阅读说明：本技术 高动态信道的符号检测方法、装置、设备及存储介质 (Symbol detection method, device, equipment and storage medium of high dynamic channel ) 是由 吕宣涛 廖晨 于 2021-01-08 设计创作，主要内容包括：本发明公开一种高动态信道的符号检测方法、装置、设备及存储介质,其中,所述高动态信道的符号检测方法包括步骤：根据预设调制阶数获取多个符号集序列；对每个符号集序列中的待检测符号进行降采样和滤波处理,得到与待检测符号对应的采样序列；将采样序列输入预设长短期记忆神经网络中,得到似然函数值序列；根据似然函数值序列对符号集序列进行检测,将采样序列作为神经网络的输入,极大的降低了神经网络的计算复杂度,而不是常规的以通信符号作为最小处理单元,解决了高通信道下,信道在一个符号周期内发生变化的难点。(The invention discloses a symbol detection method, a device, equipment and a storage medium of a high dynamic channel, wherein the symbol detection method of the high dynamic channel comprises the following steps: obtaining a plurality of symbol set sequences according to a preset modulation order; performing down-sampling and filtering processing on the symbols to be detected in each symbol set sequence to obtain sampling sequences corresponding to the symbols to be detected; inputting the sampling sequence into a preset long-short term memory neural network to obtain a likelihood function value sequence; the symbol set sequence is detected according to the likelihood function value sequence, the sampling sequence is used as the input of the neural network, the calculation complexity of the neural network is greatly reduced, the conventional communication symbol is not used as the minimum processing unit, and the difficulty that the channel changes in a symbol period under a high-pass channel is solved.)

1. A symbol detection method of a high dynamic channel is characterized in that the symbol detection method of the high dynamic channel comprises the following steps:

obtaining a plurality of symbol set sequences according to a preset modulation order;

performing down-sampling and filtering processing on the symbols to be detected in each symbol set sequence to obtain sampling sequences corresponding to the symbols to be detected;

inputting the sampling sequence into a preset long-short term memory neural network to obtain a likelihood function value sequence;

and detecting the symbol set sequence according to the likelihood function value sequence.

2. The method for symbol detection of high dynamic channel according to claim 1, wherein the step of down-sampling and filtering the symbol to be detected in each of the symbol set sequences to obtain the sampling sequence corresponding to the symbol to be detected comprises:

calculating the number Q of orthogonal bases used for filtering;

acquiring the Q sampling points of the symbol to be detected;

and constructing a filtering formula according to the number of the orthogonal bases, and substituting the Q sampling points into the filtering formula to obtain a sampling sequence corresponding to the symbol to be detected.

3. The method for symbol detection of a high dynamic channel as claimed in claim 2, wherein said Q is calculated by the formula:

Q＝2[f_dTT_s]+1, wherein, said f_dIs a frequency extension parameter, T is the number of sampling points of the symbol to be detected, T_sIs the sampling time interval.

4. The method for symbol detection of a high dynamic channel as in claim 3, wherein said filtering formula is:

r ≈ gx + n, where r represents a sampling sequence, g represents a high dynamic channel, and g is calculated by:said u is_kRepresents an orthogonal base, said q_kIs represented by the formula_kThe corresponding coefficients.

5. The method for symbol detection of high dynamic channel as claimed in claim 4, wherein u is a symbol of a channel of a high dynamic state_kThe calculation formula of (2) is as follows:

Cu_k＝λ_ku_kwherein C is a matrix of Q × Q, and the elements of CWherein, i and j respectively represent the ith row and the jth column of the matrix C, i is more than or equal to 1, and Q is more than or equal to j.

6. The method as claimed in claim 5, wherein said step of inputting said sample sequence into a predetermined long-short term memory neural network to obtain a sequence of likelihood function values comprises:

carrying out real-number conversion on the sampling sequence, and arranging the sampling sequence after the real-number conversion according to the sequence of receiving the symbol time to be detected corresponding to the sampling sequence;

according to the sequence of the time, taking the first sampling sequence as a starting point, and sequentially arranging the sampling sequences with the adjacent preset symbol number as a group until obtaining a sampling sequence group with the group number as the preset symbol number;

inputting all the sampling sequence groups of each symbol set sequence into a preset long-short term memory neural network respectively to obtain likelihood function values;

forming a sequence of likelihood function values based on the likelihood function values for each of the sequences of symbol sets.

7. The method for symbol detection of a highly dynamic channel as claimed in claim 6, wherein said likelihood function value is calculated by the formula:

wherein P is the likelihood function value, N_bFor the preset number of symbols, the p_nAnd estimating the probability of the sampling sequence group.

8. A symbol detection apparatus for a high dynamic channel, the symbol detection apparatus for the high dynamic channel comprising:

the acquisition module is used for acquiring a plurality of symbol set sequences according to a preset modulation order;

the down-sampling module is used for performing down-sampling and filtering processing on the symbols to be detected in each symbol set sequence to obtain a sampling sequence corresponding to the symbols to be detected;

the input module is used for inputting the sampling sequence into a preset long-short term memory neural network to obtain a likelihood function value sequence;

and the detection module is used for detecting the symbol set sequence according to the likelihood function value sequence.

9. A symbol detection device for a high dynamic channel, characterized in that the symbol detection device for a high dynamic channel comprises a memory, a processor and a symbol detection method program for a high dynamic channel stored on the memory and operable on the processor, which when executed by the processor implements the steps of the symbol detection method for a high dynamic channel according to any one of claims 1 to 7.

10. A storage medium, characterized in that the storage medium is a computer readable storage medium, on which a symbol detection method program of a high dynamic channel is stored, which when executed by a processor implements the steps of the symbol detection method of a high dynamic channel according to any one of claims 1 to 7.

Technical Field

The present invention relates to the field of communications, and in particular, to a method, an apparatus, a device, and a storage medium for symbol detection of a high dynamic channel.

Background

The high dynamic channel refers to that under some complex dynamic scenes, due to the dynamic change of the environment, the electromagnetic wave is changed violently in the process of propagation and is reflected on the channel, namely, the high dynamic characteristic of the channel is shown. High dynamic channels typically exhibit non-stationary characteristics and strongly time-varying characteristics. The high dynamic channel has non-stationarity and strong time-varying characteristics, so that the calculation is very complex, and because the number of sampling points of the high dynamic channel symbol is huge, it is more difficult to accurately establish a model of the channel, so that the symbol detection under the high dynamic channel is also quite difficult, and a new detection method of the high dynamic channel needs to be provided to overcome the problems.

Disclosure of Invention

The invention mainly aims to provide a symbol detection method, a device, equipment and a storage medium of a high dynamic channel, aiming at reducing the calculation complexity in the channel symbol detection process, wherein the symbol detection method of the high dynamic channel comprises the following steps:

obtaining a plurality of symbol set sequences according to a preset modulation order;

performing down-sampling and filtering processing on the symbols to be detected in each symbol set sequence to obtain sampling sequences corresponding to the symbols to be detected;

inputting the sampling sequence into a preset long-short term memory neural network to obtain a likelihood function value sequence;

and detecting the symbol set sequence according to the likelihood function value sequence.

In an embodiment, the down-sampling and filtering the symbol to be detected in each symbol set sequence to obtain a sampling sequence corresponding to the symbol to be detected includes:

calculating the number Q of orthogonal bases used for filtering;

acquiring the Q sampling points of the symbol to be detected;

and constructing a filtering formula according to the number of the orthogonal bases, and substituting the Q sampling points into the filtering formula to obtain a sampling sequence corresponding to the symbol to be detected.

In one embodiment, Q is calculated as:

Q＝2[f_dTT_s]+1, wherein, said f_dIs a frequency extension parameter, T is the number of sampling points of the symbol to be detected, T_sIs the sampling time interval.

In one embodiment, the filtering formula is: r ≈ gx + n, where r represents a sampling sequence, theg represents a high dynamic channel, and the calculation formula of the g is as follows:said u is_kRepresents an orthogonal base, said q_kIs represented by the formula_kThe corresponding coefficients.

In one embodiment, u is a halogen atom_kThe calculation formula of (2) is as follows:

Cu_k＝λ_ku_kwherein C is a matrix of Q × Q, and the elements of CWherein, i and j respectively represent the ith row and the jth column of the matrix C, i is more than or equal to 1, and Q is more than or equal to j.

In one embodiment, the step of inputting the sample sequence into a preset long-short term memory neural network to obtain the sequence of likelihood function values includes:

carrying out real-number conversion on the sampling sequence, and arranging the sampling sequence after the real-number conversion according to the sequence of receiving the symbol time to be detected corresponding to the sampling sequence;

according to the sequence of the time, taking the first sampling sequence as a starting point, and sequentially arranging the sampling sequences with the adjacent preset symbol number as a group until obtaining a sampling sequence group with the group number as the preset symbol number;

inputting all the sampling sequence groups of each symbol set sequence into a preset long-short term memory neural network respectively to obtain likelihood function values;

forming a sequence of likelihood function values based on the likelihood function values for each of the sequences of symbol sets.

In one embodiment, the likelihood function values are calculated by the formula:

wherein P is the likelihood function value, N_bFor the preset number of symbols, the p_nAnd estimating the probability of the sampling sequence group.

In addition, to achieve the above object, the present invention provides a symbol detection apparatus for a high dynamic channel, including:

the acquisition module is used for acquiring a plurality of symbol set sequences according to a preset modulation order;

the down-sampling module is used for performing down-sampling and filtering processing on the symbols to be detected in each symbol set sequence to obtain a sampling sequence corresponding to the symbols to be detected;

the input module is used for inputting the sampling sequence into a preset long-short term memory neural network to obtain a likelihood function value sequence;

and the detection module is used for detecting the symbol set sequence according to the likelihood function value sequence.

In addition, to achieve the above object, the present invention further provides a symbol detection device for a high dynamic channel, which includes a memory, a processor, and a symbol detection method program for a high dynamic channel stored in the memory and operable on the processor, wherein the symbol detection method program for a high dynamic channel, when executed by the processor, implements the steps of the symbol detection method for a high dynamic channel as described above.

Furthermore, to achieve the above object, the present invention further provides a computer readable storage medium, on which the symbol detection method program of the high dynamic channel is stored, and when executed by a processor, the symbol detection method program of the high dynamic channel implements the steps of the symbol detection method of the high dynamic channel as described above.

According to the invention, a plurality of symbol set sequences are obtained according to a preset modulation order, the symbol to be detected in each symbol set sequence is subjected to down-sampling and filtering processing to obtain a sampling sequence corresponding to the symbol to be detected, the calculation amount of a neural network is greatly reduced, the sampling sequence is input into a preset long-short term memory neural network to obtain a likelihood function value sequence, the preset long-short term memory neural network is more consistent with the time storage characteristic of a high dynamic channel, the symbol set sequences are detected according to the likelihood function value sequence, the sampling sequence is used as the input of the neural network instead of the conventional communication symbol as the minimum processing unit, and the difficulty that the channel changes in a symbol period under a high-pass channel is solved.

Drawings

FIG. 1 is a diagram of a hardware architecture of a device implementing an embodiment of the invention;

FIG. 2 is a flow chart of an embodiment of a symbol detection method for high dynamic channel according to the present invention;

FIG. 3 is a schematic representation of the high dynamic channel of the present invention;

FIG. 4 is a graph comparing the performance of the symbol detection method for high dynamic channel according to the present invention;

FIG. 5 is a schematic diagram of a sample sequence input method according to the present invention;

FIG. 6 is a diagram of a long term memory neural network element detector architecture according to the present invention;

fig. 7 is a schematic diagram of a symbol detection method based on a long-term and short-term memory neural network.

The implementation, functional features and advantages of the present invention will be described with reference to the accompanying drawings.

Detailed Description

It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The invention provides a symbol detection device of a high dynamic channel, and referring to fig. 1, fig. 1 is a schematic structural diagram of a hardware operating environment according to an embodiment of the present invention.

It should be noted that fig. 1 is a schematic structural diagram of a hardware operating environment of a symbol detection apparatus which can be a high dynamic channel. The symbol detection device of the high dynamic channel in the embodiment of the present invention may be a PC (Personal Computer), a portable Computer, a server, or the like.

As shown in fig. 1, the symbol detection apparatus for a high dynamic channel may include: a processor 1001, such as a CPU, a memory 1005, a user interface 1003, a network interface 1004, a communication bus 1002. Wherein a communication bus 1002 is used to enable connective communication between these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1005 may be a high-speed RAM memory or a non-volatile memory (e.g., a magnetic disk memory). The memory 1005 may alternatively be a storage device separate from the processor 1001.

Optionally, the symbol detection device of the high dynamic channel may further include an RF (Radio Frequency) circuit, a sensor, a WiFi module, and the like.

Those skilled in the art will appreciate that the high dynamic channel symbol detection device architecture shown in fig. 1 does not constitute a limitation of a high dynamic channel symbol detection device and may include more or fewer components than shown, or some components may be combined, or a different arrangement of components.

As shown in fig. 1, a memory 1005, which is a computer storage readable storage medium, may include therein an operating system, a network communication module, a user interface module, and a symbol detection method program of a high dynamic channel. Among them, the operating system is a program that manages and controls hardware and software resources of the symbol detection device for the high dynamic channel, a symbol detection method program that supports the high dynamic channel, and the execution of other software or programs.

The symbol detection apparatus for a high dynamic channel shown in fig. 1 may be used to reduce the complexity of channel symbol detection, and the user interface 1003 is mainly used to detect a symbol to be detected or output various information such as a likelihood function value; the network interface 1004 is mainly used for interacting with a background server and communicating; the processor 1001 may be configured to invoke a symbol detection method program for a high dynamic channel stored in the memory 1005 and perform the following operations:

obtaining a plurality of symbol set sequences according to a preset modulation order;

performing down-sampling and filtering processing on the symbols to be detected in each symbol set sequence to obtain sampling sequences corresponding to the symbols to be detected;

inputting the sampling sequence into a preset long-short term memory neural network to obtain a likelihood function value sequence;

and detecting the symbol set sequence according to the likelihood function value sequence.

According to the invention, a plurality of symbol set sequences are obtained according to a preset modulation order, the symbol to be detected in each symbol set sequence is subjected to down-sampling and filtering processing to obtain a sampling sequence corresponding to the symbol to be detected, the calculation amount of a neural network is greatly reduced, the sampling sequence is input into a preset long-short term memory neural network to obtain a likelihood function value sequence, the preset long-short term memory neural network is more consistent with the time storage characteristic of a high dynamic channel, the symbol set sequences are detected according to the likelihood function value sequence, the sampling sequence is used as the input of the neural network instead of the conventional communication symbol as the minimum processing unit, and the difficulty that the channel changes in a symbol period under a high-pass channel is solved.

The specific implementation of the mobile terminal of the present invention is substantially the same as the following embodiments of the symbol detection method for a high dynamic channel, and is not described herein again.

Based on the above structure, an embodiment of the symbol detection method for a high dynamic channel of the present invention is provided.

The invention provides a symbol detection method of a high dynamic channel.

Referring to fig. 2, fig. 2 is a flowchart illustrating an embodiment of a symbol detection method for a high dynamic channel according to the present invention.

In the present embodiment, an embodiment of a symbol detection method for a high dynamic channel is provided, and it should be noted that, although a logical sequence is shown in the flowchart, in some cases, the steps shown or described may be performed in a different sequence from that here, and the method of the symbol detection method for a high dynamic channel provided in the present embodiment may be applied to an intelligent terminal.

In this embodiment, the symbol detection method for a high dynamic channel includes:

step S10, obtaining a plurality of symbol set sequences according to a preset modulation order;

the preset modulation order is a device parameter, and is 16 if Quadrature amplitude modulation is selected, and is 4 if Quadrature Phase Shift Keying (QPSK) is selected, for example, when the preset modulation order is M, a set of a plurality of symbol set sequences is represented by S, and each symbol set sequence is represented by S, then S ═ S { (S) } S₁,s₂,…,s_M}. And acquiring the symbol set sequence through the receiving end.

Step S20, performing down-sampling and filtering processing on the symbols to be detected in each symbol set sequence to obtain a sampling sequence corresponding to the symbols to be detected;

each symbol set sequence comprises a plurality of symbols to be detected, each symbol to be detected comprises a plurality of sampling points, the symbols to be detected are subjected to down sampling and filtering processing, and the symbols to be detected are converted into data capable of being subjected to symbol detection.

In some embodiments, step S20 includes:

step a, calculating the number Q of orthogonal bases for filtering;

b, acquiring the Q sampling points of the symbol to be detected;

and c, constructing a filtering formula according to the number of the orthogonal bases, and substituting the Q sampling points into the filtering formula to obtain a sampling sequence corresponding to the symbol to be detected.

The method comprises the steps of carrying out down-sampling and filtering processing on a symbol to be detected, firstly calculating the number of sampling points after carrying out down-sampling processing on the symbol to be detected, and assuming that the number of the sampling points before carrying out down-sampling on the symbol to be detected is T, then the number Q after carrying out down-sampling processing on the symbol to be detected is 2[ f_dTT_s]+1, wherein, f_dFor frequency extension parameters, T is the number of sampling points of the symbol to be detected, T_sFor the sampling time interval, the general sampling points can be represented by time, and the Q sampling points are selected from the T sampling points according to the time sequence. It can be understood that T>>And Q. Then filtering the symbol to be detected for filteringThe number of orthogonal bases is the same as the number of sampling points of the symbol to be detected after down-sampling, a filtering formula is constructed according to the number of the orthogonal bases, the filtering formula is r ≈ gx + n, wherein r represents a sampling sequence, g represents a high dynamic channel, and the calculation formula of g is as follows:

u_krepresents an orthogonal base, q_kIs represented by the formula_kCorresponding coefficients, fig. 3 is a diagram illustrating a highly dynamic channel using orthogonal bases. Substituting Q sampling points of each symbol to be detected into a filtering formula to obtain a sampling sequence r of each symbol to be detected, such as a symbol set sequence s₁The number of symbols to be detected contained is N, then the sequence s of symbol sets₁The corresponding sample sequence includes r₁,r₂,…r_N。

In some embodiments, u_kThe calculation formula of (2) is as follows: cu_k＝λ_ku_kWherein C is a matrix of Q × Q, and the elements of CWherein, i and j respectively represent the ith row and the jth column of the matrix C, i is more than or equal to 1, and Q is more than or equal to j.

Step S30, inputting the sampling sequence into a preset long-short term memory neural network to obtain a likelihood function value sequence;

and performing down-sampling and filtering processing on the symbol to be detected, designing input data of a neural network, and greatly reducing the calculated amount of the neural network.

The preset long-short term memory neural network is a multilayer neural network constructed by adopting long-short term memory neural units as basic units of the symbol detector, and the cell state c in the long-short term memory neural network units_tAnd hidden state h_tThe time memory characteristics of the high dynamic channel can be better learned when the next time step is transmitted to the long and short term memory network unit, and fig. 4 is a performance comparison graph of a symbol detection method based on an orthogonal basis and the long and short term memory neural network. When long and shortThe memory neural network has a data driving characteristic, and the long-time memory neural network-based symbol detection method does not depend on a specific channel model, so that the difficulty that the channel model is difficult to accurately obtain under a high dynamic channel is solved.

And inputting the sampling sequences corresponding to the symbols to be detected in each symbol set sequence into a preset long-short term memory neural network to obtain the likelihood function value of each symbol set sequence, and further forming a likelihood function value sequence.

In some embodiments, step S30 further includes:

d, carrying out real-number conversion on the sampling sequence, and arranging the sampling sequence after the real-number conversion according to the sequence of receiving the symbol time to be detected corresponding to the sampling sequence;

it will be understood that the resulting sample sequence r belongs to a complex set, r has a length Q, and the filtered output is further real-valued, i.e. the complex sequence is converted into a new sequence of real and imaginary parts, and the real-valued sample sequence is y ═ f_R(r)＝[R(r),U(r)]And the length is 2Q. The symbols to be detected are received in chronological order, and the sampling sequences after the quantization, for example the sequence s of the symbol set, are arranged in chronological order according to which symbols to be detected are received₁The number of symbols to be detected contained is N, then the sequence s of symbol sets₁The corresponding sampling sequence after the real quantization is y_1,t1,y_1,t2,…y_1,tNIf the time sequence of receiving the N symbols to be detected is t1, t1, … and tN, the arranged sampling sequence after real number is regarded as y_1,t1,y_1,t2,…y_1,tN。

Step e, according to the sequence of the time, taking the first sampling sequence as a starting point, and sequentially arranging the sampling sequences with the adjacent preset symbol number as a group until obtaining a sampling sequence group with the group number as the preset symbol number;

the number of the predetermined symbols is determined by the receiving end, and referring to fig. 5, the number of the predetermined symbols is Nb, and if the real number is real, the sampling sequence is y_1,t1,y_1,t2,…y_1,tNThen fromFirst bit y_1,t1Taking Nb-1 sampling sequences as a starting point, counting backwards according to the time sequence to obtain a first sampling sequence group, and sequentially counting backwards from y_1,t2Counting Nb sampling sequences according to time sequence to obtain a second sampling sequence group, and repeating the steps until the last Nb sampling sequence group is y_1,tNb,y_1,t(Nb+1),…,y_1,t(2Nb-1). Each symbol set sequence obtains Nb sample sequence groups in this manner.

Step f, inputting all the sampling sequence groups of each symbol set sequence into a preset long-short term memory neural network respectively to obtain likelihood function values;

the preset long-short term memory neural network can only process the sampling sequences corresponding to the Nb symbols to be detected each time, so that only Nb sampling sequences are input each time, specifically, the input can be performed according to the sampling sequence group of each symbol set sequence, for example, firstly, the symbol set sequence s is input₁The first sampling sequence set, then the second, the third, etc., the likelihood function value of each symbol set sequence is calculated by presetting the long-short term memory neural network, the symbol set sequence s₁The likelihood function values of (a) can also be understood as a sequence of symbol sets s₁And the likelihood function value corresponding to the Nth symbol to be detected is obtained.

In some embodiments, the likelihood function values are calculated by the formula:wherein P is the likelihood function value, N_bFor the preset number of symbols, the p_nAnd estimating the probability of the sampling sequence group.

Similarly, the calculation formula of the likelihood function value is explained with reference to fig. 5. P in FIG. 5₁For a set of sampling sequences y_1,t1,y_1,t2,…y_1,tNIs estimated from the probability of P₂For a set of sampling sequences y_2,t2,y_2,t3,…y_2,t(Nb+1)Is estimated from the probability of P, then P_nFor a set of sampling sequences y_1,tn,y_1,t(n+1),…y_1,t(n+Nb-1)Is estimated by a probability estimator, a symbolSet sequence s₁The value of the likelihood function of (c) is the average of the Nb probability estimates obtained.

And g, forming a likelihood function value sequence based on the likelihood function values of each symbol set sequence. The sequence of likelihood function values being formed by the likelihood function values of each sequence of symbol sets, i.e. the sequence of likelihood function values

And step S40, detecting the symbol set sequence according to the likelihood function value sequence.

Specifically, a current symbol to be detected is represented by a time point t, and a sampling sequence corresponding to the symbol to be detected t is y_t. Hidden state h_tRepresenting the current state and saving short-term memory of the long-term memory network. In addition, cell status c_tRepresenting long-term memory of the memory network. h is_tAnd c_tAre all N_hThe x 1 vector and is passed to the next neural unit at the next point in time. The long and short term memory neural unit in fig. 6 includes 3 control thresholds: forget door j_tInput door i_tAnd an output gate o_t。j_t＝σ(W_j·[h_t-1,y_t]+b_j)；i_t＝σ(W_i·[h_t-1,y_t]+b_i)；o_t＝σ(W_o·[h_t-1,y_t]+b_o) Wherein [ h ]_t-1,y_t]Is one (N)_hX 2Q) x 1 vector containing hidden state h_t-1And input data y_t。W_j、W_iAnd W_oIs a weight of 3 thresholds, N_h×(N_h+2Q) matrix. In addition, b_j、b_iAnd b_oFor respective deviations, they are N_hX 1 vector. The gate activation function σ is a point-by-point logic sigmoid function:

for the time point t of memorizing the neural network unit in a long and short timeState of cellsComprises the following steps:wherein W_cIs N_h×(N_h+ 2Q). b_cIs N_hX 1 and the activation function tanh is a point-wise hyperbolic tangent function. Then, cell state c_tThe device consists of two parts: cell status at the last time point c_t-1And transient cell state Forget door j_tThe portion of the cell state that was passed from the last time step is determined. In addition, an input gate i_tThe provisional cell state calculated at this time step is determined. Thus, hidden state h_tFrom the cell state c_tAnd an output gate o_tDetermination of h_t＝o_t⊙tanh(c_t) Then, cell state c_tAnd hidden state h_tThe network element will be remembered at the next time step transmission to the duration.

The deep neural network symbol detector employs a feed-forward deep neural network of the L +2 layer, as shown in fig. 7. The detector has an input layer, L LSTM layers and an output layer. The input layer has N_bA symbol as input data, each symbol y_tCorresponding to a 2Q × 1 sample sequence. The LSTM layer thus has N_bAn LSTM cell as shown in fig. 7. For the first LSTM layer, y_tIs the input data for each LSTM cell. In addition, for the L (2. ltoreq. L. ltoreq.L) th LSTM layer,is the input data for each LSTM cell. Hidden stateAnd cell statusWill be transmitted to the next LSTM unit at the same LSTM layer. The output layer is a connection layer with softmax activation function. Output signal P_tComprises the following steps:wherein W_pIs MXN_hThe weights of the matrix. b_pIs a deviation value of mx 1, M being the modulation order of the signal. Activation function f_s(x):R^M→R^MIs the softmax activation function and its i-th element f_s(x)]_iE (0,1) is:it should be noted that the softmax function outputs a data vector P_tIs the symbol y_tWhich can be transmitted directly to the soft channel decoder at the receiving end. In particular, in our deep neural network-based monitoring scheme, the input data stream of the received signal is slid by one symbol period within one time step and returned by N_bThe output of each symbol is soft estimated data. In fact, for the symbol y_tThe deep nerve detector may have N_bAnd returning the soft estimation result. Suppose P_t,k(1≤k≤N_b) Is the symbol y_tSoft estimate output of y_tIs an input data stream (y)_t-Nb+k,…,y_t-1,y_t,y_t+1,…y_t+k-1)，

In the deep neural network-based detector detection process, d is assumed_t∈R^MIs the feature vector of the symbol t to be detected,

wherein 1 is in the symbol setThe feature function of the symbol to be detected means that the element is 1 in agreement with the symbol to be detected and the remaining elements are 0_tWill be used in the detection process and compared to the sequence of soft estimate output vector likelihood function values.

According to the embodiment, a plurality of symbol set sequences are obtained according to a preset modulation order, the symbol to be detected in each symbol set sequence is subjected to down-sampling and filtering processing to obtain a sampling sequence corresponding to the symbol to be detected, the calculated amount of a neural network is greatly reduced, the sampling sequence is input into a preset long-short term memory neural network to obtain a likelihood function value sequence, the preset long-short term memory neural network is more consistent with the time storage characteristic of a high dynamic channel, the symbol set sequences are detected according to the likelihood function value sequence, the sampling sequence is used as the input of the neural network instead of the conventional communication symbol as the minimum processing unit, and the difficulty that the channel changes in a symbol period under a high-pass channel is solved.

In addition, an embodiment of the present invention further provides a symbol detection apparatus for a high dynamic channel, where the symbol detection apparatus for a high dynamic channel includes:

the acquisition module is used for acquiring a plurality of symbol set sequences according to a preset modulation order;

the down-sampling module is used for performing down-sampling and filtering processing on the symbols to be detected in each symbol set sequence to obtain a sampling sequence corresponding to the symbols to be detected;

the input module is used for inputting the sampling sequence into a preset long-short term memory neural network to obtain a likelihood function value sequence;

and the detection module is used for detecting the symbol set sequence according to the likelihood function value sequence.

The embodiments of the symbol detection apparatus for a high dynamic channel according to the present invention are substantially the same as the embodiments of the symbol detection method for a high dynamic channel, and are not described herein again.

Furthermore, an embodiment of the present invention further provides a computer-readable storage medium, where a symbol detection method program of a high dynamic channel is stored, and when executed by a processor, the symbol detection method program of the high dynamic channel implements the steps of the symbol detection method of the high dynamic channel as described above.

It should be noted that the computer readable storage medium can be disposed in a symbol detection device of a high dynamic channel.

The specific implementation manner of the computer-readable storage medium of the present invention is substantially the same as that of the above embodiments of the symbol detection method for a high dynamic channel, and is not described herein again.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.

Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal device (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.

The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.