CN115834306A - Method for directly estimating multi-communication signal symbol sequence under interference condition - Google Patents

Method for directly estimating multi-communication signal symbol sequence under interference condition Download PDF

Info

Publication number
CN115834306A
CN115834306A CN202211377436.3A CN202211377436A CN115834306A CN 115834306 A CN115834306 A CN 115834306A CN 202211377436 A CN202211377436 A CN 202211377436A CN 115834306 A CN115834306 A CN 115834306A
Authority
CN
China
Prior art keywords
symbol sequence
signal
symbol
sequence
estimation
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202211377436.3A
Other languages
Chinese (zh)
Other versions
CN115834306B (en
Inventor
黄知涛
邓文
王翔
柯达
赵雨睿
陈颖
李保国
王丰华
蔡昕
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
National University of Defense Technology
Original Assignee
National University of Defense Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by National University of Defense Technology filed Critical National University of Defense Technology
Priority to CN202211377436.3A priority Critical patent/CN115834306B/en
Publication of CN115834306A publication Critical patent/CN115834306A/en
Application granted granted Critical
Publication of CN115834306B publication Critical patent/CN115834306B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Landscapes

  • Digital Transmission Methods That Use Modulated Carrier Waves (AREA)

Abstract

The invention discloses a method for directly estimating a multi-communication signal symbol sequence under an interference condition, which comprises the steps of extracting parallel symbol sequence estimation characteristics FPSSE based on single-channel observation to obtain characteristic sequences respectively containing target signal symbol information; and carrying out sequence marking based on the cyclic neural network after offline training to obtain a symbol sequence estimation result, wherein the characteristic sequence is used as input when the cyclic neural network is subjected to offline training. The method adopts the idea of parallel independent estimation of the multi-signal symbol sequence instead of joint estimation of the multi-signal symbol sequence, so that the calculation complexity of the method does not increase with the target signal number exponentially any more. Meanwhile, the method can be suitable for scenes with the same or different target signal symbol rates, and the practicability is obviously improved. Because the interference among the target signals is suppressed by the FPSSE extraction, the deep convolutional network demodulator with the estimation performance has obvious advantages under the condition that the power difference of the target signals is obvious.

Description

Method for directly estimating multi-communication signal symbol sequence under interference condition
Technical Field
The invention belongs to the technical field of digital signal processing, and particularly relates to a method for directly estimating a multi-communication signal symbol sequence under an interference condition.
Background
In current cooperative/non-cooperative communication, single channel receiving devices are widely adopted due to cost, volume and other aspects, which also makes the time-frequency aliasing digital communication signal problem often faced in complex electromagnetic environments more challenging. In order to realize signal analysis and information extraction based on single-channel aliasing observation, a target signal must be separated from the signal. The Blind Signal Separation (BSS) technique is intended to separate and acquire the signals of interest contained therein based on aliasing observation only, wherein the single-channel BSS technique is further intended to achieve the purpose based on single-channel observation. Therefore, the research on the single-channel BSS technology has important theoretical and practical significance for improving the capability of the existing cooperative/non-cooperative communication system for dealing with time-frequency aliasing digital communication signals.
The estimation of the multi-signal symbol sequence is a special single-channel blind signal separation BSS task when a target signal is a digital communication signal, and the existing single-channel BSS algorithm for realizing the estimation of the multi-signal symbol sequence mainly has the following problems: firstly, due to the adoption of a multi-signal symbol sequence joint estimation mode, the calculation amount of the mainstream model-based algorithm based on the technologies such as particle filter PF and survival path-by-survival path processing PSP is increased along with the target signal number and the target signal modulation order index, and particularly, the calculation amount of the algorithm based on the particle filter PF is too high. Secondly, the survivor path-by-survivor path processing PSP as a class of mainstream algorithm can only be applied to the problem scene that the symbol rate of each target signal is equal at present, and the practicability is limited. Thirdly, for the existing data-driven algorithms represented by the deep convolutional network demodulator DCND, the adaptive capacity to the power level difference of each target signal is not strong because effective preprocessing is not performed, and the existing data-driven algorithms are easily interfered by other signals when each signal symbol sequence is estimated.
Disclosure of Invention
The invention aims to provide a method for directly estimating a multi-communication signal symbol sequence under an interference condition, which solves the problems of higher computational complexity and insufficient adaptability to target signal parameters of the existing multi-signal symbol sequence estimation algorithm based on a model and adopting joint estimation.
In order to achieve the above object, the present invention provides a method for directly estimating a symbol sequence of multiple communication signals under an interference condition, including:
extracting parallel symbol sequence estimation characteristics FPSSE based on single-channel observation to obtain characteristic sequences respectively containing symbol information of each target signal;
and carrying out sequence marking based on the cyclic neural network after offline training to obtain a symbol sequence estimation result, wherein the characteristic sequence is used as input when the cyclic neural network is subjected to offline training.
Optionally, before extracting the symbol sequence estimation feature, a single-channel aliasing observation needs to be constructed, which specifically includes:
assuming that a single antenna receives I spectrally overlapping digital communication signals simultaneously, this single-channel aliasing observation can be expressed as:
Figure BDA0003926400850000021
wherein s is i (n-m i ) Representing a transmission delay of m i I target signal of one sampling interval, a i V (N) represents AWGN for its corresponding channel gain, N is the total number of sampling points;
for a typical amplitude/phase modulated digital communication signal, s i (n) furtherIs shown as
Figure BDA0003926400850000022
Wherein E is the signal power;
Figure BDA0003926400850000031
corresponding to the q-th of transmission i A symbol, Q i Is the total number of symbols; g i The expression comprehensively considers the equivalent pulse shaping function of the pulse shaping of the sending end and the physical transmission channel, T s And
Figure BDA0003926400850000032
respectively, a sampling interval and a symbol oversampling rate;
Figure BDA0003926400850000033
is the residual carrier frequency.
Optionally, the extracting the parallel symbol sequence estimation feature FPSSE based on single-channel observation includes:
p single-channel aliasing observations are constructed according to the method (1), and the modulation pattern of the I digital communication signals contained in each aliasing observation is selected from a set omega with the size of C;
for the p (p E [1, P ]) aliasing observation, I FPSSE subsequences are extracted from the aliasing observation, and each subsequence and the known symbol sequence of the corresponding signal can form a training sample.
Optionally, extracting the FPSSE subsequence comprises: suppose g i (.) has an FIR structure and is derived from
Figure BDA0003926400850000034
Continue until
Figure BDA0003926400850000035
Therefore, is received i Q of (n) i The sample point range of each symbol influence is:
Figure BDA0003926400850000036
for the first
Figure BDA0003926400850000037
The number of the symbols is one,
Figure BDA0003926400850000038
is provided with
Figure BDA0003926400850000039
Figure BDA00039264008500000310
When in use
Figure BDA00039264008500000311
When s is i The first order difference sequence of (n) is represented as:
Figure BDA0003926400850000041
assuming the sampling is sufficient
Figure BDA0003926400850000042
And g is i (. Cndot.) is a symmetric function, in which case:
Figure BDA0003926400850000043
Figure BDA00039264008500000412
and
Figure BDA0003926400850000044
equation (14) is then approximated as:
Figure BDA0003926400850000045
when in use
Figure BDA0003926400850000046
When d is greater than i (n) is
Figure BDA0003926400850000047
Combining equations (1), (15) and (16), the first order difference of x (n) is:
Figure BDA0003926400850000048
wherein phi i Represents the set of the following sample points:
Figure BDA0003926400850000049
optionally, extracting the FPSSE subsequence further comprises: for a target signal s i (n) subsequences thereof
Figure BDA00039264008500000410
From { d (n) | n ∈ [1, N ]]Extract as follows:
Figure BDA00039264008500000411
after being divided according to each signal, { d (n) | n ∈ [1, N]Will be divided into a set of sub-sequences
Figure BDA0003926400850000051
Wherein each subsequence corresponds to a target signal.
Optionally, extracting the FPSSE subsequence further comprises:
for I e [1I ]]Is further based on
Figure BDA0003926400850000052
The following normalized amplitude and phase sequences, denoted respectively, were calculated
Figure BDA0003926400850000053
Figure BDA0003926400850000054
Figure BDA0003926400850000055
Wherein, | - | represents the amplitude of the complex sequence,
Figure BDA0003926400850000056
to represent
Figure BDA0003926400850000057
Is the mean value of the complex number, k i ∈[1，Q i -2L i +1]Will be
Figure BDA0003926400850000058
As signal s in the received signal i (n) the corresponding FPSSE sequence.
Optionally, in the off-line training of the recurrent neural network, in order to demodulate signals with possibly different modulation patterns, a recurrent neural network set is constructed, the structure and parameters of each recurrent neural network in the set are the same, training is performed for symbol sequence estimation of target signals with different modulation patterns, and each recurrent neural network includes two BLSTM layers and one fully-connected classification layer.
Optionally, the output dimension of the classification layer is determined by the debugging order of the signal to which the classification layer is oriented.
The invention has the technical effects that: the method constructs a single-input-multiple-output symbol sequence estimator capable of carrying out parallel estimation on a multi-signal symbol sequence by modeling symbol sequence estimation as a sequence labeling problem and realizing sequence labeling based on a recurrent neural network. Different from direct single-channel aliasing observation as input of DCND, the method firstly carries out parallel symbol sequence estimation feature extraction based on single-channel observation, and mainly comprises difference operation and specific time extraction so as to obtain feature sequences respectively containing symbol information of each target signal. In off-line training of the RNN-based symbol sequence estimator, the characteristic sequence is used as a network input for estimating each target signal symbol sequence. Different from the existing multi-signal symbol sequence estimation algorithm based on a model, the method adopts the idea of parallel independent estimation of the multi-signal symbol sequence instead of the joint estimation of the multi-signal symbol sequence, so that the calculation complexity of the method does not increase along with the target signal number index any more. Meanwhile, the method breaks through the limitation of the PSP algorithm on the target signal symbol rate relation, can be suitable for scenes with the same or different target signal symbol rates, and obviously improves the practicability. Because the interference among the target signals is suppressed by the FPSSE extraction, the estimation performance has a significant advantage compared with the DCND under the condition that the power difference of the target signals is significant. Simulation experiment results show that the symbol sequence estimator provided by the method has the capability of simultaneously and accurately estimating a plurality of target signal symbol sequences under different signal power ratios, and has good adaptability to different target signals with large power level differences. In a generalization ability test, the data driving method provided by the method still shows good estimation performance under the test condition which is not seen in training.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the application and, together with the description, serve to explain the application and are not intended to limit the application. In the drawings:
fig. 1 is a schematic flowchart of a method for directly estimating a symbol sequence of multiple communication signals under an interference condition according to an embodiment of the present invention;
FIG. 2 is a RNN-based multi-signal symbol sequence estimator provided by an embodiment of the present invention;
fig. 3 is a diagram of estimated BER of parallel symbol sequences of two target signals under different SINRs according to an embodiment of the present invention;
fig. 4 is a graph comparing BER of different SIMO symbol sequence estimation algorithms according to an embodiment of the present invention.
Fig. 5 is a generalized performance test chart of the symbol sequence estimator according to the embodiment of the present invention.
Detailed Description
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present application will be described in detail below with reference to the embodiments with reference to the attached drawings.
As shown in fig. 1, the present embodiment provides a method for directly estimating a multi-communication signal symbol sequence under an interference condition, including the following steps:
s1, extracting parallel symbol sequence estimation characteristics FPSSE based on single-channel observation to obtain characteristic sequences respectively containing symbol information of each target signal;
and S2, carrying out sequence labeling based on the cyclic neural network after off-line training to obtain a symbol sequence estimation result, wherein the characteristic sequence is used as input when the cyclic neural network is off-line trained.
The recurrent neural network in this embodiment adopts a bidirectional RNN, i.e., a BRNN structure, integrated with an LSTM structure to form a BLSTM structure, each hidden layer includes paired forward and backward LSTM nodes, and adjacent layers are connected by a tanh activation function. The following equation describes the operation performed by the forward LSTM node in the hidden layer at each instant:
Figure BDA0003926400850000071
Figure BDA0003926400850000081
Figure BDA0003926400850000082
Figure BDA0003926400850000083
s t ＝g t ⊙i t +s t-1 ⊙f t (9)
Figure BDA0003926400850000084
wherein the content of the first and second substances,
Figure BDA0003926400850000085
and
Figure BDA0003926400850000086
respectively representing the output of each forward node at t-1 and t moments; g t ，i t ，f t ，o t Activation function values corresponding to an input node, an input gate, a forgetting gate and an output gate at the time t respectively; s t-1 And s t The states of the memory cells at t-1 and t, respectively; w gx ，W ix ，W fx ，W ox Respectively inputting traditional weight matrixes between each corresponding unit and each corresponding unit; w gh ，W ih ，W fh ，W oh Respectively are a 'circulation' weight matrix between the forward hidden layer node and the self at the adjacent time; b g ，b i ，b f ，b o Biasing vectors for each cell; phi (·) and phi indicate a tanh function and a point multiplication operation, respectively. Here, the vector represents the values of the corresponding portions of all the forward nodes in the entire hidden layer. For example, s is the state value of all the forward node memory cells in the entire hidden layer.
While the remaining calculations are similar for the backward hidden layer nodes of BLSTM, except for the direction reversal of the cyclic concatenation. Specifically, to calculate the output of each backward node at time t, it is necessary to calculate the cell state based on the input at time t and the backward node at time t + 1. The same pair of BLSTM hidden layers share inputs, and the outputs of the two layers are integrated and sent to the next network layer.
Time back propagation technique BPTT or its variant form, i.e. truncated time back propagation technique TBPTT, is used. Unlike standard back propagation techniques, in training based on BPTT, a single RNN layer is assumed to be spread out in time, which is treated as a multi-layer network sharing parameters between different time steps, and network parameters are optimized in two directions while considering both vertical propagation of gradients between adjacent network layers and horizontal propagation over adjacent time steps. In TBPTT, however, the number of time steps for back propagation is limited to a certain extent in order to improve the processing efficiency for longer sequences. By including training for a sufficient number of iterations, the network training is considered complete when the loss function value on the training set is below a certain threshold or the error rate of the sequence annotation is below a certain level. In the online test of sequence marking, the input sequence is directly sent to the trained network, and the network output is the sequence marking result of the current test sample.
In the further optimization scheme, before extracting the symbol sequence estimation features, a single-channel aliasing observation needs to be constructed:
assuming that a single antenna receives I spectrally overlapping digital communication signals simultaneously, this single-channel aliasing observation can be expressed as:
Figure BDA0003926400850000091
wherein s is i (n-m i ) Representing a transmission delay of m i I target signal of one sampling interval, a i V (N) represents AWGN for its corresponding channel gain, N is the total number of sampling points;
for typical amplitude/phase modulated digital communication signals, s i (n) is further represented as
Figure BDA0003926400850000092
Wherein E is the signal power;
Figure BDA0003926400850000101
q-th corresponding to transmission i A symbol, Q i Is the total number of symbols; g i The expression comprehensively considers the equivalent pulse shaping function of the pulse shaping of the sending end and the physical transmission channel, T s And
Figure BDA0003926400850000102
respectively, a sampling interval and a symbol oversampling rate;
Figure BDA0003926400850000103
is the residual carrier frequency.
Further optimizing the scheme, the parallel symbol sequence estimation feature FPSSE extraction based on single-channel observation comprises the following steps: in order to construct a training sample set, P single-channel aliasing observations are constructed according to the method (1), and the modulation pattern of I digital communication signals contained in each aliasing observation is selected from a set omega with the size of C; for the p (p E [1, P)]) And (3) aliasing observation, I FPSSE subsequences are extracted from the aliasing observation, and each subsequence and the known symbol sequence of the corresponding signal can form a training sample. For example, the training sample constructed based on the ith target signal in the p-th aliased observation can be expressed as:
Figure BDA0003926400850000104
wherein
Figure BDA0003926400850000105
From [ L ] th of signal i i ，Q i -L i ]And each coded symbol is composed. After the initial generation and extraction, all PI training samples in S are divided into C subsets according to their corresponding signal modulation patterns. Thus, there are a total of C training sample subsets S d |d∈[1，C]Are generated, each subset of training samples dedicated to RNN training for a certain modulation pattern signal.
Fig. 2 shows the RNN-based SIMO multi-signal symbol sequence estimator proposed by the present method. Single-input herein refers to the output of multiple-signal symbol sequences based on single-channel aliasing observation only, and multiple-output herein refers to the output of multiple-signal symbol sequencesThe results are listed. The symbol sequence estimator is first trained offline before online testing. In offline training, the C RNNs are each at { S } d |d∈[1，C]Get the training completion sequence marking task. In the online test, first, I FPSSE sub-sequences are extracted from an aliasing observation to be tested, and then the I FPSSE sub-sequences are respectively sent to trained RNNs corresponding to modulation patterns, so that finally, the output of a plurality of RNNs are symbol sequences of I signals obtained through estimation respectively.
In a further optimization scheme, a conventional symbol sequence estimator (e.g., a coherent demodulator) generally performs symbol estimation directly based on a decision on a synchronous decimation result of a down-converted received signal, and in a case where a received signal simultaneously includes multiple communication signals, an error rate of the received signal is greatly increased due to mutual interference among the signals. Therefore, when estimating the aliasing multi-signal symbol sequence, a new feature extraction mode needs to be designed, the key point is to avoid mutual interference of signals in the symbol sequence estimation process, and then the multi-signal symbol sequence estimation can be finished in a sequence labeling mode. In formula (2), g i (.) has an FIR structure and is derived from
Figure BDA0003926400850000111
Continue until
Figure BDA0003926400850000112
Therefore, is received i Q of (n) i The sample point range of each symbol influence is:
Figure BDA0003926400850000113
for the first
Figure BDA0003926400850000114
The number of the symbols is one,
Figure BDA0003926400850000115
comprises the following steps:
Figure BDA0003926400850000116
Figure BDA0003926400850000117
when in use
Figure BDA0003926400850000118
When s is i The first order difference sequence of (n) is represented as:
Figure BDA0003926400850000121
assuming the sampling is sufficient
Figure BDA0003926400850000122
And g is i (. Cndot.) is a symmetric function, in which case:
Figure BDA0003926400850000123
Figure BDA0003926400850000124
and
Figure BDA0003926400850000125
equation (14) can be approximated as:
Figure BDA0003926400850000126
when in use
Figure BDA0003926400850000127
When d is greater than i (n) is
Figure BDA0003926400850000128
Formulas (15) and (16) indicate that:
1)
Figure BDA0003926400850000129
therein is implicitly about
Figure BDA00039264008500001210
And
Figure BDA00039264008500001211
the information of (1). As a whole, { s } i (n)|i∈[1，I]The symbol information of which is contained in the symbol information of which is located
Figure BDA00039264008500001212
In the first order difference sequence at the point.
2){s i (n)|i∈1，I]The first order difference sequence of is sparse in nature, i.e. only at a few points will have significant values. The positions of these significant points are determined by the symbol oversampling rate of each target signal.
Figure BDA00039264008500001213
At a point exactly the target signal s i The symbol switching time of (n) is different, so that the symbol subsequences influencing the values of two adjacent sampling points are different, and when the difference operation is performed, the difference value contains the information of the front and rear symbols (as shown in formula (15)), and the amplitude is larger. At the target signal s i At the non-symbol switching time of (n), the symbol subsequence affecting the values of the adjacent sampling points remains unchanged, so that the differential value is smaller (as shown in formula (16), and the differential value at the time is only g i The value change between adjacent sampling points is determined, and the change is weaker under the condition of sufficient sampling).
Combining equations (1), (15) and (16), the first order difference of x (n) is:
Figure BDA0003926400850000131
wherein phi i Represents the set of the following sample points:
Figure BDA0003926400850000132
as shown in the formula (17), the signal s i The symbol information of (n) is contained in
Figure BDA0003926400850000133
While interference from other target signals therein is effectively suppressed. Because of the fact that
Figure BDA0003926400850000134
And
Figure BDA0003926400850000135
are all related to each other, therefore
Figure BDA0003926400850000136
In fact structured. This requirement is therefore based on
Figure BDA0003926400850000137
Element pair signal s in (1) i The symbol of (n) is used to estimate, taking into account all the context information in the input sequence. Here, consider the use of an RNN with BLSTM structure to accomplish this labeling task.
In a further optimization scheme, due to the fact that the over-sampling rates of target signal symbols may be different and the transmission delay of each signal is random, it can be assumed that
Figure BDA0003926400850000138
With such orthogonality, the differential sequence of the received signal may actually be decomposed into a plurality of differential subsequences corresponding to the respective target signals. Thus, extracting the FPSSE subsequence further comprises: for a target signal s i (n) subsequences thereof
Figure BDA0003926400850000139
From { d (n) | n ∈[1，N]The following are extracted:
Figure BDA00039264008500001310
it is noted that the reason is that
Figure BDA00039264008500001311
And has the formula (16) with other signals to
Figure BDA00039264008500001312
The influence of (b) will be effectively suppressed.
After being divided according to each signal, { d (n) | n ∈ [1, N]Will be divided into a set of sub-sequences
Figure BDA0003926400850000141
Wherein each subsequence corresponds to a target signal.
Further optimising the scheme, it is desirable that the input to a symbol sequence estimator is ideally related only to the transmitted symbols. This is to avoid the influence of other symbol-independent signals and channel parameters on the symbol estimation result. And, as can be seen from equation (17), in addition to transmitting symbols,
Figure BDA0003926400850000142
the amplitude and phase of (A) are still respectively influenced
Figure BDA0003926400850000143
And
Figure BDA0003926400850000144
the influence of (c). Thus, extracting the FPSSE subsequence further comprises: for I e [1I ]]Is further based on
Figure BDA0003926400850000145
The following normalized amplitude and phase sequences, denoted respectively, were calculated
Figure BDA0003926400850000146
Figure BDA0003926400850000147
Figure BDA0003926400850000148
Wherein, | - | represents the amplitude of the complex sequence,
Figure BDA0003926400850000149
to represent
Figure BDA00039264008500001410
Is the mean value of the complex number, k i ∈[1，Q i -2L i +1]Will be
Figure BDA00039264008500001411
As signal s in the received signal i (n) the corresponding FPSSE sequence.
In the further optimization scheme, in the off-line training of the recurrent neural networks, in order to demodulate signals possibly existing in different modulation patterns, a recurrent neural network set is constructed, the structures and parameters of all recurrent neural networks in the set are the same, training is carried out on symbol sequence estimation of target signals oriented to different modulation patterns, and each recurrent neural network comprises two BLSTM layers and a full-connection classification layer. The activation function of the classification layer is softmax.
In a further optimization scheme, the output dimension of the classification layer is determined by the debugging order of the signal to which the classification layer faces.
Performance and test condition index
The RNN-based SIMO multi-signal symbol sequence estimator obtained by the method is tested under different conditions and compared with the existing multi-signal symbol sequence estimation algorithm based on single-channel observation.
The simulation experiment 1 tests the estimation performance of the symbol sequence estimator under different target signal power ratios, and compares the selected RNN structure in the method with other RNN network structures. Simulation experiment 2 compares the symbol sequence estimator provided by the method with the existing symbol sequence estimation algorithm based on model and data driving in both the aspects of symbol sequence estimation precision and time complexity. In consideration of the importance of generalization capability to the data-driven method, simulation experiment 3 tests the test performance of the sequence estimator provided by the method under the condition of not seeing the training.
The Bit Error Rate (BER) of each target signal is used as an index. The BER corresponding to the ith target signal may be defined as follows:
Figure BDA0003926400850000151
wherein the content of the first and second substances,
Figure BDA0003926400850000152
converting the estimated symbol sequence into a corresponding bit sequence, and then obtaining the number of bits inconsistent with the actual bit sequence; b is i Is the total length of the target signal bit sequence.
When a certain target signal symbol sequence is estimated, error bits mainly originate from other aliasing target signals and the disturbance of environmental noise, so that the performance of the symbol sequence estimator under different disturbance intensities is mainly considered in the simulation experiment of the method, and the signal-to-interference-and-noise ratio SINR is used as a measurement index of the disturbance intensity in aliasing observation. It should be noted that the target signal and the interference signal are relative concepts, and when a symbol sequence estimation is performed on a certain target signal, all the remaining target signals are regarded as interference to the current signal, and vice versa. Thus, for a target signal i (i ∈ [1, i ]), its SINR in aliased observations is defined as:
Figure BDA0003926400850000161
simulation experiment
The basic training parameters and settings of the symbol sequence estimator are as follows: RNN is realized based on a PyTorch0.4.0 platform, and training/verification and test data are generated on an Intel (R) Core (TM) i7-6500UCPU @2.50GHz processor based on MatlabR2018 a. Training/validation/testing of a single RNN is done based on 4 × 105/1 × 105/4 × 106 modulation symbols, respectively. Different channel gains are simulated in the training sample generation process, and the SNR is set to be 25dB. The number of nodes in the forward and backward directions of each hidden layer is 128. The training optimizer and learning rate are set to classical Adam and 0.001, respectively. The training batch size is 100 and the number of training rounds is 200.
Simulation experiment 1 the symbol sequence estimator provided by the method is used for performance test and analysis under different target signal power ratios.
In the simulation experiment, two target signals are considered, namely BPSK and 2PAM signals. To examine the applicability of the proposed symbol sequence estimator to different scenarios of target signal symbol rates, the symbol rates of the two target signals are set to 0.112KB (Baud) and 0.1KB, respectively. Fig. 3 shows BER after the symbol sequence estimator provided by the present method performs parallel symbol sequence estimation on two target signals under different power ratios. Wherein, (a) BPSK (b) 2PAM. SNR =25dB. L is 1 ＝L 2 And =1. When calculating the SINR of a certain signal, another target signal is considered as interference. The BER results of the conventional coherent demodulator and the symbol sequence estimator constructed based on the original RNN are also shown (in the original RNN symbol sequence estimator, the BLSTM layer is replaced by the original RNN layer with the same number of neurons). It can be seen that, in the case that coherent demodulation (i.e. the conventional symbol sequence estimation process) is almost completely failed, the symbol sequence estimator provided in the present method successfully performs parallel estimation on two target signal symbol sequences under different SINRs. The symbol sequence estimator provided by the method still completes accurate estimation of the target signal symbol sequence even when the interference signal is significantly stronger than the current target signal (for example, when SINR = -10 dB)Both target signals BER are below 10 "2), when the conventional coherent demodulator has substantially failed. This is mainly due to the FPSSE extraction performed. As can be seen from the formulas (16) and (17), in
Figure BDA0003926400850000171
Has been substantially suppressed, the estimation of the current target signal symbol sequence can be made less affected by interference from other target signals. On the other hand, the results of the BLSTM network are seen to be superior to those of the original RNN network, since long-term and bi-directional context information in input/output can be exploited. This also indicates the necessity of utilizing the bi-directional context information and long-term context information in the sequence tagging problem modeled by the method.
Simulation experiment 2: the symbol sequence estimator provided by the method is compared and analyzed with the performance of the existing symbol sequence estimation algorithm.
In the simulation experiment, the SIMO multi-signal symbol sequence estimator provided by the method is compared with the existing SIMO symbol sequence estimation algorithm based on model and data driving. To make a fair comparison, two aliased BPSK signals with the same symbol rate are considered, limited by the applicability of the PSP algorithm. The relative delay between the two signals is 0.1 times the symbol duration. For the PSP algorithm, the step size of its LMS iteration is set to 0.005. In DCND, the number of convolution kernels included in the two convolution layers is 128 and 64, respectively. Fig. 4 shows the test BER for different symbol sequence estimation algorithms (two target signal results shown separately). Wherein (a) BPSK #1 and (b) BPSK #2. It can be seen that the symbol sequence estimator proposed by the present method has a significant BER reduction compared to the existing model-based and data-driven symbol sequence estimator, especially when the power of the two target signals has a significant difference. This is mainly because the input of the existing symbol sequence estimation algorithm is not differentiated for different target signals (in PSP, joint demodulation is based on symbol rate decimation of aliased observations, and in DCND, directly on aliased observations), i.e. symbol sequence estimation is performed for different target signals based on the same input, which makes interference between different target signals inevitably present. Different from the above, in the symbol sequence estimator provided by the present method, respective fpse sub-sequences are extracted for different target signals, and the sparsity of the fpse sub-sequences determines that the mutual interference of the target signals after extraction is very weak, so that the present method can be well adapted to a scene with a significant target signal power difference. Meanwhile, the above results also verify that the symbol sequence estimator provided by the method has good adaptability to the scene with the same target signal symbol rate, and thus is a more practical algorithm compared with the PSP in this respect. On the other hand, table 1 lists the number of trainable parameters and the average required running time in the symbol sequence estimator proposed by the present method and the existing symbol sequence estimation algorithm (the number of trainable parameters is only for the data-driven method). It can be seen that under the test conditions of the simulation experiment, the time complexity of the estimator provided by the method is similar to that of the PSP algorithm. It should be noted that the current simulation experiment is a test under the condition that the target signal number is only 2 and the modulation order of the target signal is 2. The estimator provided by the method adopts a parallel estimation mode, namely symbol sequence estimation is carried out on a plurality of target signals synchronously and independently, and the calculation complexity of the estimator does not increase with the modulation order index of the target signals. In contrast, if the number of target signals is increased or the modulation order of the target signals is increased, the time complexity of the PSP algorithm will increase exponentially, and it is expected that the average operation time will increase significantly.
TABLE 1 different SIMO symbol sequence estimation algorithms can train the number of parameters and the average estimation time (every 1000 symbols)
Figure BDA0003926400850000191
Simulation experiment 3: the symbol sequence estimator provided by the method can be used for generalization capability test and analysis.
For data-driven methods, generalization capability is always an important issue. Since the training process cannot traverse all possible data in general, whether to have the ability to deal with unseen data during training is one of the keys for determining whether the data-driven method is practical. In the simulation experiment, the performance of the symbol sequence estimator provided by the method under the generalization condition is tested. Here, the test under the generalization condition mainly means that a signal/channel parameter or the like used when generating the test data does not appear in the training data generation. In the first case of fig. 5, the test SNR is reduced from 25dB to 15dB. In the second case, the relative time delay of the two target signals is set to be randomly generated between (0.1-0.9) times the symbol duration, and then the average BER is calculated over a number of tests. It can be seen that although the symbol sequence estimator proposed by the present method performs off-line training only under certain conditions, it can still cope well with the generalization test conditions not seen in training. Even if the SNR is reduced by 10dB, the resulting BER increase is still within an acceptable range. It should also be noted that if different SNRs are considered during training, the performance of the symbol sequence estimator proposed by the present method at lower test SNR will be improved predictably. This is achieved by simply generating the training data over a wider range of SNRs.
Finally, it should be noted that: the above-mentioned embodiments are only specific embodiments of the present invention, which are used for illustrating the technical solutions of the present invention and not for limiting the same, and the protection scope of the present invention is not limited thereto, although the present invention is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope of the present disclosure; such modifications, changes or substitutions do not depart from the spirit and scope of the present invention in its spirit and scope. Are intended to be covered by the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (8)

1. A method for directly estimating a symbol sequence of multiple communication signals under an interference condition, comprising:
extracting parallel symbol sequence estimation characteristics FPSSE based on single-channel observation to obtain characteristic sequences respectively containing symbol information of each target signal;
and carrying out sequence marking based on the cyclic neural network after offline training to obtain a symbol sequence estimation result, wherein the characteristic sequence is used as input when the cyclic neural network is subjected to offline training.
2. The method according to claim 1, wherein before extracting the symbol sequence estimation feature, constructing a single-channel aliasing observation specifically comprises:
assuming that a single antenna receives I spectrally overlapping digital communication signals simultaneously, this single-channel aliasing observation can be expressed as:
Figure FDA0003926400840000011
wherein s is i (n-m i ) Representing a transmission delay of m i I target signal of one sampling interval, a i V (N) represents AWGN for its corresponding channel gain, N is the total number of sampling points;
for typical amplitude/phase modulated digital communication signals, s i (n) is further represented as
Figure FDA0003926400840000012
Wherein E is i Is the signal power;
Figure FDA0003926400840000013
q-th corresponding to transmission i A symbol, Q i Is the total number of symbols; g i The expression comprehensively considers the equivalent pulse shaping function of the pulse shaping of the sending end and the physical transmission channel, T s And
Figure FDA0003926400840000014
respectively, a sampling interval and a symbol oversampling rate;
Figure FDA0003926400840000015
is the residual carrier frequency.
3. The method of claim 2, wherein performing parallel symbol sequence estimation feature FPSSE extraction based on single channel observations comprises:
p single-channel aliasing observations are constructed according to the method (1), and the modulation pattern of the I digital communication signals contained in each aliasing observation is selected from a set omega with the size of C;
for the p (p E [1, P ]) aliasing observation, I FPSSE subsequences are extracted from the aliasing observation, and each subsequence and the known symbol sequence of the corresponding signal can form a training sample.
4. The method of claim 3, wherein extracting the FPSSE subsequence comprises: suppose g i (.) has a finite long single-bit impulse response FIR structure, and
Figure FDA0003926400840000021
continue until
Figure FDA0003926400840000022
Therefore s to i Q of (n) i The sample point range of each symbol influence is:
Figure FDA0003926400840000023
for the first
Figure FDA0003926400840000024
The number of the symbols is one,
Figure FDA0003926400840000025
is provided with
Figure FDA0003926400840000026
Figure FDA0003926400840000027
When in use
Figure FDA0003926400840000028
When s is i The first order difference sequence of (n) is represented as:
Figure FDA0003926400840000029
assuming the sampling is sufficient
Figure FDA0003926400840000031
And g is i (. Cndot.) is a symmetric function, in which case:
Figure FDA0003926400840000032
Figure FDA0003926400840000033
and
Figure FDA0003926400840000034
equation (14) is then approximated as:
Figure FDA0003926400840000035
when in use
Figure FDA0003926400840000036
When d is greater than i (n) is
Figure FDA0003926400840000037
Combining equations (1), (15) and (16), the first order difference of x (n) is:
Figure FDA0003926400840000038
wherein phi i Represents the set of the following sample points:
Figure FDA0003926400840000039
5. the method of claim 4, wherein extracting the FPSSE subsequence further comprises: for a target signal s i (n) subsequences thereof
Figure FDA00039264008400000310
From { d (n) | n ∈ [1, N ]]Extract as follows:
Figure FDA00039264008400000311
after being divided according to each signal, { d (n) | n ∈ [1, N]Will be divided into a set of sub-sequences
Figure FDA00039264008400000312
Wherein each subsequence corresponds to a target signal.
6. The method of claim 5, wherein extracting the FPSSE subsequence further comprises:
for i e [1, I]Is further based on
Figure FDA0003926400840000041
The following normalized amplitude and phase sequences, denoted respectively, were calculated
Figure FDA0003926400840000042
Figure FDA0003926400840000043
Figure FDA0003926400840000044
Wherein, | - | represents the amplitude of the complex sequence,
Figure FDA0003926400840000045
to represent
Figure FDA0003926400840000046
Is the mean value of the complex number, k i ∈[1，Q i -2L i +1]Will be
Figure FDA0003926400840000047
As signal s in the received signal i (n) the corresponding FPSSE sequence.
7. The method according to claim 1, wherein in the off-line training of the recurrent neural networks, in order to demodulate signals with possibly different modulation patterns, a recurrent neural network set is constructed, the structure and parameters of each recurrent neural network in the set are the same, and the estimation of the symbol sequence of the target signal facing different modulation patterns is trained, and each recurrent neural network comprises two BLSTM layers and one fully-connected classification layer.
8. The method of claim 7, wherein the output dimension of the classification layer is determined by the debugging order of the signal it is facing.
CN202211377436.3A 2022-11-04 2022-11-04 Method for directly estimating symbol sequence of multiple communication signals under interference condition Active CN115834306B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211377436.3A CN115834306B (en) 2022-11-04 2022-11-04 Method for directly estimating symbol sequence of multiple communication signals under interference condition

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211377436.3A CN115834306B (en) 2022-11-04 2022-11-04 Method for directly estimating symbol sequence of multiple communication signals under interference condition

Publications (2)

Publication Number Publication Date
CN115834306A true CN115834306A (en) 2023-03-21
CN115834306B CN115834306B (en) 2023-09-05

Family

ID=85526656

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211377436.3A Active CN115834306B (en) 2022-11-04 2022-11-04 Method for directly estimating symbol sequence of multiple communication signals under interference condition

Country Status (1)

Country Link
CN (1) CN115834306B (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101394388A (en) * 2008-10-24 2009-03-25 武汉大学 Demodulating and modulation processing method for radio digital domain
CN106850470A (en) * 2017-02-17 2017-06-13 中国人民解放军海军工程大学 A kind of Interference Cancellation based on affine precoding and the channel estimation methods of two-way cooperation
CN108206799A (en) * 2016-12-16 2018-06-26 中国航天科工飞航技术研究院 A kind of carrier frequency bias estimation for burst communication
CN110048982A (en) * 2019-04-04 2019-07-23 中国人民解放军战略支援部队信息工程大学 The blind separating method and device of the paired carrier multiple access signal of distinct symbols rate
CN113259283A (en) * 2021-05-13 2021-08-13 侯小琪 Single-channel time-frequency aliasing signal blind separation method based on recurrent neural network
CN114464206A (en) * 2022-04-11 2022-05-10 中国人民解放军空军预警学院 Single-channel blind source separation method and system

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101394388A (en) * 2008-10-24 2009-03-25 武汉大学 Demodulating and modulation processing method for radio digital domain
CN108206799A (en) * 2016-12-16 2018-06-26 中国航天科工飞航技术研究院 A kind of carrier frequency bias estimation for burst communication
CN106850470A (en) * 2017-02-17 2017-06-13 中国人民解放军海军工程大学 A kind of Interference Cancellation based on affine precoding and the channel estimation methods of two-way cooperation
CN110048982A (en) * 2019-04-04 2019-07-23 中国人民解放军战略支援部队信息工程大学 The blind separating method and device of the paired carrier multiple access signal of distinct symbols rate
CN113259283A (en) * 2021-05-13 2021-08-13 侯小琪 Single-channel time-frequency aliasing signal blind separation method based on recurrent neural network
CN114464206A (en) * 2022-04-11 2022-05-10 中国人民解放军空军预警学院 Single-channel blind source separation method and system

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
XIN CAI; XIANG WANG; ZHITAO HUANG; FENGHUA WANG: "Single-Channel Blind Source Separation of Communication Signals Using Pseudo-MIMO Observations", IEEE COMMUNICATIONS LETTERS *
XIN CAI; ZHITAO HUANG; BAOGUO LI: "Asynchronous and Non-Stationary Interference Cancellation in Multiuser Interference Channels", IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS *
崔鹏辉: "成对载波多址信号单通道盲分离关键技术研究", 中国博士学位论文全文数据库 (信息科技辑) *

Also Published As

Publication number Publication date
CN115834306B (en) 2023-09-05

Similar Documents

Publication Publication Date Title
CN112418014B (en) Modulated signal identification method based on wavelet transformation and convolution long-term and short-term memory neural network
CN109412725B (en) Radio communication PCMA signal blind demodulation method and device
CN109951214B (en) Signal detection method suitable for large-scale MIMO system
CN112115821B (en) Multi-signal intelligent modulation mode identification method based on wavelet approximate coefficient entropy
Zhang et al. Deep learning aided OFDM receiver for underwater acoustic communications
CN112347871B (en) Interference signal modulation identification method for communication carrier monitoring system
CN111614584B (en) Transform domain adaptive filtering channel estimation method based on neural network
CN111884976B (en) Channel interpolation method based on neural network
CN112910812B (en) Modulation mode identification method for deep learning based on space-time feature extraction
CN113572708B (en) DFT channel estimation improvement method
WO2021088465A1 (en) Fast modulation recognition method using multilayer perceptron, and employing data fusion of multiple distribution tests
Zhang et al. Deep learning and expert knowledge based underwater acoustic OFDM receiver
Zhang et al. Deep learning based underwater acoustic OFDM receiver with joint channel estimation and signal detection
CN112422208B (en) Signal detection method based on antagonistic learning under unknown channel model
CN110474663B (en) Iterative intelligent signal detection method based on neural network
CN114614920B (en) Signal detection method based on data and model combined driving of learning factor graph
CN115834306A (en) Method for directly estimating multi-communication signal symbol sequence under interference condition
Yuan et al. Channel estimation and pilot design for uplink sparse code multiple access system based on complex-valued sparse autoencoder
Yıldırım et al. Deep receiver design for multi-carrier waveforms using cnns
CN112821971A (en) Time-varying channel signal detection method based on countermeasure learning
Park et al. MetaSSD: Meta-Learned Self-Supervised Detection
Ramezanpour et al. DNN‐based interference mitigation beamformer
Siu et al. Decision feedback equalization using neural network structures
CN111865489B (en) Multiple-input multiple-output detection method based on graph neural network
CN114662523A (en) Spread spectrum signal detection method integrating cognitive enhanced compressed sensing and artificial neural network

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