CN115514596B

CN115514596B - OTFS communication receiver signal processing method and device based on convolutional neural network

Info

Publication number: CN115514596B
Application number: CN202210979456.1A
Authority: CN
Inventors: 王斌; 袁壮; 潘寅飞; 郑仕链; 周华吉; 孙彦景; 张育芝; 刘洋
Original assignee: Xian University of Science and Technology
Current assignee: Xian University of Science and Technology
Priority date: 2022-08-16
Filing date: 2022-08-16
Publication date: 2024-06-04
Anticipated expiration: 2042-08-16
Also published as: CN115514596A

Abstract

The invention discloses a signal processing method and a device of an OTFS communication receiver based on a convolutional neural network, wherein the method comprises the following steps: at a transmitting end of an OTFS communication system, acquiring a training data set generated by an OTFS transmitter; training the deep convolutional neural network through the training data set to obtain a trained OTFS receiver signal processing model; taking a signal received by a receiver as an input of an OTFS receiver signal processing model; and acquiring a processed received signal according to the output of the OTFS receiver signal processing model. The information can be recovered with a lower error rate, the signal received by the receiver can be recovered, and the reliability of wireless communication can be improved.

Description

OTFS communication receiver signal processing method and device based on convolutional neural network

Technical Field

The invention relates to the technical field of wireless communication, in particular to an OTFS communication receiver signal processing method and device based on a convolutional neural network.

Background

With the large-scale construction and deployment of highways and highways, the popularity of autopilot technology is increasing. The internet of vehicles (IoV) is one of the key application scenarios of the fifth generation (5G) mobile communication technology, and has become an indispensable requirement for users in the future. However, orthogonal Frequency Division Multiplexing (OFDM), which is widely used in 5G mobile communication systems, is very sensitive to the effect of high doppler shift, which makes OFDM less effective in fast time-varying channels, and it is difficult to meet the increasing demands of future internet of vehicles systems. The method aims to solve the problem of low-delay and high-reliability communication of the Internet of vehicles system in a high-speed mobile scene. Orthogonal Time Frequency Space (OTFS) this is a two-dimensional modulation technique suitable for use with a double dispersion fading channel. Meanwhile, OTFS can be implemented on an OFDM basis, i.e., compatible with long term evolution (long Term Evolution, LTE) architecture by adding additional pre-and post-processing modules.

At present, edge computing based on deep learning plays an important role in internet of vehicles resource scheduling and load balancing. However, in the edge calculation based on the internet of vehicles, the study of deep learning is mostly kept at the network layer. On the other hand, for the study of the physical layer communication of the internet of vehicles, deep learning is mostly adopted to optimize the performance of each communication module, but the local optimization of each module of the communication system is not the overall performance of the receiver.

Therefore, a new technical solution is needed to solve the above problems.

Disclosure of Invention

In order to overcome the problems in the related art, the invention discloses an OTFS communication receiver signal processing method and device based on a convolutional neural network.

According to a first aspect of the disclosed embodiment of the present invention, there is provided an OTFS communication receiver signal processing method based on a convolutional neural network, the method including:

At a transmitting end of an OTFS communication system, acquiring a training data set generated by an OTFS transmitter;

Training the deep convolutional neural network through the training data set to obtain a trained OTFS receiver signal processing model;

taking a signal received by a receiver as an input of an OTFS receiver signal processing model;

And acquiring a processed received signal according to the output of the OTFS receiver signal processing model.

Optionally, the inputting the signal received by the receiver as the OTFS receiver signal processing model includes:

mapping signals received by a receiver in a preset rectangular coordinate system to obtain a real part Re (r) and an imaginary part Im (r) of an IQ signal;

The real part Re (r) and the imaginary part Im (r) of the IQ signal are taken as inputs of an OTFS receiver signal processing model.

Optionally, the method further comprises:

the training data set is:

The loss function is determined as:

Where N _B is the number of samples contained in the small lot, T _ni is the true label on the ith class of the nth sample, and P _ni is the output probability of the ith class of the nth sample.

Optionally, after the receiving the signal received by the receiver as an input to the OTFS receiver signal processing model, the method includes:

Extracting shallow features in the input of the OTFS receiver signal processing model through a shallow feature extraction layer, wherein the shallow feature extraction layer comprises three convolution layers, a batch normalization layer and a ReLU activation layer;

And carrying out depth feature extraction on the output of the shallow feature extraction layer through a backbone network, wherein the backbone network comprises a plurality of Bneck blocks.

Optionally, the convolutional neural network model is MobileNetV a lightweight one-dimensional convolutional neural network model;

The MobileNetV light one-dimensional convolutional neural network model further comprises: convolution layer, batch normalization, reLu activation and global average pooling operations.

According to a second aspect of the disclosed embodiment of the present invention, there is provided an OTFS communication receiver signal processing device based on a convolutional neural network, the receiving card including: the device comprises:

the data set acquisition module is used for acquiring a training data set generated by an OTFS transmitter at a transmitting end of the OTFS communication system;

The model acquisition module is connected with the data set acquisition module, and trains the deep convolutional neural network through the training data set to obtain a trained OTFS receiver signal processing model;

the input module is connected with the model acquisition module and takes a signal received by the receiver as the input of an OTFS receiver signal processing model;

and the output module is connected with the input module and acquires the processed received signal according to the output of the OTFS receiver signal processing model.

Optionally, the input module includes:

The mapping unit maps the signal received by the receiver in a preset rectangular coordinate system to obtain a real part Re (r) and an imaginary part Im (r) of the IQ signal;

and the input unit is connected with the mapping unit and takes a real part Re (r) and an imaginary part Im (r) of the IQ signal as input of an OTFS receiver signal processing model.

Optionally, the training data set is:

The loss function is determined as:

Optionally, the apparatus further includes:

the shallow feature extraction module is used for extracting shallow features in the input of the OTFS receiver signal processing model through a shallow feature extraction layer, wherein the shallow feature extraction layer comprises three convolution layers, a batch normalization layer and a ReLU activation layer;

and the deep feature extraction module is used for carrying out deep feature extraction on the output of the shallow feature extraction layer through a backbone network, and the backbone network comprises a plurality of Bneck blocks.

In summary, the present disclosure relates to a signal processing method and apparatus for an OTFS communication receiver based on a convolutional neural network, where the method includes: at a transmitting end of an OTFS communication system, acquiring a training data set generated by an OTFS transmitter; training the deep convolutional neural network through the training data set to obtain a trained OTFS receiver signal processing model; taking a signal received by a receiver as an input of an OTFS receiver signal processing model; and acquiring a processed received signal according to the output of the OTFS receiver signal processing model. The information can be recovered with a lower error rate, the signal received by the receiver can be recovered, and the reliability of wireless communication can be improved.

In addition, the novel signal processing method of the OTFS receiver in the embodiment of the invention is different from optimizing a certain information recovery module of the receiver by deep learning, and optimizes all communication modules of the receiver as a whole. The neural network is adopted to replace all modules (including carrier and symbol synchronization, channel estimation, equalization, demodulation, channel decoding and the like) of a receiving end to complete the whole process of information recovery, so that the influence of imperfect Channel State Information (CSI) and accumulated errors caused by modularized processing is avoided. Therefore, the influence of factors such as intersymbol interference (Inter Symbol Interference, ISI) caused by multipath effect, inter-carrier interference (Inter-CARRIERINTERFERENCE, ICI) caused by Doppler frequency shift, doppler interference (Inter-Doppler Interference, IDI) and noise in a wireless channel is overcome, and the low-delay and high-reliability wireless communication of a communication system in various complex scenes is ensured.

Additional features and advantages of the present disclosure will be set forth in the detailed description which follows.

Drawings

The accompanying drawings are included to provide a further understanding of the disclosure, and are incorporated in and constitute a part of this specification, illustrate the disclosure and together with the description serve to explain, but do not limit the disclosure. In the drawings:

FIG. 1 is a flow chart illustrating a method of signal processing for an OTFS communication receiver based on a convolutional neural network, in accordance with an exemplary embodiment;

fig. 2 is a schematic diagram of the mapping relationship of the delay-doppler grid Γ and the time-frequency grid Λ;

FIG. 3 is a schematic diagram of the signal transformation relationship of the delay-Doppler domain and the time-frequency domain of an OTFS communication system;

FIG. 4 is a schematic diagram of a signal processing method model of an OTFS communication receiver;

FIG. 5 is a schematic diagram of a convolutional neural network;

FIG. 6 is a block diagram illustrating a signal processing apparatus of an OTFS communication receiver based on a convolutional neural network, according to an exemplary embodiment;

FIG. 7 is a block diagram of an input module according to the one shown in FIG. 6;

FIG. 8 illustrates the bit error rate performance of a conventional OTFS communication receiver algorithm and a novel signal processing method;

FIG. 9 illustrates the bit error rate performance of a conventional OTFS communication receiver in EVA channels and a novel signal processing method employing (7, 4) Hamming codes;

FIG. 10 illustrates the bit error rate performance of a conventional OTFS communication receiver in an ETU channel and a novel signal processing method employing (7, 4) Hamming codes;

FIG. 11 illustrates the bit error rate performance of a conventional OTFS communication receiver and a novel signal processing method employing (7, 4) Hamming codes under EVA channels;

Fig. 12 shows the performance of different implementation models in QPSK and 16QAM modulation modes using the novel signal processing method of (7, 4) hamming codes.

Detailed Description

The following describes in detail the embodiments of the present disclosure with reference to the drawings. It should be understood that the detailed description and specific examples, while indicating and illustrating the disclosure, are not intended to limit the disclosure.

FIG. 1 is a flow chart illustrating a method of signal processing for an OTFS communication receiver based on a convolutional neural network, as shown in FIG. 1, according to an exemplary embodiment, the method comprising:

before describing the signal processing method of the OTFS communication receiver based on the convolutional neural network in the disclosed embodiment of the invention, an OTFS communication system and a deep convolutional neural network are described.

The OTFS communication system comprises a transmitting end and a receiving end, wherein at the OTFS transmitting end, an input signal X can obtain a matrix-form delay Doppler domain symbol X _DD through X _DD＝vec^-1 (X), a time-frequency symbol X _TF＝F_MX_DDF_N ^H is obtained through ISFFT transformation, a transmitting symbol matrix S=G _txF_MX_TF＝G_txX_DDF_N ^H is obtained through Hassenberg transformation, and finally a transmitting signal vector of MN×1 is obtained through s=vec (S)

In a delay-doppler channel, the delay-doppler channel response h (τ, v) better matches the true physical channel through the delay-doppler domain relative to the time-frequency domain channel response h (t, f), visually displaying the different distances and relative velocities of the multiple reflectors. As shown in fig. 4, with respect to the time-frequency channel response extending to the entire time-frequency domain. The channel response is sparse in the delay-doppler domain due to the limited number of reflectors over the channel, and the channel response is only present in a grid of (τ _max,±v_max) ranges, where τ _max is the maximum delay and v _max is the maximum doppler shift.

Due to the limited transmission path and associated delay and doppler spread, the channel can be sparsely represented as:

where P is the number of paths of the multipath channel and h _i、τ_i、v_i is the channel gain, delay and doppler spread corresponding to the ith path. The delay τ _i and the Doppler spread value v _i may be converted from an index l _i,k_i in the delay Doppler domain, where

The channels in matrix form can be expressed as:

H _eff is a matrix of mn×mn, expressed as follows: where n (Delay) is the permutation matrix (forward cyclic shift), (Doppler) is a diagonal matrix, wherein/>

For virtually equivalent channels in matrix form of arbitrary pulses

At the receiving end of the OTFS communication system, the received signal r (t) is obtained by adding noise w (t) after the transmitted signal s (t) passes through Doppler channel h (tau, v)

r(t)＝∫∫h(τ,v)s(t-τ)e^j2πv(t-τ)dτdv+w(t)

Receiving signals r=Hs+w are obtained through channels, at a receiving end, the receiving signals are matrixed R=vec ^-1 (R) to obtain a receiving signal matrix R, then the time-frequency symbol Y _TF＝F_MG_rx R is obtained through the Wiegner transformation, the delay Doppler domain symbol Y _DD＝F_M ^HY_TFF_N＝G_rxRF_N is obtained through the SFFT transformation, and finally the Y _DD is vectorized to obtain the receiving signals

According to the equivalent reduced channel, the received signal can be represented as a received signal which can be reduced to

The mapping relationship between the delay-doppler grid Γ and the time-frequency grid Λ is shown in fig. 2. N-point samples are taken along the doppler axis at intervals of doppler shift resolution Δv=1/NT and M-point samples are taken along the delay axis at intervals of delay spread Δτ=1/mΔf, respectively, in the delay-doppler plane. Resulting in Γ= { (k/NT, l/mΔf), k=0, …, N-1, l=0, …, M-1}. The delay-doppler plane is converted to a time-frequency domain grid Λ by a two-dimensional inverse-octave-finite fourier transform (ISFFT), in which N-point samples are taken along the time axis at intervals T, respectively, and M-point samples are taken along the frequency axis at intervals Δf=1/T, resulting in Λ= { (nT, mΔf), n=0, …, N-1, m=0, …, M-1}. Let t=1/Δf, in the time-frequency domain, the entire packet duration is NT, the occupied bandwidth mΔf (where M is the number of subcarriers in the frequency domain; Δf is the subcarrier spacing; N is the number of slots in the time domain; T is the symbol duration).

CNNs have the advantage of sharing convolution kernels, which allows the network to become deeper. By completing the gradual expression from shallow learning to deep learning of input data, more accurate features can be extracted, and a better visual effect is achieved. However, some CNN models with excellent performance tend to have deeper network model structures and higher model complexity, which is not beneficial for mobile deployment. Wherein MobileNetV is a lightweight CNN model suitable for deployment on mobile or embedded devices. MobileNetV2 inherits the depth separable convolution of MobileNetV 1. The reverse residual structure of MobileNetV also references the residual connection idea of ResNet. Instead MobileNetV a first maps the input tensor from the low-dimensional space to the high-dimensional space by an expansion layer to expand the dimensions. Features are then extracted using depth separable convolutions. Finally, the depth-separable convolution output is compressed through the projection layer, so that the data is mapped from a high-dimensional space to a low-dimensional space, and the network is reduced again. Meanwhile, as the expansion layer and the projection layer both contain the learnable parameters, the whole network structure can learn how to better expand and compress data so as to realize light weight and better performance.

Although Relu activation of the functional model may make the model more robust under low-precision calculations, it is inevitable that some features are lost. To reduce this loss of information, instead of using Relu a batch normalization is added once after convolving the projection layer.

In step 101, at a transmitting end of an OTFS communication system, a training data set generated by an OTFS transmitter is acquired.

Illustratively, in the disclosed embodiment of the invention, the signal in the receiver is processed through the lightweight CNN model, so that reliable information recovery from the IQ signal waveform to the information bit stream is realized. In the embodiment of the invention, a certain module is not optimized in the signal receiving process, a global optimization strategy is considered, and meanwhile, the intelligent processing of OTFS communication receiving signals is realized through a 1D-Conv-MobileNetV structure.

It will be appreciated that before processing the signal received by the receiver, the convolutional neural network model needs to be trained to obtain a trained OTFS receiver signal processing model, so as to process the signal received by the receiver according to the OTFS receiver signal processing model. When training a convolutional neural network model, a training data set needs to be acquired first.

Wherein the training data set is:

Meanwhile, it is also necessary to determine the loss function as:

In step 102, the deep convolutional neural network is trained through the training data set, and a trained OTFS receiver signal processing model is obtained.

Illustratively, the 1D-Conv-MobileNetV convolutional neural network model in the disclosed embodiments of the present invention is shown in FIG. 5, it being understood that the disclosed embodiments of the present invention are architectural designs and improvements over the original MobileNetV.

In order to accurately extract the characteristics of an OTFS communication system under a double dispersion channel, we first designed a shallow characteristic extraction layer consisting of three convolution layers, a batch normalization layer and a ReLU activation layer. The feature extraction layer performs shallow feature extraction on the input. And then, the output of shallow feature extraction is connected in series with the input of a backbone network by utilizing the dense connection idea in DenseNet, depth features are further extracted through the backbone network, and the feature transmission is enhanced by utilizing linear activation, so that the classification precision of the network is improved. The backbone network consists of several Bneck blocks, where n is Bneck repetitions. The last part of 1D-Conv-MobileNetV then consists of convolutional layers, batch normalization, reLu6 activation, and global average pooling operations. The use of an average global pool can reduce the number of network parameters while avoiding overfitting.

The training aim is to optimize the network parameters according to the training data set generated by the transmitter, so that the training model can achieve excellent performance, and meanwhile, the training model tries to be popularized to other data outside the training set. The training set of the novel signal processing method is as follows:

The loss function is the key to model training. The loss function employed in the method herein is cross entropy, defined as:

Where N _B is the number of samples contained in the small lot and T _ni is the true tag on the ith class of the nth sample. P _ni is the output probability of the ith class of the nth sample. The optimization algorithm employed herein, adaptive moment estimation (Adam), is an extension of the random gradient descent (SGD) method, which combines the advantages of adaptive gradient (Adagrad) and root mean square back propagation (RMSProp) on the SGD basis and takes into account momentum. This means Adam has done two optimizations, gradient sliding average and bias correction, solving sparse gradients while maintaining a relatively small computational effort, and good performance can be achieved on non-stationary problems. Adam can not only further reduce parameter update type jitter, but also balance the update speed of each previous parameter, accelerate convergence speed and ensure final convergence.

The Adam algorithm is optimized by calculating an exponentially weighted average of the momentums

V_dθ＝β₁V_dθ+(1-β₁)dθ

S_dθ＝β₂S_dθ+(1-β₂)dθ²

Updated with RMSprop to obtain first and second moment estimates, respectively:

Finally updating parameters

Where β ₁,β₂ is the exponential decay rate of the instantaneous estimate, α is the learning rate, ε is a constant that prevents division by zero. The weight W and the deviation b are based onIt is necessary to update the type parameters in the network.

The training algorithm of the novel signal processing method of the OTFS communication receiver is shown in table 1:

table 1 training algorithm for OTFS Signal processing method

In step 103, the signal received by the receiver is taken as input to the OTFS receiver signal processing model.

The difficulty of the OTFS receiver signal processing method is to design a data preprocessing method and a neural network structure suitable for complex OTFS communication signal processing. CNNs generally use a two-dimensional black-and-white image or a three-dimensional color image as input data. Whereas for communication problems the form of the input data is different from the form of the image data. The input data of the model are the real part Re (r) and the imaginary part Im (r) of the received IQ signal, specifically, the signal received by the receiver is mapped in a preset rectangular coordinate system to obtain the real part Re (r) and the imaginary part Im (r) of the IQ signal; the real part Re (r) and the imaginary part Im (r) of the IQ signal are taken as inputs to the OTFS receiver signal processing model.

Thus in the convolutional layer of the network, the channels of the convolutional kernel are set in one dimension herein. The input of 1D-Conv-MobileNetV is the processed received signal, which can be expressed as:

In step 104, a processed received signal is obtained from the output of the OTFS receiver signal processing model.

By way of example, the signal processing method of the OTFS communication receiver under the vehicle networking is adopted to recover the signal information sent by the traditional OTFS communication transmitter, wherein the 1D-Conv-MobileNetV structure is adopted to replace the receiver of the traditional OTFS communication system, and the whole process of recovering the information of the receiving end is completed. The aim is to understand the complex relationship between the received signal and the transmitted information sequence so as to recover the information under various non-ideal conditions as reliably as possible and to improve the generalization capability of the receiver to non-ideal conditions. The reliability of a wireless communication system is mainly manifested in bit error rate. Thus, the signal processing method of the OTFS communication receiver is designed to minimize the bit error rate, which can be expressed as:

Wherein the method comprises the steps of An information bit stream recovered for a signal processing method of an OTFS communication receiver, delta being a model parameter of the method, F (·; delta) representing a function mapping of the method from input to output.

In addition, the deployment of the deep learning algorithm to the terminal device requires consideration of the memory and computing power requirements, and thus requires model complexity analysis. For the novel signal processing method, the calculation amount of the general convolution layer can be expressed as:

and the calculated amount of depth-separable convolution of MobileNetV2 can be expressed as:

Where H _l,W_l represents the length and width of the feature map, K _l represents the size and length of the convolution kernel, and C _l-1,C_l represents the number of input and output channels of the first convolution layer, respectively.

The calculated amounts of the batch normalization layer and the ReLU layer are as follows:

Fig. 6 is a block diagram illustrating a signal processing apparatus of an OTFS communication receiver based on a convolutional neural network according to an exemplary embodiment, and as shown in fig. 6, the apparatus 600 includes:

The data set obtaining module 610 obtains a training data set generated by the OTFS transmitter at a transmitting end of the OTFS communication system;

The model acquisition module 620 is connected to the data set acquisition module 610, and trains the deep convolutional neural network through the training data set to obtain a trained OTFS receiver signal processing model;

An input module 630, connected to the model acquisition module 620, for taking the signal received by the receiver as an input of an OTFS receiver signal processing model;

And the output module 640 is connected with the input module 630, and obtains a processed received signal according to the output of the signal processing model of the OTFS receiver.

Fig. 7 is a block diagram of an input module according to the one shown in fig. 6, and as shown in fig. 7, the input module 630 includes:

a mapping unit 631 that maps the signal received by the receiver in a preset rectangular coordinate system to obtain a real part Re (r) and an imaginary part Im (r) of the IQ signal;

An input unit 632 is connected to the mapping unit 631, and takes the real part Re (r) and the imaginary part Im (r) of the IQ signal as inputs to the OTFS receiver signal processing model.

Optionally, the training data set is:

The loss function is determined as:

Optionally, the apparatus further comprises:

The method for processing the OTFS communication receiver signal based on the convolutional neural network provided by the invention is used for performing performance simulation:

(1) Simulation parameter setting

At the transmitting end of the OTFS wireless communication system, the OTFS frame structure (M, N) is (4, 7), and the information bit stream is randomly generated. Carrier frequency f _c is 4GHz and subcarrier spacing af is 15KHz. The modulation uses Binary Phase Shift Keying (BPSK), quadrature Phase Shift Keying (QPSK) and Quadrature Amplitude Modulation (QAM). The channel coding adopts (7, 4) hamming coding, and the equalization algorithm adopts a message passing algorithm (MP algorithm).

The training set, the verification set and the test set of the OTFS communication receiving signal processing method are generated through MA TLAB simulation. In the model training process, the Eb/N0 range is 0-8 dB, and the interval is 1dB. At each Eb/N0, the number of samples for the training set is 40,000 and the number of samples for the validation set is 20,000. In the test set, the sample capacity was 40,000. To test the generalization ability of the model, eb/N0 samples in the untrained test set were valued in the range of 0-8 dB during the test with an interval of 0.5dB. The optimization algorithm adopts Adam, the default learning rate alpha is 0.001, and the exponential decay rates beta 1 and beta 2 are respectively 0.9 and 0.999. The initial learning rate was set to 0.001. During training, the small lot size was set to 256 and the maximum training period was 8.

1) Influence of additive white gaussian noise channel:

fig. 8 shows the bit error rate performance of a conventional OTFS communication receiver algorithm and a novel signal processing method. The receiver decodes with hard decisions and Maximum Likelihood (ML) estimates, respectively. The traditional use of receiver hard decisions refers to a decoding method that is not affected by any other factors, except for Additive White Gaussian Noise (AWGN). Since the equiprobable distribution of information bits simulates random generation, ML decisions represent optimal performance under ideal conditions. It can be seen that when Eb/N0 is 8dB, the bit error rate of the conventional OTFS communication receiver in BPSK and QPSK modulation modes can reach 10 ^-5. Meanwhile, the performance of the signal processing method is superior to that of a traditional OTFS receiver. In QPSK modulation mode, the error rate of the proposed method can be close to 10 ^-6 when Eb/N0 is 7dB, and can reach 10 ^-6 when Eb/N0 is 8 dB. When Eb/N0 is 7dB, the error rate of the method in the BPSK modulation mode can reach 10 ^-6, and when Eb/N0 is 8dB, the error rate is 0.

The performance of the novel signal processing method in terms of bit error rate is very close to that of ideal ML decision, and is far superior to that of the traditional hard decision method, and the novel signal processing method also has the potential of approaching the performance limit. On untrained Eb/N0, the novel signal processing method also achieves performance close to ML decision, which shows that the method has good generalization capability for Eb/N0.

2) Influence of different car channel conditions:

Fig. 9 shows the bit error rate performance of a conventional OTFS communication receiver under an EVA channel and a novel signal processing method using (7, 4) hamming codes. At different terminal movement speeds, it can be seen that the error rate value is between 10 ^-2 and 10 ^-3 when Eb/N0 is 8dB, regardless of whether the conventional receiver uses BPSK or QPSK modulation. At the moving speed of the terminal 350Kmph, the error rate performance of the conventional receiver is substantially the same as 500 Kmph. The performance of the algorithm is obviously superior to that of the traditional algorithm under different terminal moving speeds. The bit error rates in the above cases are 10 ^-5 and 10 ^-6 when Eb/N0 is 8 dB. In the BPSK modulation mode, when Eb/N0 is 8dB and the terminal moving speed is 350Kmph, the error rate is close to 10 ^-6.

Fig. 10 shows bit error rate performance and novel signal processing method employing (7, 4) hamming codes in ETU channels for conventional OTFS communication receivers. As can be seen from fig. 10, the new signal processing method can recover information with a lower error rate at different terminal moving speeds when Eb/N0 is 8dB, regardless of whether the conventional receiver adopts BPSK modulation or QPSK modulation, compared with the performance of the conventional algorithm. The performance of the method presented herein in ETU channel and EVA channel is similar and verifies the stability and reliability of the method under different channel conditions.

3) Effects of different modulation modes:

The high-order modulation mode adopted in the practical communication system can improve the spectrum efficiency and the anti-interference capability, but has higher requirements on signal detection. The modulation uses QPSK, 8QAM and 16QAM, respectively. Fig. 11 shows that the bit error rate performance terminal movement speed under EVA channel of the conventional OTFS communication receiver and the novel signal processing method employing (7, 4) hamming code is 500 km/h. In the case of a moving speed of 500Kmph, when Eb/N0 is 8dB, different modulation modes are adopted, and the error rate performance of the conventional receiver is between 10 ^-1 and 10 ^-2. In the novel signal processing method, when Eb/N0 is 8dB, the BER of 16QAM modulation reaches 10 ^-4, the BER of 8QAM modulation is between 10 ^-4 and 10 ^-5, and the BER of QPSK modulation is between 10 ^-5 and 10 ^-6. Simulation results show that the novel signal processing method is still superior to the traditional receiver adopting the high-order modulation under the (7, 4) Hamming code, and also prove that the method provided by the invention still has higher stability when adopting the high-order modulation.

4) The signal processing method has different realization model effects:

Fig. 12 shows the performance of different implementation models in QPSK and 16QAM modulation modes using the novel signal processing method of (7, 4). And the mobile speed of the terminal of the Internet of vehicles channel under the Hamming code is 500km/h. The improved network design is superior to the original network model, and the potential of the novel signal processing method in the anti-interference aspect is reflected. In the QPSK modulation mode, the BER of the native ResNet model is close to 10 ^-4, the BER of the native DeneNet model, the native MobileNetV model, and the model presented herein are all between 10 ^-4 and 10 ^-5 when Eb/N0 is 8 dB. In the 16QAM modulation mode, when Eb/N0 is 8dB, the BER of the native ResNet model reaches 10 ^-5, and the BER of the native DeneNet model, the native MobileNetV model and the proposed model are all between 10 ^-5 and 10 ^-6. The method presented herein achieves the best bit error rate performance by comparing four different implementations.

The intelligent signal processing method is realized by using different CNN models, and the influence of the different CNN models on the reliability of the novel signal processing method is verified. From the above results, it is clear that 1D-Conv-MobileNetV not only reduces the amount of calculation in the process of information recovery, but also realizes lower error rate under different conditions, further improving the reliability of the receiver. The different implementation modes of the novel signal processing method have great influence on the system reliability, and the performance gap between the different implementation modes is obvious. Therefore, selecting a proper CNN model and a reasonable optimization design have an important influence on the reliability of the novel signal processing method.

The preferred embodiments of the present disclosure have been described in detail above with reference to the accompanying drawings, but the present disclosure is not limited to the specific details of the embodiments described above, and various simple modifications may be made to the technical solutions of the present disclosure within the scope of the technical concept of the present disclosure, and all the simple modifications belong to the protection scope of the present disclosure.

In addition, the specific features described in the foregoing embodiments may be combined in any suitable manner, and in order to avoid unnecessary repetition, the present disclosure does not further describe various possible combinations.

Moreover, any combination between the various embodiments of the present disclosure is possible as long as it does not depart from the spirit of the present disclosure, which should also be construed as the disclosure of the present disclosure.

Claims

1. An OTFS communication receiver signal processing method based on a convolutional neural network, the method comprising:

Acquiring a processed received signal according to the output of the OTFS receiver signal processing model;

the method further comprises the steps of: the training data set is: ，

The loss function is determined as: Wherein/> For the number of samples contained in a small batch,/>For/>First/>, of the samplesTrue tags on individual categories,/>Is/>First/>, of the samplesThe output probabilities of the individual categories;

the training method of the OTFS receiver signal processing model comprises the following steps: inputting training data sets Maximum number of iterations/>Instantaneous estimation/>And learning rate/>; Randomly initializing network parameters; the following steps are circularly executed: /(I); From training set/>Random selection/>A sample number; calculating a loss function according to a formula; updating network parameters according to an Adam optimizer; the steps are circularly executed until the function mapping/>, from input to output, is obtained according to the minimized error training；

After said receiving the signal received by the receiver as input to the OTFS receiver signal processing model, the method comprises: extracting shallow features in the input of the OTFS receiver signal processing model through a shallow feature extraction layer, wherein the shallow feature extraction layer comprises three convolution layers, a batch normalization layer and a ReLU activation layer; deep feature extraction is carried out on the output of the shallow feature extraction layer through a backbone network, wherein the backbone network comprises a plurality of Bneck blocks; the convolutional neural network model is MobileNetV light one-dimensional convolutional neural network model; the MobileNetV light one-dimensional convolutional neural network model further comprises: convolution layer, batch normalization, reLu activation and global average pooling operations.

2. The OTFS communication receiver signal processing method based on a convolutional neural network according to claim 1, wherein the inputting the signal received by the receiver as the OTFS receiver signal processing model includes:

3. An OTFS communication receiver signal processing device based on a convolutional neural network, the device comprising:

The output module is connected with the input module and acquires a processed received signal according to the output of the OTFS receiver signal processing model;

the training data set is: The loss function is determined as: Wherein/> For the number of samples contained in a small batch,/>For/>First/>, of the samplesTrue tags on individual categories,/>Is/>First/>, of the samplesThe output probabilities of the individual categories;

The training process of the OTFS receiver signal processing model comprises the following steps: inputting training data sets Maximum number of iterations/>Instantaneous estimation/>And learning rate/>; Randomly initializing network parameters; the following steps are circularly executed: /(I); From training set/>Random selection/>A sample number; calculating a loss function according to a formula; updating network parameters according to an Adam optimizer; the steps are circularly executed until the function mapping/>, from input to output, is obtained according to the minimized error training；

The apparatus further comprises: the shallow feature extraction module is used for extracting shallow features in the input of the OTFS receiver signal processing model through a shallow feature extraction layer, wherein the shallow feature extraction layer comprises three convolution layers, a batch normalization layer and a ReLU activation layer; the deep feature extraction module is used for extracting depth features of the output of the shallow feature extraction layer through a backbone network, and the backbone network comprises a plurality of Bneck blocks; the convolutional neural network model is MobileNetV light one-dimensional convolutional neural network model; the MobileNetV light one-dimensional convolutional neural network model further comprises: convolution layer, batch normalization, reLu activation and global average pooling operations.

4. The OTFS communication receiver signal processing device based on a convolutional neural network according to claim 3, wherein the input module comprises: