CN113973041B - Terahertz signal demodulation method and device and electronic equipment - Google Patents
Terahertz signal demodulation method and device and electronic equipment Download PDFInfo
- Publication number
- CN113973041B CN113973041B CN202010721836.6A CN202010721836A CN113973041B CN 113973041 B CN113973041 B CN 113973041B CN 202010721836 A CN202010721836 A CN 202010721836A CN 113973041 B CN113973041 B CN 113973041B
- Authority
- CN
- China
- Prior art keywords
- neural network
- demodulation
- information
- received signal
- soft information
- 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.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L27/00—Modulated-carrier systems
- H04L27/32—Carrier systems characterised by combinations of two or more of the types covered by groups H04L27/02, H04L27/10, H04L27/18 or H04L27/26
- H04L27/34—Amplitude- and phase-modulated carrier systems, e.g. quadrature-amplitude modulated carrier systems
- H04L27/38—Demodulator circuits; Receiver circuits
- H04L27/3845—Demodulator circuits; Receiver circuits using non - coherent demodulation, i.e. not using a phase synchronous carrier
- H04L27/3854—Demodulator circuits; Receiver circuits using non - coherent demodulation, i.e. not using a phase synchronous carrier using a non - coherent carrier, including systems with baseband correction for phase or frequency offset
- H04L27/3863—Compensation for quadrature error in the received signal
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L27/00—Modulated-carrier systems
- H04L27/32—Carrier systems characterised by combinations of two or more of the types covered by groups H04L27/02, H04L27/10, H04L27/18 or H04L27/26
- H04L27/34—Amplitude- and phase-modulated carrier systems, e.g. quadrature-amplitude modulated carrier systems
- H04L27/38—Demodulator circuits; Receiver circuits
- H04L27/3845—Demodulator circuits; Receiver circuits using non - coherent demodulation, i.e. not using a phase synchronous carrier
- H04L27/3854—Demodulator circuits; Receiver circuits using non - coherent demodulation, i.e. not using a phase synchronous carrier using a non - coherent carrier, including systems with baseband correction for phase or frequency offset
- H04L27/3872—Compensation for phase rotation in the demodulated signal
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The embodiment of the application provides a terahertz signal demodulation method and device and electronic equipment. The method comprises the following steps: performing feature extraction on the received signal to acquire input data of a neural network; inputting the neural network input data to the demodulating neural network to obtain quantized values of various dimensions of the soft information; and determining a demodulation result corresponding to the received signal according to the quantization value of each dimension of the soft information. According to the method, the mixed distortion caused by the non-ideal characteristics of the low-cost device and the influence of the complex noise on the demodulation of the terahertz signal can be effectively reduced, and therefore the problem that the terahertz demodulation performance cannot be guaranteed by the existing signal demodulation system is solved.
Description
Technical Field
The application relates to the technical field of intelligent terminals, in particular to a terahertz signal demodulation method and device and electronic equipment.
Background
The terahertz wave band has a large amount of available spectrum resources, can support high-speed transmission of hundreds of Gbps and even Tbps, and meets the requirement of future mobile communication on data transmission rate, so the terahertz communication becomes one of the key technologies of 6G.
In a terahertz frequency band, signal demodulation faces serious technical challenges, mainly because the characteristics of a terahertz channel are complex, and the problems of complex noise, mixed distortion and the like are caused by the attenuation characteristics in the transmission process and the non-ideal characteristics of a radio frequency device, so that the difficulty in signal demodulation is increased. The terahertz channel is subjected to composite influence of superposition of various noises such as thermal noise, molecular absorption noise, quantum noise and the like, the quality of a received signal is poor, and the demodulation error rate is high. Meanwhile, the cost of the Terahertz radio frequency device is high, and low-cost Terahertz devices and emission structures are often adopted, for example, a solid electronic device TMIC (Terahertz Monolithic Integrated Circuit) used at the Terahertz radio frequency end has low output power and poor linearity, and a series connection mode of a solid electronic amplifier and a vacuum tube amplifier is required for realizing high-power output, so that great nonlinear distortion is introduced, the amplitude phase of a signal is deformed, and the difficulty of signal judgment is increased. In addition, in-phase (Inphase) and Quadrature (Quadrature) branches in the structure of the terahertz transmitter often have I/Q (Inphase/Quadrature) path imbalance caused by amplitude and phase imbalance, so that an EVM (Error Vector Magnitude) of a received signal is increased, and the decision difficulty in signal decision is increased. The instability of the local oscillator of the high-frequency device will cause phase noise to the signal, the phase noise usually includes thermal noise of the local oscillator/mixer, flicker noise of the local oscillator/mixer, etc., and the phase noise will further distort the signal, increasing the error rate of the system. In addition, due to path loss, molecular absorption, electromagnetic properties of device materials, directional transmission and the like, the terahertz channel has complex transient property, and the demodulation performance of signals can be ensured only by compensating the channel properties in real time, so that a robust demodulation algorithm needs to be designed.
In order to reduce the influence of channel irrational characteristics on transmission and improve the accuracy of demodulation decision, the main idea of the existing work is to compensate the channel irrational characteristics, i.e. to correct signal distortion before demodulation decision, thereby improving demodulation performance. The main processes of the compensation technology are as follows: 1) Estimating the channel characteristics, 2) obtaining corresponding inverse characteristics according to the estimated channel characteristics, and adding the inverse characteristics into the signals by a digital signal processing method to obtain the inverse characteristics. And after the signal distortion compensation is finished, demodulating and judging the signal. The compensation technology comprises a transmitting end compensation technology and a receiving end compensation technology, wherein the transmitting end compensation technology is mainly a predistortion technology for compensating the nonlinearity of the power amplifier characteristic, and the technology is characterized in that a power amplifier nonlinear model is constructed, a feedback link is used for obtaining a signal of a power amplifier output end, then parameter estimation is carried out on the power amplifier model, and finally the nonlinear compensation of the signal is completed. However, the predistortion technique needs to accurately acquire the output signal of the power amplifier, and thus, higher requirements are placed on the precision and sampling rate of an Analog to Digital (a/D) converter of a feedback link. However, in terahertz transmission, the signal bandwidth is extremely wide, tens of G or even hundreds of G of bandwidth is used, the a/D sampling requirement is extremely high, and the implementation of a radio frequency link is complex, so that the terahertz transmission is not suitable for being used in terahertz communication. The existing receiving end compensation technology is mainly a channel equalization technology, a receiving end estimates the characteristics of a channel by using pilot frequency data, and compensates the characteristics of the channel based on a method of zero-forcing equalization or Minimum Mean Square Error (MMSE) equalization, the method mainly compensates intersymbol interference of signals, cannot compensate mixed distortion caused by irrational characteristics of devices, and is limited in demodulation performance. Therefore, a more optimal demodulation scheme is required to solve the signal demodulation problem in the terahertz channel.
Disclosure of Invention
Aiming at the problem of signal demodulation under a terahertz channel in the prior art, the application provides a terahertz signal demodulation method, a terahertz signal demodulation device and electronic equipment, and also provides a computer-readable storage medium.
The embodiment of the application adopts the following technical scheme:
in a first aspect, the present application provides a terahertz signal demodulation method, including:
carrying out feature extraction on the received signal to obtain neural network input data, wherein: the received signal is a terahertz signal; the neural network input data corresponds to an input item of a demodulation neural network; the demodulation neural network is a multi-output deep feedforward neural network; the input items of the demodulation neural network comprise amplitude information and phase information; the output item of the deep feedforward neural network is multi-dimensional soft information; the dimension number of the soft information output by the depth feedforward neural network is consistent with the number of the demodulation results corresponding to the terahertz signals; the demodulation neural network is used for fitting a mapping relation between the terahertz signal and the demodulation result, and multiple dimensions of the soft information are used for quantitatively describing a mapping relation between currently input neural network input data and each demodulation result;
inputting the neural network input data to the demodulating neural network to obtain quantized values of various dimensions of the soft information;
and determining a demodulation result corresponding to the received signal according to the quantization value of each dimension of the soft information.
In a feasible implementation manner of the first aspect, the soft information is probability information, multiple dimensions of the soft information are used for describing probabilities that a received signal corresponding to currently input neural network input data respectively corresponds to each demodulation result;
the determining a demodulation result corresponding to the received signal according to the quantization value of each dimension of the soft information includes:
and determining a demodulation result corresponding to the dimension with the maximum probability value in the soft information as a demodulation result corresponding to the received signal.
In a feasible implementation manner of the first aspect, the determining, according to the quantization value of each dimension of the soft information, a demodulation result corresponding to the received signal further includes:
performing normalization processing on the quantization values of all dimensions of the soft information to obtain a multi-dimensional normalization processing result, wherein the sum of the quantization values of all dimensions of the multi-dimensional normalization processing result is 1;
and determining a demodulation result corresponding to the received signal according to the multi-dimensional normalization processing result.
In a feasible implementation manner of the first aspect, the soft information is distance information, multiple dimensions of the soft information are used to describe distances between a received signal corresponding to currently input neural network input data and ideal decision points corresponding to each demodulation result, respectively;
the determining a demodulation result corresponding to the received signal according to the quantization value of each dimension of the soft information includes:
and determining a demodulation result corresponding to the dimension with the minimum distance value in the soft information as a demodulation result corresponding to the received signal.
In a possible implementation manner of the first aspect, after determining a demodulation result corresponding to the received signal according to quantization values of respective dimensions of the soft information, the method further includes:
and outputting error prompt information when the received signals correspond to a plurality of demodulation results according to the quantization values of all dimensions of the soft information.
In a feasible implementation manner of the first aspect, the demodulation neural network corresponds to a modulation manner of the received signal, and the degree of dimension of the soft information is the number of demodulation results corresponding to the modulation manner of the received signal.
In a possible implementation manner of the first aspect, the input item of the demodulating neural network further includes channel characteristic information and/or signal-to-noise ratio information.
In a possible implementation manner of the first aspect, before performing feature extraction on the received signal and acquiring neural network input data, the method further includes:
acquiring a sample set required by neural network sequence training, wherein each training sample in the sample set consists of neural network input and output information corresponding to a sample receiving signal, the input information of the neural network input and output information comprises amplitude information and phase information of the sample receiving signal, and the output information of the neural network input and output information is a quantitative description of a mapping relation between the sample receiving signal and each demodulation result;
constructing a multi-output deep feedforward neural network;
training the multi-output deep feedforward neural network with the sample set to obtain the demodulation neural network.
In a second aspect, the present application provides a terahertz signal demodulation apparatus, including:
the received signal analysis module is used for carrying out feature extraction on the received signal and acquiring input data of the neural network, wherein: the receiving signal is a terahertz signal; the neural network input data corresponds to an input item of a demodulation neural network; the input items of the demodulation neural network comprise amplitude information and phase information; the output item of the deep feedforward neural network is multi-dimensional soft information; the dimension number of the soft information output by the deep feedforward neural network is consistent with the number of demodulation results; the demodulation neural network is used for fitting a mapping relation between the terahertz signal and a demodulation result, and multiple dimensions of the soft information are used for quantitatively describing the mapping relation between currently input neural network input data and each demodulation result;
a neural network module, configured to load the demodulation neural network, and input the neural network input data to the demodulation neural network to obtain quantized values of various dimensions of the soft information;
and the demodulation judging module is used for determining a demodulation result corresponding to the received signal according to the quantization value of each dimension of the soft information.
In a third aspect, the present application provides an electronic device comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein the computer program instructions, when executed by the processor, trigger the electronic device to perform the method steps as described in embodiments of the present application.
In a fourth aspect, the present application provides a computer-readable storage medium having stored thereon a computer program, which, when run on a computer, causes the computer to perform the method of an embodiment of the present application.
According to the technical scheme provided by the embodiment of the application, at least the following technical effects can be realized:
according to the method, demodulation is achieved based on the neural network, extra radio frequency feedback links are not needed, and demodulation of the terahertz signal can be completed only through an algorithm, so that the hardware requirement for demodulation of the terahertz signal is greatly reduced;
furthermore, according to the method of the embodiment of the application, demodulation is realized based on the deep feedforward neural network, the input of the neural network is the signal characteristic information of the received signal, a complete received waveform does not need to be input, the complexity of the neural network construction and calculation is low, and the realization difficulty of demodulation operation is greatly reduced;
furthermore, according to the method of the embodiment of the application, the demodulation result corresponding to the received signal is determined through the multi-dimensional soft information, and because the multi-dimensional soft information describes the mapping relationship between the received signal and each demodulation result, the inaccurate neural network output caused by the overlapping of constellation points can be effectively avoided, so that the accuracy of the demodulation result is greatly improved, and the correct demodulation is realized under the conditions of mixed distortion and complex noise;
according to the method provided by the embodiment of the application, the mixed distortion caused by the non-ideal characteristics of a low-cost device and the influence of complex noise on the demodulation of the terahertz signal can be effectively reduced, so that the problem that the terahertz demodulation performance cannot be ensured by the conventional signal demodulation system is solved.
Drawings
FIG. 1 is a diagram illustrating a demodulator according to an embodiment of the present application;
FIG. 2 is a diagram of a ReLU function;
FIG. 3 is a diagram showing a Tansig function;
FIG. 4 is a flow chart of a method according to an embodiment of the present application;
FIG. 5 is a schematic diagram of a demodulating neural network according to an embodiment of the present application;
fig. 6 shows modulation constellations corresponding to QPSK, 8PSK, and 16 QAM;
FIG. 7 is a flow diagram of a portion of a method according to an embodiment of the present application;
FIG. 8 is a schematic diagram of a demodulating neural network according to an embodiment of the present application;
FIG. 9 is a schematic diagram of a demodulating neural network according to an embodiment of the present application;
fig. 10 is a schematic diagram illustrating a structure of a demodulation neural network for QPSK modulation according to an embodiment of the present application;
fig. 11 is a schematic structural diagram of a demodulation neural network for 8PSK modulation according to an embodiment of the present application;
fig. 12 is a schematic structural diagram of a neural network for demodulating 16PSK modulation according to an embodiment of the present application;
fig. 13 is a schematic structural diagram of a terahertz signal demodulating apparatus according to an embodiment of the present application;
fig. 14 is a schematic structural diagram of a communication device in an application scenario according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the technical solutions of the present application will be described in detail and completely with reference to the following specific embodiments of the present application and the accompanying drawings. It should be apparent that the described embodiments are only some of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The terminology used in the description of the embodiments section of the present application is for the purpose of describing particular embodiments of the present application only and is not intended to be limiting of the present application.
Due to mixed distortion and complex noise of signals under the terahertz channel, a demodulation judgment area of the terahertz signal is changed, a new judgment area is obtained according to a traditional fitting mode with high difficulty, and the performance is limited by the complexity of a fitting function. In an actual application scene, the neural network can fit any nonlinear relation, and a judgment region under the characteristic of an irrational channel can be obtained. Therefore, for the problem of signal demodulation under a terahertz channel in the prior art, in order to improve demodulation performance, in a feasible demodulation scheme, a deep neural network is used for demodulation decision of the terahertz signal. Specifically, based on a demodulation scheme of a learning algorithm, the judgment of the received signal is completed by training a deep neural network.
For example, a demodulation structure based on a deep convolutional neural network performs feature extraction on a received waveform, and then performs demodulation decision on the received waveform. However, this scheme requires a complete waveform to be input into the convolutional neural network to demodulate the signal, and is relatively complex.
For another example, based on a deep belief network in combination with a demodulation algorithm of a support vector machine, features of a received signal are extracted through a neural network, and then a demodulation decision result is obtained by performing classification decision through the support vector machine. However, this scheme requires separately training the neural network and the classifier, and the complexity of demodulator training is high.
In order to solve the problem that the complexity of the demodulation neural network is high in the above scheme, in an embodiment of the present application, demodulation of the terahertz signal is implemented based on a Deep feed forward neural network (DFNN).
Specifically, in an embodiment of the present application, since the input amount of the neural network must be a real amount, the received signal (terahertz signal) is subjected to feature extraction, a real-value signal feature is obtained, and the signal feature is input to the deep feedforward neural network.
Further, for a complex signal (e.g., quadrature Phase Shift Keying (QPSK) signal, 8Phase Shift Keying (8 psk) signal, 16Phase Shift Keying (169psk) signal), a received signal can be completely represented by two pieces of information, i.e., amplitude information and Phase information, and thus, the amplitude information and the Phase information of the received signal are used as input of the deep feedforward neural network.
Fig. 1 is a schematic diagram of a demodulator according to an embodiment of the present application. As shown in fig. 1, the deep feedforward neural network is composed of an input layer 110, a hidden layer 120, and an output layer 130. Characteristic information of the received signal r (111 and 112, where 111 is amplitude information | r | of r and 112 is phase information of r, and r is (r)) is input to the input layer 110, the output layer 130 outputs modulation decision informationAfter obtaining the modulation decision informationThen, the demodulation result of the received signal is further obtained through the decision module 120
According to the scheme of the embodiment shown in fig. 1, the demodulation of the terahertz signal can be realized based on deep feedforward neural network demodulation; due to the fact that the depth feedforward neural network can well fit the nonlinear relation, a more accurate judgment result under the irrational channel characteristic can be obtained by demodulating the terahertz signal based on the depth feedforward neural network, and more accurate terahertz signal demodulation is achieved.
To implement the demodulation scheme of the embodiment shown in fig. 1, a deep feed-forward neural network for demodulation decisions needs to be constructed first. The input layer, the hidden layer and the output layer of the deep feedforward neural network and the layers inside the hidden layer are connected in a feedforward mode to realize fitting of the nonlinear mapping relation between input and output. In the deep feedforward neural network, the input of the next layer is generally the output of the previous layer, wherein the input of the ith layer can be expressed as:
x i =g(W i-1 x i-1 +b i-1 ), (1)
in formula 1, x i Representing the input of the i-th layer, g () representing the activation function of the neural network, W i Weight matrix representing the i-th layer, b i Indicating the bias of the ith layer.
In one embodiment, in the process of constructing the deep feedforward neural network, a constrained linear unit (ReLU) and a tannig function are respectively selected as the activation functions of the output layer and the hidden layer. Fig. 2 is a diagram illustrating a ReLU function. FIG. 3 is a diagram showing the Tansig function.
Specifically, the ReLU function is used as an activation function of the output layer to generate the classification probability, and the input-output relationship thereof can be expressed as:
in formula 2, g R (x R ) Representing the output of the ReLU function, x R Representing the input of the ReLU function.
And by adopting the Tansig function as the activation function of the hidden layer, a nonlinear relationship can be introduced between different layers of the neural network, and the input-output relationship of the Tansig function can be expressed as:
in formula 3, g T (x T ) Representing the output, x, of the Tansig function T Representing the input to the Tansig function.
Further, in the process of constructing the deep feedforward neural network, besides the activation function of each layer of the neural network, the number of hidden layers of the neural network and the number of neurons of each layer need to be determined. The number of layers, the number of neurons and other super parameters are key factors influencing the performance of the neural network, the fitting capability of the neural network can be optimized by selecting reasonable super parameters, and under-fitting and over-fitting of the neural network are avoided.
Further, after the parameters such as the activation function of each layer of the deep feedforward neural network, the number of layers of the hidden layer, and the number of neurons in each layer are set, in order to obtain the deep feedforward neural network satisfying the demodulation requirement, a training sample set including N training samples needs to be generated to train the deep feedforward neural network.
Specifically, in an embodiment of the present application, the training sample set is obtained through analog channel or data acquisition. For example, a known data set is transmitted, a distorted signal after passing through a channel is received at a receiving end, and amplitude information and phase information of the distorted signal are extracted, so that a required training sample can be obtained.
In an embodiment of the present application, each training sample in the training sample set is randomly generated, and each training sample can represent the received signal r and the modulation decision informationThe mapping relationship between them.
Since the deep feed forward neural network for demodulation decisions takes as input the amplitude information as well as the phase information of the received signal. Therefore, only the amplitude information and the phase information of the training samples need to be extracted as the input information of the deep feedforward neural network to be trained.
The output of the deep feedforward neural network to be trained is the modulation decision informationModulation decision information to facilitate utilization of deep feedforward neural network outputDemodulating the received signal in such a way thatIs as close as possible to the actual demodulation result of the received signal, i.e. in the training samplesShould be the demodulation result y ∈ (0, 1, \8230;, 2 M -1), where M is the modulation order of the signal. In summary, the training samples are represented as:
after the training sample set is generated, the deep feedforward neural network can be trained to meet the fitting function of demodulating terahertz signals. The accuracy of the demodulation result is determined by the accuracy of the modulation decision information output by the deep feedforward neural network. Therefore, in the process of training the deep feedforward neural network, the error between the modulation decision information output by the deep feedforward neural network and the optimal modulation decision information needs to be reduced as much as possible.
In order to reduce the decision error and improve the performance of demodulation decision, in an embodiment of the present application, mean Squared Error (MSE) is used as a loss function of the deep feedforward neural network, that is, the MSE is obtained by final trainingNetwork weight W = [ W = 1 ,…,W N ]The following constraints need to be satisfied:
in equation 5, J () represents a Jacobian matrix, epsilon is the target training error, and y is the accurate demodulation result of the received signal.
Further, in order to obtain the neural network parameters satisfying the constraint of equation 5, in an embodiment of the present application, a Levenberg-Marquardt algorithm is used to train the deep feedforward neural network. The Levenberg-Marquardt algorithm is an iterative neural network parameter training method and is suitable for network parameter training when a loss function is the square sum of a nonlinear function. After the (k + 1) th iteration, the parameters of the neural network can be expressed as:
in formula 7, J () represents a Jacobian matrix, e (W) k )=[e 1 (W k ),…,e n (W k ),…,e N (W k )]Is that the network parameter is W k Error between the output of the neural network and the actual index, e n (W k ) Is the error of the nth training sample.
Finally, for a trained deep feedforward neural network, the mapping relationship between the received signal and the modulation decision information can be expressed by the deep feedforward neural network as:
in equation 8, DFNN () represents the input-output relationship of the deep feedforward neural network, r is the received signal,is the modulation decision information obtained by the deep feed-forward neural network.
Further, according to the embodiment shown in fig. 1, although the demodulation structure based on the deep feedforward neural network can perform demodulation decision on the received signal, due to the overlap between constellation points in a low signal-to-noise ratio, these regions will become fuzzy regions in the neural network training process, which affects the training of the neural network. Meanwhile, the two constellation points may have a cross part due to noise, and the decision information output by the neural network may not be accurate information, and may be decided into any one of adjacent modulation indexes when the decision is made, so that demodulation misdecision is caused, and the error rate of the system is increased.
In order to solve the above problems, in order to improve the demodulation performance based on the deep feedforward neural network and enable the deep feedforward neural network to complete effective demodulation decision even under a low signal-to-noise ratio, in an embodiment of the present application, the deep feedforward neural network based on multiple outputs demodulates the terahertz signal. Specifically, the mapping relation between the received signal and the demodulation result is fitted by utilizing the fitting capacity of the depth feedforward neural network for fitting the mapping relation with any precision; respectively outputting mapping relations between the received signals and a plurality of different demodulation results by using a multi-output deep feedforward neural network; and determining one demodulation result which is most matched with the received signal from the plurality of demodulation results as a final demodulation result of the received signal according to the mapping relation between the received signal and the plurality of different demodulation results.
Specifically, in an embodiment of the present application, the deep feedforward neural network for multiple outputs multi-dimensional soft information, each dimension of the soft information is used to describe a mapping relationship between one demodulation result and a received signal in a quantized manner, and the dimension number of the soft information is consistent with the number of demodulation results corresponding to the terahertz signal. Therefore, the demodulation result corresponding to the received signal can be determined from all possible demodulation results of the terahertz signal.
Compared with a single-output neural network structure, the multi-output neural network structure can effectively avoid the training fuzzy problem of different constellation point intersection areas through the accumulation effect of multiple training sample points in the training process, so that the neural network can effectively work under the low signal-to-noise ratio. Furthermore, the multi-output neural network can obtain the description that the received signal belongs to each demodulation result, and the most possible demodulation result of the received signal is obtained through comparison, so that the possibility of misjudgment is reduced, and the demodulation error rate is reduced.
FIG. 4 is a flow chart of a method according to an embodiment of the present application. As shown in fig. 4, in an embodiment of the present application, the following steps are performed to demodulate the received signal:
Fig. 5 is a schematic diagram illustrating a structure of a demodulating neural network according to an embodiment of the present application. As shown in the figureAs shown in fig. 5, the demod neural network is composed of an input layer 310, a hidden layer 320, and an output layer 330. The number of output terms of the output layer 330 is n (n is the number of demodulation results that may exist in the terahertz signal), the characteristic information of the received signal r (311 and 312, where 311 is the amplitude information | r | of r and 312 is the phase information angle (r) of r) is input to the input layer 310, and the output layer 330 outputs the multi-dimensional soft information P 3 (P 3 0,P 3 1,…,P 3 (n-1))。
According to the method shown in the embodiments of fig. 4 and 5, demodulation is realized based on a neural network, and demodulation of the terahertz signal can be completed only through an algorithm without an additional radio frequency feedback link, so that the hardware requirement for demodulation of the terahertz signal is greatly reduced;
further, according to the methods shown in the embodiments of fig. 4 and 5, demodulation is implemented based on the depth feedforward neural network, the input of the neural network is the signal characteristic information of the received signal, a complete received waveform does not need to be input, the complexity of the neural network construction and calculation is low, and the implementation difficulty of demodulation operation is greatly reduced;
further, according to the method shown in the embodiments of fig. 4 and 5, the demodulation result corresponding to the received signal is determined by using the multi-dimensional soft information, and since the multi-dimensional soft information describes the mapping relationship between the received signal and each demodulation result, the output inaccuracy of the neural network due to the overlapping of constellation points can be effectively avoided, so that the accuracy of the demodulation result is greatly improved, and the correct demodulation is realized under the conditions of mixed distortion and complex noise;
according to the method, the mixed distortion caused by the non-ideal characteristics of the low-cost device and the influence of the complex noise on the demodulation of the terahertz signal can be effectively reduced, and therefore the problem that the terahertz demodulation performance cannot be guaranteed by the existing signal demodulation system is solved.
Further, in an actual application scenario, various different modulation modes exist for the terahertz signal. The number of demodulation results that one terahertz signal can be modulated is different corresponding to different modulation modes. For example, for the QPSK modulation scheme, each terahertz signal can carry two bits of binary information (4 th order modulation), that is, for the terahertz signal modulated by the QPSK scheme, the number of corresponding possible demodulation results is 4. For another example, for an 8PSK modulation scheme, each terahertz signal may carry three bits of binary information (8 th order modulation), that is, for a terahertz signal modulated by a QPSK scheme, the number of corresponding possible demodulation results is 8. For another example, for a 16PSK modulation scheme, each terahertz signal can carry four bits of binary information (16 th order modulation), that is, for a terahertz signal modulated by a QPSK scheme, the number of corresponding possible demodulation results is 16. The corresponding modulation constellations of QPSK, 8PSK, and 16QAM are shown in fig. 6.
In order to implement corresponding demodulation for different modulation schemes, in an embodiment of the present application, the demodulation neural network corresponds to a modulation scheme of the received signal, and the dimension number of the soft information output by the demodulation neural network is the number of demodulation results corresponding to the modulation scheme of the received signal. Specifically, in one embodiment, the dimension number of the soft information output by the demodulation neural network is 2 M And M is the order of the modulation mode. For example, for QPSK, the demodulation neural network outputs soft information of 4 dimensions, for 8PSK, the demodulation neural network outputs soft information of 8 dimensions, and for 16PSK, the demodulation neural network outputs soft information of 16 dimensions.
Further, in a practical application scenario, the purpose of demodulating the multidimensional soft information output by the neural network is not limited to determining the demodulation result. Those skilled in the art can apply the multi-dimensional soft information to other purposes according to actual requirements. For example, a network condition analysis is performed based on the multi-dimensional soft information, or a modulation signal offset condition is determined based on the multi-dimensional soft information.
Further, in practical application scenarios, the soft information used for quantizing the mapping relationship between the received signal and the demodulation result may adopt a plurality of different types of formats.
Specifically, in an embodiment of the present application, the soft information is described in a probabilistic manner. Specifically, the soft information is probability information, multiple dimensions of the soft information are used for describing the probability that the received signal corresponding to the currently input neural network input data respectively corresponds to each demodulation result.
For the soft information described in the probability manner, in step 230, the demodulation result corresponding to the dimension with the maximum probability value in the soft information is determined as the demodulation result corresponding to the received signal.
Further, in an application scenario where the soft information is probability information, because a certain error exists in the output of the demodulation neural network and there is no correlation constraint between output results, the multidimensional soft information of the demodulation neural network may not meet the actual probability statistics, thereby being unfavorable for subsequent determination of the demodulation result or other processing procedures. In order to facilitate subsequent processing, in an embodiment of the present application, after the demodulating neural network outputs the multidimensional soft information, normalization processing is performed on the multidimensional soft information to obtain a multidimensional normalization processing result, where a sum of quantization values of each dimension of the multidimensional normalization processing result is 1. And then, according to the multi-dimensional normalization processing result, determining a demodulation result corresponding to the received signal.
In particular, assume that the output of the demodulating neural network isThe following steps are performed:
1) Summing, sum up B:
2) Normalization, dividing each item of B by s to obtain a normalization processing result, wherein each dimension of the normalization processing result is p i =b i /s。
Specifically, in an embodiment of the present application, the soft information is described in a distance manner. Specifically, the soft information is distance information, multiple dimensions of the soft information are used for describing distances between a received signal corresponding to currently input neural network input data and ideal decision points corresponding to each demodulation result.
For the soft information described in the distance manner, in step 230, the demodulation result corresponding to the dimension with the minimum distance value in the soft information is determined as the demodulation result corresponding to the received signal.
Further, in an ideal state, one received signal can correspond to only one demodulation result. However, in an actual application scenario, since the demodulation neural network has a decision error, when a demodulation result corresponding to a received signal is determined from quantized values of respective dimensions of soft information, there is a possibility that one received signal corresponds to a plurality of demodulation results. For example, in an application scenario in which soft information is probability information, two dimensions with equal quantization values exist in certain soft information, and the quantization values of the two dimensions are greater than the quantization values of other dimensions; for another example, in an application scenario in which the soft information is distance information, two dimensions having equal quantization values exist in a certain soft information, and the quantization values of the two dimensions are smaller than the quantization values of the other dimensions.
To avoid the above situation, in an embodiment of the present application, after step 230, the method further includes:
and outputting error prompt information when the received signals correspond to a plurality of demodulation results according to the quantization values of all dimensions of the soft information.
Further, in order to implement the embodiment shown in fig. 2, before step 210 is performed, the demodulating neural network needs to be trained. FIG. 7 is a partial flow diagram of a method according to an embodiment of the present application. Specifically, in an embodiment of the present application, before performing step 210, as shown in fig. 7, the method further includes:
Specifically, taking an application scenario in which soft information is probability information as an example, in order to enable the demodulation neural network to output multi-dimensional soft information, in an implementation manner of step 510, each training sample in a sample set used for training the demodulation neural network is composed of neural network input/output information corresponding to one received signal, where input information of the neural network input/output information includes amplitude phase information and channel information of the received signal, and the output information is a probability that the received signal corresponds to each demodulation result. That is, the training samples used in training the deep feedforward neural network should include probability information of the received signal for each demodulation result. Since the training samples are known data, that is, the accurate demodulation result corresponding to the received signal is known, if the modulation information included in the received signal is the mth demodulation result, the probability that the received signal corresponds to the mth demodulation result is 1, and the probability that the received signal corresponds to other modulation results than the mth demodulation result is 0. That is, the dimension value of the soft information corresponding to the correct demodulation result among the output values of the training samples should be 1, and the other dimension values should be 0. Each training sample can be represented as:
using training samples S n The formed training sample set trains the multi-output deep feedforward neural network, and the obtained demodulation neural network can output one dimension to each received signalDegree of 2 M As a result of (1).
It is to be understood that some or all of the steps or operations in the above-described embodiments are merely examples, and other operations or variations of various operations may be performed by the embodiments of the present application. Further, the various steps may be performed in a different order presented in the above-described embodiments, and it is possible that not all of the operations in the above-described embodiments are performed.
According to the embodiments shown in fig. 4 and fig. 5, the accuracy of the demodulation result can be effectively improved, but the demodulation neural network shown in fig. 5 can only demodulate a specific channel environment, and when the channel environment changes, the network parameters of the demodulation neural network need to be updated to train a new demodulation neural network. However, in practical applications, the overhead of updating the network parameters in real time and training the demodulation neural network is huge and is not easy to implement, and a large amount of training time is required in the training process of the demodulation neural network. Therefore, in an embodiment of the present application, an adaptive demodulation neural network is adopted, so that the demodulation neural network is robust to changes in the channel environment.
In an embodiment of the present application, in order to enable the demodulating neural network to adapt to variations in the signal-to-noise ratio of the received signal, the input of the demodulating neural network further comprises signal-to-noise ratio information.
Fig. 8 is a schematic diagram illustrating a structure of a demodulating neural network according to an embodiment of the present application. As shown in fig. 8, the demod neural network is composed of an input layer 610, a hidden layer 620, and an output layer 630. The number of output terms of the output layer 630 is n (n is the number of demodulation results that may exist in the terahertz signal), the characteristic information of the received signal r is input to the input layer 610 (611, 612, and 613), and the output layer 630 outputs multi-dimensional soft information P 6 (P 6 0,P 6 1,…,P 6 (n-1))。
In one embodiment of the present application, 611 is the amplitude information | r | of r, 612 is the phase information angle (r) of r, 613 is the signal-to-noise ratio information E of r s /N 0 ),
Further, taking an application scenario in which soft information is used as probability information as an example, when 611, 612, and 613 are:
amplitude information | r |, phase information of r, and signal-to-noise ratio information E of r s /N 0 Then (c) is performed.
To train the demodulated neural network shown in fig. 6, the training samples need to have the following form:
in formula 11, E s /N 0 Signal to noise ratio information of the current training sample.
When the demodulation neural network shown in fig. 8 is used for demodulation, 611, 612, and 613 respectively are:
amplitude information of r |, phase information of r angle (r), and signal-to-noise ratio information E of r s /N 0 Then (c) is performed.
Besides extracting the amplitude information and the phase information of each received signal, the signal-to-noise ratio of the received signal of the current frame needs to be estimated and input into a demodulation neural network as an input quantity, so that the demodulation of each received signal is realized.
In an embodiment of the present application, in order to enable the demodulating neural network to adapt to changes in the channel characteristics of the received signal, the input entries of the demodulating neural network further include channel characteristic information.
As shown in fig. 8, in an embodiment of the present application, 611 is amplitude information | r | of r, 612 is phase information angle (r) of r, and 613 is channel characteristic information of r
Further, taking an application scenario in which soft information is used as probability information as an example, when 611, 612, and 613 are:
amplitude information of r |, phase information of r angle (r), and channel characteristic information of rThen (c) is performed. To train the demod neural network shown in fig. 7, the training samples need to have the following form:
in the formula (12), the compound represented by the formula (I),is the channel characteristic information of the current training sample.
When the demodulation neural network shown in FIG. 8 is used for demodulation, the amplitude information | r |, the phase information of r and the channel characteristic information of r are 611, 612 and 613 respectivelyThen (c) is performed. Besides extracting the amplitude information and the phase information of each received signal, the channel characteristics of the received signal of the current frame are estimated and input into a demodulation neural network as input quantity, so that the demodulation of each received signal is realized.
Further, in an actual application scenario, the channel characteristic information input to the demodulation neural network during demodulation can only be a value obtained through channel estimation, and a certain error exists between the estimated channel characteristic and the actual channel characteristic. Then, if the actual channel characteristics are used in the training process, the actual application scenario may not match the training samples, and the performance of the demodulation neural network may be reduced. Therefore, in the training process of the demodulation neural network, the channel characteristic information in the training samples also needs to be the channel characteristic information estimated according to the same channel characteristic method as the actual demodulation application scenario.
In an embodiment of the present application, in order to enable the demodulating neural network to adapt to the signal-to-noise ratio of the received signal and the variation of the channel characteristics, the input items of the demodulating neural network further include signal-to-noise ratio information and channel characteristic information.
Fig. 9 is a schematic diagram illustrating a demodulating neural network according to an embodiment of the present application. As shown in fig. 9, the demod neural network is composed of an input layer 810, a hidden layer 820, and an output layer 830. Output items of output layer 830N (n is the number of demodulation results that may exist in the terahertz signal), and characteristic information (811, 812, 813, and 814) of the received signal r is input to the input layer 810, the output layer 830 outputs multi-dimensional soft information P 8 (P 8 0,P 8 1,…,P 8 (n-1)). Where 811 is amplitude information | r | of r, 812 is phase information of r and 813 is channel characteristic information of rSignal to noise ratio information E of 814 r s /N 0
Further, taking an application scenario in which soft information is probability information as an example, in order to train the demodulation neural network shown in fig. 8, the training samples need to have the following form:
when the demodulation neural network shown in fig. 9 is used for demodulation, in addition to extracting the amplitude information and the phase information of each received signal, the signal-to-noise ratio and the channel characteristics of the received signal of the current frame need to be estimated and input into the demodulation neural network as input quantities, so that demodulation of each received signal is realized.
According to the method, aiming at the robustness requirement of complex transient of the terahertz channel on the demodulator design, the demodulation neural network with the adaptive capacity to the change of the channel is designed, and the received signal under the dynamic channel is correspondingly demodulated, so that the robust demodulation problem of the terahertz signal is solved.
Taking a specific application scenario as an example, fig. 10 is a schematic structural diagram of a demodulation neural network for a QPSK modulation scheme according to an embodiment of the present application. As shown in fig. 10, the demod neural network is composed of an input layer 910, a hidden layer 920, and an output layer 930. The inputs to the input layer 910 of the demod neural network are:
911, receiving amplitude information | r | of the signal r;
912, receiving phase information angle (r) of the signal r;
914, signal-to-noise ratio information E of the received signal r s /N 0
Demodulating inputs to neural network input layer 930 as soft information P 9 (P 9 0,P 9 1,P 9 2,P 9 3)。
Fig. 11 is a schematic structural diagram of a demodulation neural network for 8PSK modulation according to an embodiment of the present application. As shown in fig. 11, the demod neural network is composed of an input layer 1010, a hidden layer 1020, and an output layer 1030. The inputs to the input layer 1010 of the demod neural network are:
1011, receiving amplitude information | r | of the signal r;
1012, receiving phase information angle (r) of the signal r;
1014, signal-to-noise ratio information E of the received signal r s /N 0
Demodulating inputs to the neural network input layer 1030 as soft information P 10 (P 10 0,P 10 1,P 10 2,P 10 3,P 10 4,P 10 5, P 10 6,P 10 7)。
Fig. 12 is a schematic structural diagram of a demodulation neural network for a 16PSK modulation scheme according to an embodiment of the present application. As shown in fig. 12, the demod neural network is composed of an input layer 1110, a hidden layer 1120, and an output layer 1130. The inputs to the input layer 1110 of the demodulation neural network are:
1111, receiving amplitude information | r | of the signal r;
1112, receiving phase information angle (r) of the signal r;
1114, receiving signal-to-noise ratio information E of the signal r s /N 0
The input to the input layer 1130 of the demod neural network is soft information P 11 (P 11 0,P 11 1,…,P 11 15)。
Further, based on the terahertz signal demodulation method provided in the embodiment of the present application, the embodiment of the present application also provides a terahertz signal demodulation device. Fig. 13 is a schematic structural diagram of a terahertz signal demodulation apparatus according to an embodiment of the present application. As shown in fig. 13, the terahertz signal demodulating apparatus includes:
a received signal analyzing module 1310, configured to perform feature extraction on the received signal to obtain neural network input data, where: the received signal is a terahertz signal; the neural network input data corresponds to the input items of the demodulation neural network; the input items of the demodulation neural network comprise amplitude information and phase information; the output item of the deep feedforward neural network is multi-dimensional soft information; the dimension number of the soft information output by the deep feedforward neural network is consistent with the number of the demodulation results; the demodulation neural network is used for fitting the mapping relation between the terahertz signals and the demodulation results, and multiple dimensions of the soft information are used for quantitatively describing the mapping relation between the currently input neural network input data and each demodulation result;
a neural network module 1320, configured to load a demodulation neural network, and input neural network input data into the demodulation neural network to obtain quantized values of each dimension of the soft information;
a demodulation decision module 1330 configured to determine a demodulation result corresponding to the received signal according to the quantization values of the dimensions of the soft information.
Further, in the 90 s of the 20 th century, improvements in a technology could clearly distinguish between improvements in hardware (e.g., improvements in circuit structures such as diodes, transistors, switches, etc.) and improvements in software (improvements in process flow). However, as technology advances, many of today's process flow improvements have been seen as direct improvements in hardware circuit architecture. Designers almost always obtain the corresponding hardware circuit structure by programming an improved method flow into the hardware circuit. Thus, it cannot be said that an improvement in the process flow cannot be realized by hardware physical modules. For example, a Programmable Logic Device (PLD) (e.g., a Field Programmable Gate Array (FPGA)) is an integrated circuit whose Logic functions are determined by an accessing party programming the Device. A digital device is "integrated" on a PLD by the designer's own programming without requiring the chip manufacturer to design and fabricate specialized integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing an Integrated Circuit chip, such Programming is often implemented by "logic compiler" software, which is similar to a software compiler used in program development, but the original code before compiling is also written in a specific Programming Language, which is called Hardware Description Language (HDL), and the HDL is not only one kind but many kinds, such as abll (Advanced boot Expression Language), AHDL (alternate hard Description Language), traffic, CUPL (computer universal Programming Language), HDCal (Java hard Description Language), lava, lola, HDL, PALASM, software, rhydl (Hardware Description Language), and vhul-Language (vhyg-Language), which is currently used in the field. It will also be apparent to those skilled in the art that hardware circuitry that implements the logical method flows can be readily obtained by merely slightly programming the method flows into an integrated circuit using the hardware description languages described above.
The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, atmel AT91SAM, microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic for the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller in purely computer readable program code means, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may thus be considered a hardware component, and the means included therein for performing the various functions may also be considered as a structure within the hardware component. Or even means for performing the functions may be regarded as being both a software module for performing the method and a structure within a hardware component.
In the description of the embodiments of the present application, for convenience of description, the device is described as being divided into various modules/units by functions, the division of each module/unit is only a division of logic functions, and the functions of each module/unit can be implemented in one or more pieces of software and/or hardware when the embodiments of the present application are implemented.
Specifically, the apparatuses proposed in the embodiments of the present application may be wholly or partially integrated into one physical entity or may be physically separated when actually implemented. And these modules can be realized in the form of software called by processing element; or can be implemented in the form of hardware; and part of the modules can be realized in the form of calling by the processing element in software, and part of the modules can be realized in the form of hardware. For example, the detection module may be a separately established processing element, or may be integrated into a chip of the electronic device. The other modules are implemented similarly. In addition, all or part of the modules can be integrated together or can be independently realized. In implementation, each step of the above method or each module above may be implemented by an integrated logic circuit of hardware in a processor element or an instruction in the form of software.
For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as: one or more Application Specific Integrated Circuits (ASICs), or one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs), etc. For another example, these modules may be integrated together and implemented in the form of a System-On-a-Chip (SOC).
An embodiment of the present application also proposes an electronic device comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein the computer program instructions, when executed by the processor, trigger the electronic device to perform the method steps as described in the embodiments of the present application.
In particular, in an embodiment of the present application, the one or more computer programs are stored in the memory, and the one or more computer programs include instructions that, when executed by the apparatus, cause the apparatus to perform the method steps described in the embodiment of the present application.
Specifically, in an embodiment of the present application, the processor of the electronic device may be an on-chip device SOC, and the processor may include a Central Processing Unit (CPU), and may further include other types of processors. Specifically, in an embodiment of the present application, the processor of the electronic device may be a PWM control chip.
Specifically, in an embodiment of the present application, the processors referred to may include, for example, a CPU, a DSP, a microcontroller, or a digital Signal processor, and may further include a GPU, an embedded Neural Network Processor (NPU), and an Image Signal Processing (ISP), and the processors may further include necessary hardware accelerators or logic Processing hardware circuits, such as an ASIC, or one or more integrated circuits for controlling the execution of the programs according to the technical solution of the present application. Further, the processor may have the functionality to operate one or more software programs, which may be stored in the storage medium.
Specifically, in one embodiment of the present application, the memory of the electronic device may be a read-only memory (ROM), other types of static memory devices that can store static information and instructions, a Random Access Memory (RAM), or other types of dynamic memory devices that can store information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compact disc, laser disc, optical disc, digital versatile disc, blu-ray disc, etc.), a magnetic disc storage medium, or other magnetic storage devices, or any computer-readable medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer.
In particular, in an embodiment of the present application, the processor and the memory may be combined into a processing device, and more generally, independent components, and the processor is configured to execute the program code stored in the memory to implement the method described in the embodiment of the present application. In particular implementations, the memory may be integrated with the processor or may be separate from the processor.
Fig. 14 is a schematic structural diagram of a communication device in an application scenario according to an embodiment of the present application. As shown in fig. 14, in a specific application scenario, the communication apparatus 1400 at least includes: one or more transceivers 1401, one or more processors 1402, one or more memories 1403, and one or more antennas 1404. The memory 1403 is used to store instructions. The processor 1402 can invoke instructions in the memory 1403 to cause the communication device 1400 to perform the demodulation methods described herein in embodiments. The processor 1402 and the transceiver 1401, the memory 1403 are connected by a bus so as to realize data exchange. The transceiver 1401 enables terahertz signal communication between the user equipment and the network equipment under the control of the processor 1402.
Further, the apparatuses, devices, or modules illustrated in the embodiments of the present application may be implemented by a computer chip or an entity, or by a product with certain functions.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, apparatus, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media having computer-usable program code embodied in the media.
In the several embodiments provided in the present application, any function, if implemented in the form of a software functional unit and sold or used as a standalone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the methods described in the embodiments of the present application.
Specifically, an embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored, and when the computer program runs on a computer, the computer is caused to execute the method provided by the embodiment of the present application.
An embodiment of the present application further provides a computer program product, which includes a computer program and when the computer program runs on a computer, the computer is caused to execute the method provided by the embodiment of the present application.
The embodiments herein are described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (devices), and computer program products according to embodiments herein. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In the embodiments of the present application, "at least one" means one or more, "and" a plurality "means two or more. "and/or" describes the association relationship of the associated objects, and indicates that three relationships may exist, for example, a and/or B, and may indicate that a exists alone, a and B exist simultaneously, and B exists alone. Wherein A and B can be singular or plural. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship. "at least one of the following" and the like, refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c may represent: a, b, c, a and b, a and c, b and c or a and b and c, wherein a, b and c can be single or multiple.
In the embodiments of the present application, the terms "include", "include" or any other variations are intended to cover non-exclusive inclusions, so that a process, method, article, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising a," "8230," "8230," or "comprising" does not exclude the presence of other like elements in a process, method, article, or apparatus comprising the element.
The application may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The application may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
The embodiments in the present application are described in a progressive manner, and the same and similar parts among the embodiments can be referred to each other, and each embodiment focuses on differences from other embodiments. In particular, as for the apparatus embodiment, since it is substantially similar to the method embodiment, the description is relatively simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
Those of ordinary skill in the art will appreciate that the various elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of electronic hardware and computer software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the technical solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described apparatuses, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
The above description is only for the specific embodiments of the present application, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present disclosure, and all the changes or substitutions should be covered by the protection scope of the present application. The protection scope of the present application shall be subject to the protection scope of the claims.
Claims (11)
1. A terahertz signal demodulation method is characterized by comprising the following steps:
carrying out feature extraction on the received signal to obtain neural network input data, wherein: the received signal is a terahertz signal; the neural network input data corresponds to an input item of a demodulation neural network; the demodulation neural network is a multi-output deep feedforward neural network; the input items of the demodulation neural network comprise amplitude information and phase information; the output item of the demodulation neural network is multi-dimensional soft information; the dimension number of the soft information output by the demodulation neural network is consistent with the number of demodulation results corresponding to the terahertz signals; the demodulation neural network is used for fitting a mapping relation between the terahertz signal and the demodulation result, and multiple dimensions of the soft information are used for quantitatively describing a mapping relation between currently input neural network input data and each demodulation result; the soft information is probability information or distance information; the probability information is used for describing the probability that the received signal corresponding to the input data of the neural network corresponds to the demodulation result; the distance information is used for describing the distance between a received signal corresponding to the input data of the neural network and an ideal decision point corresponding to the demodulation result;
inputting the neural network input data to the demodulating neural network to obtain quantized values of various dimensions of the soft information;
and determining a demodulation result corresponding to the received signal according to the quantization value of each dimensionality of the soft information.
2. The method of claim 1, wherein the soft information is probability information, and multiple dimensions of the soft information are used to describe a probability of each demodulation result corresponding to a received signal corresponding to currently input neural network input data;
the determining a demodulation result corresponding to the received signal according to the quantization value of each dimension of the soft information includes:
and determining the demodulation result corresponding to the dimension with the maximum probability value in the soft information as the demodulation result corresponding to the received signal.
3. The method of claim 2, wherein the determining the demodulation result corresponding to the received signal according to the quantization value of each dimension of the soft information further comprises:
performing normalization processing on the quantization values of all dimensions of the soft information to obtain a multi-dimensional normalization processing result, wherein the sum of the quantization values of all dimensions of the multi-dimensional normalization processing result is 1;
and determining a demodulation result corresponding to the received signal according to the multi-dimensional normalization processing result.
4. The method of claim 1, wherein the soft information is distance information, and multiple dimensions of the soft information are used to describe distances between a received signal corresponding to currently input neural network input data and ideal decision points corresponding to each demodulation result;
the determining a demodulation result corresponding to the received signal according to the quantization value of each dimension of the soft information includes:
and determining a demodulation result corresponding to the dimension with the minimum distance value in the soft information as a demodulation result corresponding to the received signal.
5. The method according to any one of claims 1 to 4, wherein after determining the demodulation result corresponding to the received signal according to the quantization value of each dimension of the soft information, the method further comprises:
and outputting error prompt information when the received signals correspond to a plurality of demodulation results according to the quantization values of all dimensions of the soft information.
6. The method according to any one of claims 1 to 4, wherein the demodulation neural network corresponds to a modulation scheme of the received signal, and the dimensional number of the soft information is the number of demodulation results corresponding to the modulation scheme of the received signal.
7. The method according to any one of claims 1 to 4, wherein the input item of the demod neural network further comprises channel characteristic information and/or signal to noise ratio information.
8. The method according to any one of claims 1 to 4, wherein before the feature extraction is performed on the received signal to obtain the neural network input data, the method further comprises:
acquiring a sample set required by neural network sequence training, wherein each training sample in the sample set consists of neural network input and output information corresponding to a sample receiving signal, the input information of the neural network input and output information comprises amplitude information and phase information of the sample receiving signal, and the output information of the neural network input and output information is a quantitative description of a mapping relation between the sample receiving signal and each demodulation result;
constructing a multi-output deep feedforward neural network;
training the multi-output deep feedforward neural network with the sample set to obtain the demodulation neural network.
9. A terahertz signal demodulation apparatus, comprising:
the received signal analysis module is used for carrying out feature extraction on the received signal and acquiring input data of the neural network, wherein: the receiving signal is a terahertz signal; the neural network input data corresponds to an input item of a demodulation neural network; the input items of the demodulation neural network comprise amplitude information and phase information; the output item of the demodulation neural network is multi-dimensional soft information; the dimension number of the soft information output by the demodulation neural network is consistent with the number of demodulation results; the demodulation neural network is used for fitting a mapping relation between the terahertz signal and a demodulation result, and multiple dimensions of the soft information are used for quantitatively describing the mapping relation between currently input neural network input data and each demodulation result; the soft information is probability information or distance information; the probability information is used for describing the probability that the received signal corresponding to the input data of the neural network corresponds to the demodulation result; the distance information is used for describing the distance between a received signal corresponding to the input data of the neural network and an ideal decision point corresponding to the demodulation result;
a neural network module, configured to load the demodulation neural network, and input the neural network input data to the demodulation neural network to obtain quantized values of various dimensions of the soft information;
and the demodulation judging module is used for determining a demodulation result corresponding to the received signal according to the quantization value of each dimension of the soft information.
10. An electronic device, characterized in that the electronic device comprises a memory for storing computer program instructions and a processor for executing the program instructions, wherein the computer program instructions, when executed by the processor, trigger the electronic device to perform the method steps of any of claims 1-8.
11. A computer-readable storage medium, in which a computer program is stored which, when run on a computer, causes the computer to carry out the method according to any one of claims 1-8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010721836.6A CN113973041B (en) | 2020-07-24 | 2020-07-24 | Terahertz signal demodulation method and device and electronic equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010721836.6A CN113973041B (en) | 2020-07-24 | 2020-07-24 | Terahertz signal demodulation method and device and electronic equipment |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113973041A CN113973041A (en) | 2022-01-25 |
CN113973041B true CN113973041B (en) | 2023-01-06 |
Family
ID=79585662
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010721836.6A Active CN113973041B (en) | 2020-07-24 | 2020-07-24 | Terahertz signal demodulation method and device and electronic equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113973041B (en) |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107147603B (en) * | 2017-05-05 | 2019-10-08 | 西安电子科技大学 | DBPSK demodulation method based on multiple neural network |
KR102310001B1 (en) * | 2017-06-19 | 2021-10-08 | 버지니아 테크 인터렉추얼 프라퍼티스, 인크. | Encoding and decoding of information for wireless transmission using multi-antenna transceivers |
CN110535486B (en) * | 2019-08-07 | 2021-03-02 | 东南大学 | Radio frequency signal direct processing type wireless transceiver based on super surface neural network |
-
2020
- 2020-07-24 CN CN202010721836.6A patent/CN113973041B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN113973041A (en) | 2022-01-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Sánchez-Fernández et al. | SVM multiregression for nonlinear channel estimation in multiple-input multiple-output systems | |
Tang et al. | A new low-complexity near-ML detection algorithm for spatial modulation | |
US8290096B2 (en) | Radius adaptive sphere decoding with probabilistic noise constraint | |
CN101499840B (en) | Iteration detection method for MIMO system | |
CN114097202B (en) | Decoder and method for decoding a signal | |
CN102457470B (en) | Method and apparatus for soft-decision sphere decoding | |
Li et al. | IQ symbols processing schemes with LSTMs in OFDM system | |
Kim et al. | Deep learning-assisted multi-dimensional modulation and resource mapping for advanced OFDM systems | |
CN110417515B (en) | Large-scale MIMO signal detection method based on discrete iterative estimation | |
US8306139B2 (en) | Systems and methods for low-complexity MIMO detection using leaf-node prediction via look-up tables | |
Weon et al. | Learning-aided deep path prediction for sphere decoding in large MIMO systems | |
Zhang et al. | Lightweight automatic modulation classification via progressive differentiable architecture search | |
Bakulin et al. | Bayesian estimation with successive rejection and utilization of a priori knowledge | |
CN114142905A (en) | Improved method of MIMO detection algorithm based on Newton iteration method under time-varying channel | |
CN113973041B (en) | Terahertz signal demodulation method and device and electronic equipment | |
CN109818891B (en) | Lattice reduction assisted low-complexity greedy sphere decoding detection method | |
EP2786541B1 (en) | Metric corrections for demodulators using serial localization with indecision | |
CN106788630B (en) | Power distribution method based on reduction of upper bound of real-time error bit probability | |
CN109639618B (en) | Low-complexity soft output demodulation method suitable for high-order quadrature amplitude modulation signal | |
Che et al. | A low-complexity transceiver design for terahertz communication based on deep learning | |
Krishna Reddy et al. | Enhancing PAPR performance in MIMO-OFDM system using hybrid optimal MMSE-MLSE equalizers | |
US20080232491A1 (en) | Systems and methods for low-complexity mimo detection with analytical leaf-node prediction | |
Chang et al. | Complexity‐reduced maximum‐likelihood hybrid detection for MIMO systems | |
Feng et al. | The design of satellite equalizer based on CNN | |
Bai et al. | An innovative low-complexity detection algorithm for spatial modulation |
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 |