNN-AE comprises an encoder and a decoder, wherein the encoder is arranged at a signal transmitting end of the MIMO-OFDM system, and the decoder is arranged at a signal receiving end of the MIMO-OFDM system; after obtaining data to be transmitted, a signal transmitting end of the MIMO-OFDM system performs constellation modulation, such as 4-QAM modulation, on the data to be transmitted to obtain a modulated signal x to be transmitted,
Figure BDA0002401984620000051
c represents a complex set, and because the neural network can only process real numbers and can not directly process modulated signals x to be transmitted, real parts and imaginary parts of the modulated signals x to be transmitted need to be separated to obtain input signals x with the real parts and the imaginary parts separatedin，
Figure BDA0002401984620000052
Figure BDA0002401984620000053
Where Re (x) represents the real part of the modulated signal x to be transmitted, im (x) represents the imaginary part of the modulated signal x to be transmitted, R represents the real number set, and the signal x is inputinInputting into an encoder, wherein the output of the encoder is a frequency domain signal X to be transmitted and a frequency domain to be transmittedThe signal X is converted into a plurality of paths of signals in a serial/parallel mode and is used as a transmitting signal of the MIMO-OFDM system after OFDM modulation; the input of the decoder is a frequency domain signal obtained by detecting, OFDM demodulating and parallel/serial converting a received signal of the MIMO-OFDM system
Figure BDA0002401984620000061
The output of the decoder is recombined into a complex signal to obtain a reconstructed signal to be demodulated by a constellation
Figure BDA0002401984620000062
The encoder and the decoder respectively comprise 1 input layer, 4 one-dimensional convolutional layers, 4 normalization layers, 1 full-connection layer and 1 output layer;
step 2, training a self-encoder model CNN-AE by using a sample data set:
step 2.1, constructing a cost function: reconstructing the signal
Figure BDA0002401984620000063
The two-norm of the difference with the modulated signal x to be transmitted as a first cost function L of the self-encoder model CNN-AE1(ii) a In order to achieve the purpose of reducing PAPR of each path of OFDM symbols, the PAPR of a frequency domain signal X to be transmitted output by an encoder after OFDM modulation is taken as a second cost function L of a self-encoder model CNN-AE2；
Step 2.2, in order to reduce the cost function, adopting Adam optimization algorithm, using the sample data set to perform offline joint training on the encoder and decoder of the self-encoder model CNN-AE, i.e. traversing the self-encoder model CNN-AE through the training set, so that the first cost function L1Or a second cost function L2Minimizing to respectively obtain feedback model parameters of the encoder and feedback model parameters of the decoder; feedback model parameters for an encoder include the convolution kernel W of the encoderconv1,Wconv2,…,Wconv4Full connection layer weight Wfc1Offset b1,b2,…,b5And normalization layer parameter γ1,γ2,…,γ4,β1,β2,…,β4(ii) a The feedback model parameters of the decoder containConvolution kernel W of decoderconv5,Wconv6,…,Wconv8Full connection layer weight Wfc2Offset b6,b7,…,b10And normalization layer parameter γ5,γ6,…,γ8,β5,β6,…,β8；
Step 2.3, respectively inputting the feedback model parameters of the encoder and the feedback model parameters of the decoder into the encoder and the decoder to obtain a trained self-encoder model CNN-AE;
and 3, applying the trained self-encoder model CNN-AE to an MIMO-OFDM system to complete the suppression of PAPR. The trained coder is positioned behind the constellation modulation at the transmitting end of the MIMO-OFDM system, and the trained decoder is positioned in front of the constellation modulation at the receiving end of the MIMO-OFDM system.
In step 1, the output of the decoder is the input signal xinReconstruction data for a target
Figure BDA0002401984620000064
Specifically, in the on-line test or use process, a transmitting end processes data to be transmitted, and inputs a modulated signal x to be transmitted into an encoder after real and imaginary parts are separated and connected, and the modulated signal x is transmitted into a channel after serial/parallel conversion and OFDM modulation; the receiving end carries out signal detection processing, OFDM demodulation and parallel/serial conversion on the received signal, and then inputs the signal into a decoder to obtain a reconstructed signal
Figure BDA0002401984620000071
In step 1, each one-dimensional convolution layer contains 1 convolution kernel WconvAnd 1 of the offsets b,
Figure BDA0002401984620000072
Figure BDA0002401984620000073
where M represents the convolution kernel size, CinAnd CoutRespectively representing the number of input channels and the number of output channels; aThe dimension convolution layer performs convolution and bias on input data of the dimension convolution layer and then outputs the input data through a hyperbolic tangent excitation function;
for each normalization layer, input data xnormPerforming special normalization processing to normalize the output of the layer
Figure BDA0002401984620000074
Where γ 'and β' are both parameters to be trained, γ 'is a one-dimensional constant, β' is a constant associated with the input data xnormVectors of the same dimension, v being a constant added to prevent the denominator from being zero, v having a value of 0.001; the existence of the normalization layer can lead the training of the whole neural network to reach convergence more quickly;
the full-link layer of the encoder has 1 full-link layer weight Wfc1And an offset b5, the global layer of the decoder having 1 global layer weight Wfc2And an offset b10, their dimensions being expressed as:
Figure BDA0002401984620000075
Figure BDA0002401984620000076
Coutrepresenting the number of output channels;
the dimensions of the bias b5 and the bias b10 are the same, the full-link layer of the encoder and the full-link layer of the decoder both adopt hyperbolic tangent excitation functions, and the full-link layer mainly has the functions of sorting and reducing the dimensions of data processed by the convolutional layer and serves as a final output layer.
In step 2, a sample data set is formed by binary sequences to be transmitted randomly generated by a Matlab software simulation platform, and constellation modulation, real part separation and imaginary part separation operations are sequentially carried out on the sample data set to obtain a dimension of 2 N.Nt× 1 input signal xinAnd the composed sample data is divided into a training set and a testing set according to the proportion of 25:1, and the training set and the testing set are used as the input of a self-encoder model CNN-AE.
In step 2, a first cost function L1Expressed as:
Figure BDA0002401984620000077
w 'in the formula'conv、WfcB, y, β respectively represent convolutional layer weight, full link layer weight, bias of each layer, weight of normalization layer, bias of normalization layer, |2Is the Euclidean norm, T is the number of samples in the sample data set, xtRepresenting the t-th sample sequence in the sample data set,
Figure BDA0002401984620000078
the t-th sample sequence is used as a reconstruction sequence corresponding to the input of the self-encoder model CNN-AE,
Figure BDA0002401984620000079
expressed as:
Figure BDA0002401984620000081
in the formula (f)enAnd fdeRespectively representing an encoder and a decoder, WenAnd benRespectively representing weights and offsets, gamma, in the encoderenAnd βenRepresenting the encoder normalization parameter, WdeAnd bdeRepresenting weights and offsets in the decoder, gammadeAnd βdeRepresenting a decoder normalization parameter;
second cost function L2Expressed as:
Figure BDA0002401984620000082
in the formula, XiI is more than or equal to 1 and is more than or equal to N, representing the ith signal of the frequency domain signal X to be transmitted after serial/parallel conversiont，NtThe number of transmitting antennas used for a signal transmitting end of the MIMO-OFDM system, and the PAPR (-) represents the PAPR calculation.
In step 2.2, the offline joint training of the encoder and decoder of the self-encoder model CNN-AE using the sample data set is divided into two stages:
the first stage uses only the first cost function L1As a cost function, the signal-to-noise ratio in each iterative optimization process is randomly changed between 0dB and 30dB so as to achieve the purpose of optimizing the error rate performance;
in the second stage, on the basis of the model after the training in the first stage is finished, a cost function L is used as the cost function, and L is equal to L1+λ·L2And the signal-to-noise ratio is randomly changed between 5 dB and 30dB so as to achieve the purpose of jointly optimizing the error rate and the PAPR performance, wherein lambda is a constant value and is used for balancing the proportion of the error rate performance and the PAPR performance in training deviation.
After training is completed, the optimized Wen、ben、γen、βenAnd Wde、bde、γde、βdeThe equal parameters are respectively input into the encoder and the decoder to obtain the trained encoder and decoder.
And (3) simulation result analysis:
the simulation experiment of the invention is that the running system is Intel(R)Core(TM)i7-6700HQ CPU @2.6GHz, a 64-bit Windows operating system and a Ubuntu16.04 Linux operating system, wherein MATLAB is adopted as simulation software. The simulation experiment adopts 4 transmitting antennas, 4 receiving antennas, 4-QAM modulation, 64 subcarrier numbers and 8 cyclic prefix lengths, and the lambda is 0.45 when the simulation experiment works in a multipath Rayleigh fading channel with 3 multipath time delay. For comparison, the clipping threshold of the clipping method is 0.8, the number of blocks of the PTS is 4, and the number of phase factors is 4.
When the present invention (CNN-AE) is applied to the above environment, compared with the prior art (PTS, clipping method, ACE-SGP) without PAPR suppression algorithm, the bit error rate performance results are shown in fig. 2, and the PAPR performance results are shown in fig. 3. As can be seen from fig. 2, the performance of the present invention is slightly inferior to that of the prior art and without PAPR suppression algorithm at low snr, but the performance of the present invention is superior at medium and high snr; as can be seen from fig. 3, the PAPR performance of the present invention is much higher than that of the prior art, while the PAPR performance of the clipping method is very outstanding, but the error rate performance is deteriorated too much, and the overall performance is too poor. The overall result is reviewed, and compared with the prior art, the method can obviously improve the performance of the MIMO-OFDM system.

Claims (6)

1. A PAPR restraining method of an MIMO-OFDM system is characterized by comprising the following steps: comprises the following steps:
step 1, constructing a self-encoder model CNN-AE by using a convolutional neural network and a fully-connected feedforward neural network; the self-encoder model CNN-AE comprises an encoder and a decoder, wherein the encoder is arranged at a signal transmitting end of the MIMO-OFDM system, and the decoder is arranged at a signal receiving end of the MIMO-OFDM system; a signal transmitting end of the MIMO-OFDM system obtains data to be transmitted, then carries out constellation modulation on the data to be transmitted to obtain a modulated signal x to be transmitted, and carries out real-imaginary part separation on the modulated signal x to be transmitted to obtain an input signal x with the real part and the imaginary part separatedinWill input a signal xinInputting an encoder, wherein the output of the encoder is a frequency domain signal X to be transmitted, and the frequency domain signal X to be transmitted is converted into a plurality of paths of signals in a serial/parallel mode and is used as a transmitting signal of the MIMO-OFDM system after being modulated by OFDM; the input of the decoder is a frequency domain signal obtained by detecting, OFDM demodulating and parallel/serial converting a received signal of the MIMO-OFDM system
Figure FDA0002401984610000011
The output of the decoder is recombined into a complex signal to obtain a reconstructed signal to be demodulated by a constellation
Figure FDA0002401984610000012
The encoder and the decoder respectively comprise 1 input layer, 4 one-dimensional convolutional layers, 4 normalization layers, 1 full-connection layer and 1 output layer;
step 2, training a self-encoder model CNN-AE by using a sample data set:
step 2.1, constructing a cost function: reconstructing the signal
Figure FDA0002401984610000013
With modulated signal x to be transmittedDifference two norm as first cost function L of self-encoder model CNN-AE1(ii) a The PAPR of a frequency domain signal X to be transmitted output by an encoder after OFDM modulation is taken as a second cost function L of a self-encoder model CNN-AE2；
Step 2.2, performing offline joint training on the encoder and the decoder of the self-encoder model CNN-AE by using the sample data set by adopting an Adam optimization algorithm to enable the first cost function L1Or a second cost function L2Minimizing to respectively obtain feedback model parameters of the encoder and feedback model parameters of the decoder; the feedback model parameters of the encoder comprise convolution kernel, full-connection layer weight, bias and normalization layer parameters of the encoder; the feedback model parameters of the decoder comprise convolution kernel, full link layer weight, bias and normalization layer parameters of the decoder;
step 2.3, respectively inputting the feedback model parameters of the encoder and the feedback model parameters of the decoder into the encoder and the decoder to obtain a trained self-encoder model CNN-AE;
and 3, applying the trained self-encoder model CNN-AE to an MIMO-OFDM system to complete the suppression of PAPR.
2. The PAPR suppressing method for MIMO-OFDM system according to claim 1, wherein: in step 1, the output of the decoder is the input signal xinReconstruction data for a target
Figure FDA0002401984610000021
3. The PAPR suppressing method for MIMO-OFDM system according to claim 1, wherein: in the step 1, each one-dimensional convolutional layer contains 1 convolutional kernel and 1 bias, and the one-dimensional convolutional layer performs convolution and bias on input data of the one-dimensional convolutional layer and then outputs the input data through a hyperbolic tangent excitation function;
for each normalization layer, input data xnormNormalizing the output of the normalization layer
Figure FDA0002401984610000022
Where γ 'and β' are both parameters to be trained, γ 'is a one-dimensional constant, β' is a constant associated with the input data xnormVectors with the same dimension, the value of v is 0.001;
the full-link layer of the encoder has 1 full-link layer weight Wfc1And an offset b5, the global layer of the decoder having 1 global layer weight Wfc2And the dimensions of a bias b10, a bias b5 and a bias b10 are the same, and the full link layers of the coder and the full link layers of the decoder adopt hyperbolic tangent excitation functions.
4. The PAPR suppressing method for MIMO-OFDM system according to claim 1, wherein: in the step 2, a binary sequence to be transmitted randomly generated by a Matlab software simulation platform is used for forming a sample data set, the constellation modulation, real-part separation and imaginary-part separation operations are sequentially carried out on the sample data set to obtain sample data, the sample data set is divided into a training set and a testing set according to the proportion of 25:1, and the training set and the testing set are used as the input of a self-encoder model CNN-AE.
5. The PAPR suppressing method for MIMO-OFDM system according to claim 1, wherein: in step 2, the first cost function L1Expressed as:
Figure FDA0002401984610000023
w 'in the formula'conv、WfcB, y, β respectively represent convolutional layer weight, full link layer weight, bias of each layer, weight of normalization layer, bias of normalization layer, |2Is the Euclidean norm, T is the number of samples in the sample data set, xtRepresenting the t-th sample sequence in the sample data set,
Figure FDA0002401984610000024
the t-th sample sequence is used as a reconstruction sequence corresponding to the input of the self-encoder model CNN-AE,
Figure FDA0002401984610000025
expressed as:
Figure FDA0002401984610000026
in the formula (f)enAnd fdeRespectively representing an encoder and a decoder, WenAnd benRespectively representing weights and offsets, gamma, in the encoderenAnd βenRepresenting the encoder normalization parameter, WdeAnd bdeRepresenting weights and offsets in the decoder, gammadeAnd βdeRepresenting a decoder normalization parameter;
second cost function L2Expressed as:
Figure FDA0002401984610000031
in the formula, XiI is more than or equal to 1 and is more than or equal to N, representing the ith signal of the frequency domain signal X to be transmitted after serial/parallel conversiont，NtThe number of transmitting antennas used for a signal transmitting end of the MIMO-OFDM system, and the PAPR (-) represents the PAPR calculation.
6. The PAPR suppressing method for MIMO-OFDM system according to claim 1, wherein: in step 2.2, the offline joint training of the encoder and decoder of the self-encoder model CNN-AE using the sample data set is divided into two stages:
the first stage uses only the first cost function L1As a cost function, the signal-to-noise ratio in each iterative optimization process is randomly changed between 0dB and 30 dB;
in the second stage, on the basis of the model after the training in the first stage is finished, a cost function L is used as the cost function, and L is equal to L1+λ·L2The signal-to-noise ratio varies randomly between 5 dB and 30dB, wherein lambda is a constant value.
CN202010149663.5A 2020-03-06 2020-03-06 PAPR suppression method for MIMO-OFDM system Active CN111368979B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010149663.5A CN111368979B (en) 2020-03-06 2020-03-06 PAPR suppression method for MIMO-OFDM system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010149663.5A CN111368979B (en) 2020-03-06 2020-03-06 PAPR suppression method for MIMO-OFDM system

Publications (2)

Publication Number Publication Date
CN111368979A true CN111368979A (en) 2020-07-03
CN111368979B CN111368979B (en) 2023-03-28

Family

ID=71210336

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010149663.5A Active CN111368979B (en) 2020-03-06 2020-03-06 PAPR suppression method for MIMO-OFDM system

Country Status (1)

Country Link
CN (1) CN111368979B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111817997A (en) * 2020-07-21 2020-10-23 西安电子科技大学 Method for suppressing combined rotation precoding and PTS peak-to-average power ratio in OFDM system
CN112702095A (en) * 2020-12-28 2021-04-23 西安交通大学 PAPR suppression method based on constructive interference precoding in MIMO-OFDM
CN113938207A (en) * 2021-10-15 2022-01-14 湖南师范大学 Signal processing method applied to optical non-Hermite symmetric OFDM receiver
CN116609672A (en) * 2023-05-16 2023-08-18 国网江苏省电力有限公司淮安供电分公司 Energy storage battery SOC estimation method based on improved BWOA-FNN algorithm

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160065275A1 (en) * 2014-08-27 2016-03-03 MagnaCom Ltd. Multiple input multiple output communications over nonlinear channels using orthogonal frequency division multiplexing
CN107078982A (en) * 2014-11-18 2017-08-18 华为技术有限公司 The dispersive estimates of digital coherent optical receiver
US20180357542A1 (en) * 2018-06-08 2018-12-13 University Of Electronic Science And Technology Of China 1D-CNN-Based Distributed Optical Fiber Sensing Signal Feature Learning and Classification Method
CN109600335A (en) * 2019-01-17 2019-04-09 山东建筑大学 The comprehensive PAPR suppressing method of ACO-OFDM system neural network based and system
CN109743275A (en) * 2018-12-25 2019-05-10 北京信息科技大学 It is a kind of based on the signal modulate method for owing complete self-encoding encoder

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160065275A1 (en) * 2014-08-27 2016-03-03 MagnaCom Ltd. Multiple input multiple output communications over nonlinear channels using orthogonal frequency division multiplexing
CN107078982A (en) * 2014-11-18 2017-08-18 华为技术有限公司 The dispersive estimates of digital coherent optical receiver
US20180357542A1 (en) * 2018-06-08 2018-12-13 University Of Electronic Science And Technology Of China 1D-CNN-Based Distributed Optical Fiber Sensing Signal Feature Learning and Classification Method
CN109743275A (en) * 2018-12-25 2019-05-10 北京信息科技大学 It is a kind of based on the signal modulate method for owing complete self-encoding encoder
CN109600335A (en) * 2019-01-17 2019-04-09 山东建筑大学 The comprehensive PAPR suppressing method of ACO-OFDM system neural network based and system

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
倪嘉成等: "结合侧抑制机制的自动编码器训练新算法", 《计算机应用与软件》 *
刘元元等: "MIMO SC-FDMA系统中半盲信道估计新方法", 《南京邮电大学学报(自然科学版)》 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111817997A (en) * 2020-07-21 2020-10-23 西安电子科技大学 Method for suppressing combined rotation precoding and PTS peak-to-average power ratio in OFDM system
CN112702095A (en) * 2020-12-28 2021-04-23 西安交通大学 PAPR suppression method based on constructive interference precoding in MIMO-OFDM
CN112702095B (en) * 2020-12-28 2021-11-19 西安交通大学 PAPR suppression method based on constructive interference precoding in MIMO-OFDM
CN113938207A (en) * 2021-10-15 2022-01-14 湖南师范大学 Signal processing method applied to optical non-Hermite symmetric OFDM receiver
CN113938207B (en) * 2021-10-15 2023-02-24 湖南师范大学 Signal processing method applied to optical non-Hermite symmetric OFDM receiver
CN116609672A (en) * 2023-05-16 2023-08-18 国网江苏省电力有限公司淮安供电分公司 Energy storage battery SOC estimation method based on improved BWOA-FNN algorithm
CN116609672B (en) * 2023-05-16 2024-05-07 国网江苏省电力有限公司淮安供电分公司 Energy storage battery SOC estimation method based on improved BWOA-FNN algorithm

Also Published As

Publication number Publication date
CN111368979B (en) 2023-03-28

Similar Documents

Publication Publication Date Title
CN111368979B (en) PAPR suppression method for MIMO-OFDM system
Ye et al. Power of deep learning for channel estimation and signal detection in OFDM systems
US9960831B2 (en) Multiple input multiple output orthogonal frequency division multiplexing with index modulation, MIMO-OFDM-IM, communications system
CN106789764B (en) Joint Weighted Threshold denoises and the transform domain quadratic estimate method of balanced judgement
CN109547381B (en) Self-encoder-based DCO-OFDM system PAPR suppression method and system
US8488684B2 (en) Methods and systems for hybrid MIMO decoding
US7483364B2 (en) Method and apparatus for detecting STBC-OFDM signals in time-variant channels
CN114745230A (en) OTFS signal receiving and recovering method based on deep neural network structure
CN112953653B (en) Single-carrier multi-user underwater acoustic communication method
Qing et al. Joint model and data-driven receiver design for data-dependent superimposed training scheme with imperfect hardware
US20070206697A1 (en) Signal receiving method and signal receiving equipment for multiple input multiple output wireless communication system
CN113971430A (en) Signal detection and model training method, device, equipment and storage medium
CN114039821B (en) Wideband mmWave MIMO-OFDM system wave beam space channel estimation method
CN114499601B (en) Large-scale MIMO signal detection method based on deep learning
CN114745246A (en) Working method of orthogonal time-frequency space modulation system based on orthogonal space modulation
CN113726697A (en) OTFS symbol detection method based on confidence space dynamic decision
CN112636855A (en) OFDM signal detection method
CN102710567A (en) Part judgment method in interference elimination technology for multiple-input multiple-output (MIMO) wireless communication receiver
CN109818663B (en) Low-complexity differential orthogonal space modulation detection method
CN109167748B (en) Partial maximum likelihood detection method based on energy sorting
Huang et al. Deep learning based parallel detector for MIMO systems
Xin et al. Dmcnet: Data-driven multi-Pilot convolution neural network for MIMO-OFDM receiver
CN114528925B (en) Time-varying channel OFDM signal equalization method based on deep classification network
Yang et al. Residual network-based channel estimation for the protograph LDPC-coded OFDM systems
Lam et al. Low complexity approaches of K-Best MIMO decoder for 802.11 ac WLAN system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant