A data-driven SCA interference suppression method
Through sparse domain encoding and decoding technology, the interference suppression problem in asynchronous non-stationary interference scenarios under single-channel observation conditions is solved, and high-quality target signal recovery is achieved. It is suitable for single-channel systems, improving waveform similarity and reducing errors Symbol rate.
Patent Information
- Application Number
- CN202211543766.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-02
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2042-12-02
AI Technical Summary
The existing interference suppression method under single-channel observation conditions does not perform well in asynchronous non-stationary interference scenarios, especially the requirement for prior knowledge of the target signal and lacks applicability in time-frequency aliasing.
Using the data-driven SCA interference suppression method, the time domain aliasing observation is sparsely transformed through sparse domain encoding, the representation matrix of the target signal in the sparse domain is estimated, and the target signal is restored to the time domain through sparse domain decoding, and interference suppression is achieved. .
The target signal is restored with high quality under different interference types and intensity. It is suitable for single-channel systems without the prior requirements of aliasing observation. It is highly adaptable. After recovery, the waveform similarity is improved and the symbol error rate is significantly reduced.
Smart Images

Figure CN116015484B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of asynchronous non-stationary interference suppression, and in particular to a data-driven SCA interference suppression method. Background Art
[0002] Interference is an urgent problem to be solved in cooperative / non-cooperative communication systems. Reducing and suppressing interference is of great significance to improving the performance of cooperative / non-cooperative communication systems. In a multi-user wireless communication system, the transmitter usually uses the same medium to send messages to the corresponding receivers, so it is difficult to avoid mutual influence between the expected signals of each user, resulting in the receiver finally receiving a time-frequency aliasing observation. The received signal contains two components, the expected signal and the interference, where the interference refers to other signal components other than the expected signal. In the existing interference suppression work, it is usually assumed that the expected signal and the interference arrive at each receiver at the same time. However, in practice, each sending node generally works independently, and its working time, the type of signal sent, and the transmission delay to each receiving node may be different. Therefore, in the aliasing observation of a certain receiving node, the existence time intervals of different signals will generally be different (the existence of a signal in a certain time interval means that it has non-zero power in the interval). In the present invention, asynchronous non-stationarity refers to the number, type and arrival time of the current interference signal at the receiving node. Dynamic changes. In addition, the research on interference suppression technology should also take into account the actual system reception conditions. Limited by factors such as volume and cost, the communication system receiving modules carried by satellites, aircraft and other platforms are mostly single-channel. Therefore, the research on interference suppression technology under single-channel observation conditions will be more practical.
[0003] The existing methods of interference suppression under single-channel observation conditions mainly include algorithms based on generalized spectral Wiener filtering and algorithms based on signal subspace projection. By weighted summing the filtering results of different frequency-shifted components of aliased observations, the frequency-shifted components of the target signal are enhanced while the frequency-shifted components of the interference signal are weakened. However, the LCL-FRESH filter requires the target signal cycle frequency and even the signal waveform to be known, which limits its practicality to a considerable extent. In order to reduce the requirements of the FRESH filter for training signals, the prior art directly replaces the target signal itself with a certain frequency-shifted component of the aliased observation as the training signal, so compared with the FRESH filter, the practicality is significantly improved. However, the BA-FRESH filter still requires accurate prior knowledge of the target signal cycle frequency family, and its performance will be significantly reduced when there is an error in the cycle frequency. At the same time, there is a certain gap in its separation performance compared with the standard FRESH filter.
[0004] In addition to the above-mentioned method based on generalized spectral Wiener filtering, another important single signal waveform separation technology is signal subspace projection. Its basic principle is to first construct a subspace basis set so that the target signal and the interference signal are as orthogonal as possible in the subspace formed by the basis set, and finally achieve single signal waveform separation by projecting the aliased observation to the signal subspace. In order to solve the problem of communication anti-co-channel interference, a suitable wavelet basis is determined based on the target signal symbol rate information; then the aliased observation is projected to the wavelet basis, and the interference signal estimation is obtained based on the prior information of the interference signal pulse shaping function, and finally the target signal waveform is separated by offsetting the interference signal estimation result from the aliased observation. This method requires that the target signal is a baseband signal, and the symbol rate difference between the target signal and the interference signal is very small, and it is not very applicable to general time-frequency aliased digital communication signals. In the existing hypothetical aliased observation, there is a single signal region (SSR) of the target signal. The subspace basis of the target signal is constructed using the SSR detection result as a template to reduce the prior requirements for the target signal. However, if the SSR region does not exist, its performance will be significantly degraded or even fail.
[0005] In addition, unlike the above-mentioned algorithms based on pre-established mathematical models of target signals, channels, etc., single-channel aliasing observations are directly input into the demodulator, and the output of the network classification layer is used as the signal symbol sequence estimation result. However, it is easily affected by the power of the interference signal and cannot restore the waveform of the required signal. Therefore, it is not very practical in applications such as spectrum monitoring and specific signal detection. Summary of the invention
[0006] In order to solve the problems existing in the above-mentioned prior art, the present invention provides a data-driven SCA interference suppression method, proposes an interference suppression framework of "sparse domain coding-sparse domain representation estimation-sparse domain decoding", and performs sparse transformation on time domain aliasing observations through sparse domain coding; then, the estimation and separation recovery of the target signal representation are completed in the sparse domain; finally, the target signal in the sparse domain is converted to the time domain through sparse domain decoding, thereby completing the recovery of the target signal. It breaks through the prior limitations on aliasing observations and is more practical.
[0007] In order to achieve the above technical objectives, the present invention provides a data-driven SCA interference suppression method, comprising:
[0008] Construct an asynchronous non-stationary interference channel model to obtain aliasing observations of the receiving node;
[0009] Performing a sparse transformation on the aliased observations so that the target signal has sparseness in the transform domain;
[0010] Complete the representation matrix estimation of the target signal in the transform domain to obtain the representation matrix estimation result;
[0011] Based on the representation matrix estimation result, the target signal is restored in the transform domain to obtain a restoration result;
[0012] The recovery result is converted to the time domain through inverse transformation to complete interference suppression.
[0013] Optionally, the aliasing observation is:
[0014]
[0015] In the formula, a ii is the channel gain between the sending node i (i = 1, 2, ..., N) and the receiving node i (i = 1, 2, ..., N); s i (nm i ) is the transmission delay m i The signal transmitted by the i-th sending node in the sampling interval; a ji is the channel gain between the sending node j (j = 1, 2, ..., N) and the receiving node i (i = 1, 2, ..., N); s j (nm j ) indicates that the transmission delay is m j The signal transmitted by the jth sending node in the sampling interval; v i (n) represents the additive white Gaussian noise at the receiving node i; T is the total number of sampling points.
[0016] Optionally, the calculation formula of the sparse transformation is:
[0017] W i =f Encoder (X i )
[0018] in:
[0019] X i =A i S i +V i
[0020] In the formula, represents the aliased observation at the receiving node i; represents the aliasing matrix, where each column represents the channel coefficient between the sending node j and the receiving node i; represents the signal matrix, each row represents the signal transmitted by the sending node j received at the receiving node i; represents the AWGN at receiving node i;
[0021] is the time domain aliasing mixed measurement X iThe deep high-dimensional representation of K is the number of encoder filters, and L is the output length of a single filter; f Encoder (·) represents the mapping function of the encoder.
[0022] Optionally, the calculation formula for restoring the target signal in the sparse domain is:
[0023]
[0024] In the formula, is the estimation of the target signal in the sparse domain; is the time domain aliasing mixed measurement X i Deep high-dimensional representation of is the representation matrix of the target signal in the sparse domain, which is estimated by the separator; Represents a matrix dot product operation.
[0025] Optionally, the calculation formula for the inverse transformation is:
[0026]
[0027] is the estimate of the target signal after recovery in the time domain; f Decoder (·) represents the mapping function of the decoder; is the estimation of the target signal in the sparse domain.
[0028] Optionally, the asynchronous non-stationary interference channel model includes N main communication links and N(N-1) interference links.
[0029] Optionally, a performance evaluation is performed on the recovery result, including the waveform similarity of the recovered target signal and the demodulation symbol error rate of the recovered target signal.
[0030] Optionally, the calculation formula of the waveform similarity is:
[0031]
[0032] In the formula, ||·|| represents the 2-norm;
[0033] when hour, c is a non-zero constant; when with a ii s ii When orthogonal, The closer it is to 1, the The higher the restoration quality, the smaller the error.
[0034] Optionally, the calculation formula of the demodulation symbol error rate is:
[0035]
[0036] In the formula, Q is the number of symbols in the symbol sequence obtained by demodulating the target signal before or after recovery that are inconsistent with the real symbol sequence; i is the total number of target signal symbol sequences.
[0037] The present invention has the following technical effects:
[0038] 1. The present invention has the ability to recover the target signal with high quality under different interference types and interference strengths, and has good adaptability to strong interference situations with large differences in interference and signal power levels. Under different SNRs and SIRs, the average waveform similarity after recovery is increased by 57.189%, and the symbol error rate is reduced by an average of 3 orders of magnitude;
[0039] 2. Compared with the existing methods, the present invention is more universal, does not require a priori requirements for aliasing observation, can achieve high-quality recovery of the target signal in the case of severe time-frequency aliasing with interference, and is suitable for single-channel receiving systems;
[0040] 3. The present invention has strong adaptability to changes in interference parameters such as SNR and interference signal duration. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0042] Figure 1 It is a flowchart of a data-driven SCA interference suppression method according to an embodiment of the present invention;
[0043] Figure 2 This is a schematic diagram of an N-user interference channel according to an embodiment of the present invention;
[0044] Figure 3 This is a schematic diagram of aliasing observation according to an embodiment of the present invention;
[0045] Figure 4 This is a framework diagram of the interference suppression method according to an embodiment of the present invention;
[0046] Figure 5 A schematic diagram of a network structure of an interference suppression method according to an embodiment of the present invention;
[0047] Figure 6 This is a schematic diagram of the structure of Conv1DBlock according to an embodiment of the present invention;
[0048] Figure 7 The test WS under different SIRs of the existing method of the embodiment of the present invention;
[0049] Figure 8 The test SER under different SIRs of the existing method of the embodiment of the present invention;
[0050] Fig. 9 WS and SER under generalized test conditions in simulation experiment 3 of an embodiment of the present invention. DETAILED DESCRIPTION
[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0052] like Figure 1 As shown, the present invention discloses an asynchronous non-stationary interference suppression method, comprising:
[0053] Construct an asynchronous non-stationary interference channel model to obtain aliasing observations of the receiving node;
[0054] In a multi-user communication system, due to the broadcast characteristics of the wireless channel, the receiver not only receives the information of the desired transmitter, but also receives interference information from other transmitters. This type of communication channel can be modeled as a multi-user interference channel. The present invention constructs an asynchronous non-stationary interference channel model, which includes N pairs of "send-receive" multi-user interference channels (i.e., N main communication links and N (N-1) interference links). Its structure is shown in the figure below: Figure 1 As shown, Figure 2 The solid line in the middle represents the "main communication link", which refers to the link formed by a transmitter communicating with its desired receiver; the dotted line represents the "interference link", which refers to the link formed by a receiving node receiving communication with an undesired transmitter. Each communication node uses a single antenna, and the aliasing observation at the receiving node i (i = 1, 2, ..., N) is:
[0055]
[0056] In the formula, a ii is the channel gain between the sending node i (i = 1, 2, ..., N) and the receiving node i (i = 1, 2, ..., N); s i (nm i ) is the transmission delay m i The signal transmitted by the i-th sending node in the sampling interval; a jiis the channel gain between the sending node j (j = 1, 2, ..., N) and the receiving node i (i = 1, 2, ..., N); s j (nm j ) indicates that the transmission delay is m j The signal transmitted by the jth sending node in the sampling interval; v i (n) represents the additive white Gaussian noise (AWGN) at the receiving node i, and T is the total number of sampling points.
[0057] Here, the difference in the transmission delay of each signal is considered, which is mainly because the paths of each target signal to reach the receiving antenna are generally different in practice. i (n)(i=1,2,…,N)is:
[0058]
[0059] in, is the signal power, A iq The qth modulation symbol corresponding to the ith signal; g i (·) is the duration N ci The pulse shaping function of N sampling points, ci is the number of symbol oversampling points; the sampling interval is T s , f ci With Q i Respectively represent the carrier frequency and the total number of symbols of the i-th source signal. In addition, since each target signal is generally transmitted from a different communication system, its carrier frequency is still likely to have a certain difference when the spectrum overlaps. The present invention considers that the mixed model at the receiving node is a linear time-delay instantaneous mixed model, and temporarily does not consider the multipath effect of the signal transmitted by each sending node at the receiving node. Therefore, formula (1) can also be expressed by a vector as:
[0060] X i =A i S i +V i (3)
[0061] in, represents the aliased observation at the receiving node i; represents the aliasing matrix, where each column represents the channel coefficient between the sending node j and the receiving node i; represents the signal matrix, each row represents the signal transmitted by the sending node j received at the receiving node i; represents the AWGN at receiving node i.
[0062] The purpose of the single antenna interference suppression of the present invention is to achieve X iThe main communication link component (i.e., the target signal a ii s ii ) waveform recovery. Figure 3 This is a schematic diagram of aliasing observation of the present invention. Due to full consideration of actual application scenarios, there is no target signal SSR in the aliasing observation.
[0063] A sparse transform is performed on the aliased observations to make the target signal sparse in the transform domain. Sparsity means that in the time domain or a certain transform domain, only a few sample points have values far from zero, while most of the sample points have values close to zero. The representation matrix of the target signal is estimated in the transform domain to obtain the representation matrix estimation result. Based on the representation matrix estimation result, the target signal is restored in the transform domain to obtain the restored result. The restored result is converted to the time domain through inverse transform to complete interference suppression.
[0064] Specifically, the three core issues in the above SCA are: the selection of sparse transform domain, the estimation of aliasing matrix in the transform domain, and the recovery of target signal under the condition that the aliasing matrix is known. Among them, the degree of adaptation between the sparse transform domain and the signal determines the degree of validity of the signal sparsity assumption, that is, the overcompleteness of the signal representation basis in the sparse domain determines the performance of subsequent signal separation. However, for communication signals, a sparse transform domain with good universality has not yet been proposed, especially when multiple time-frequency aliasing signals are contained in a single-channel observation. Based on this, the present invention proposes an interference suppression method framework such as Figure 4 As shown in the figure, it includes three parts: sparse domain coding, sparse domain representation estimation and sparse domain decoding. Sparse domain coding is used to perform sparse transformation on time domain aliasing observations; sparse domain representation estimation is used to complete aliasing matrix estimation; sparse domain decoding is used to convert the target signal recovery result back to the time domain. First, the aliasing observation is converted to the corresponding representation in the sparse space through the sparse domain encoder, and then this representation is used to estimate the sparse representation matrix of the target signal in the sparse domain, so as to realize the recovery of the target source signal in the sparse space. Finally, the sparse domain decoder is used to reconstruct the target signal recovered in the sparse domain into the final target signal in the time domain.
[0065] Specifically, first, the encoder transforms the time-domain aliasing observation X i The input is converted to a sparse domain representation W i (The sparse domain here means that there is a certain area in the transform domain where only the target signal dominates, that is, only the target signal has non-zero power), and the conversion process is:
[0066] W i =f Encoder (X i ) (4)
[0067] In the formula, is the time domain aliasing mixed measurement X iThe deep high-dimensional representation of K is the number of encoder filters, and L is the output length of a single filter; f Encoder (·) represents the mapping function of the encoder.
[0068] Secondly, is the representation matrix of the target signal in the sparse domain, which is estimated by the separator, and then, M i With W i Multiply to get the estimate of the target signal in the sparse domain The calculation formula for the target signal representation matrix estimation is:
[0069]
[0070] in, Represents a matrix dot product operation.
[0071] Finally, the decoder will Convert to the time domain to get the estimated target signal after recovery The conversion process is:
[0072]
[0073] is the estimate of the target signal after recovery in the time domain; f Decoder (·) represents the mapping function of the decoder; is the estimation of the target signal in the sparse domain.
[0074] Specifically, the network structure of the interference suppression method proposed in the present invention is as follows: Figure 5 As shown in Figure 1, the network structure consists of three parts: sparse domain encoder, sparse domain representation estimator and sparse domain decoder, which correspond to the encoder, separator and decoder in ConvTasNet respectively:
[0075] (1) Sparse domain encoder: The adaptive selection of the sparse transform domain of the target signal is performed by a single one-dimensional convolutional layer and a nonlinear activation function (parametric rectified linear unit, PReLU), which is defined as:
[0076]
[0077] in, is a trainable parameter.
[0078] (2) Sparse domain representation estimator: Figure 5As shown in the figure, the adaptive learning of the sparse representation matrix of the target signal in the transform domain in the sparse domain is mainly completed by three convolutional stacking blocks with the same structure, where the output of the previous convolutional stacking block is the input of the next convolutional stacking block, the purpose of which is to extract the deep representation of the input by deepening the network depth, and then the output of the convolutional stacking block passes through a convolution layer with a convolution size of 1 and a Sigmoid activation function in turn, thereby completing the estimation of the sparse representation matrix of the target signal in the sparse domain. In addition, in order to reduce the impact of data changes on the stability of the sparse domain representation estimation module, the output W of the sparse domain encoder is i Before entering the sparse domain representation estimator, we must first i Perform layer normalization. The convolution layer after the LN layer is used to control the number of channels of the input convolution stack block. Specifically, a convolution stack block consists of eight one-dimensional convolution blocks ( Figure 5 Conv1D Block) is composed of Conv1D Block, and the structure of Conv1D Block is as follows Figure 6 As shown, the expansion factor in each Conv1D Block increases exponentially to ensure that the network has a large enough receptive field to model the long-term dependency of the signal sequence. The expansion factor in the mth (m = 1, 2, ..., 8) Conv1D Block is 2 m-1 In addition, to reduce the number of network parameters, the depth-separable convolutional network (D-conv in the figure) is used to replace the standard convolution. LN is:
[0079]
[0080]
[0081]
[0082] in, and is a trainable parameter; ε is a constant (default ε=10 -5 ).
[0083] (3) Sparse domain decoder: Automatic recovery of the target signal is performed by a single one-dimensional transposed convolutional layer.
[0084] Furthermore, the present invention uses two types of performance indicators to evaluate the signal recovery performance. One is to use the waveform similarity of the target signal after recovery as the performance evaluation indicator from the perspective of the waveform recovery quality of the target signal, and the calculation formula is:
[0085]
[0086] In the formula, ||·|| represents the 2-norm; when hour, c is a non-zero constant; when with a ii s ii When orthogonal, The closer it is to 1, the The higher the restoration quality, the smaller the error.
[0087] Second, from the perspective of the accuracy of communication information transmission, the demodulation symbol error rate (SER) of the recovered target signal is used as the performance evaluation indicator. The SER calculation formula corresponding to the i-th target signal is:
[0088]
[0089] In the formula, Q is the number of symbols in the symbol sequence obtained by demodulating the target signal before or after recovery that are inconsistent with the real symbol sequence; i is the total number of target signal symbol sequences.
[0090] On the other hand, since the error mainly comes from the disturbance of other aliased interference signals and environmental noise when restoring a target signal waveform, the embodiment of the present invention designs a simulation experiment, which focuses on the performance of the present invention under different environmental noise levels and interference signal strengths, and uses the signal-to-noise ratio (SNR) and signal-to-interference ratio (SIR) in aliased observation as the measurement indicators of disturbance intensity. For the target signal i (i∈[1,N]), its SNR and SIR in aliased observation are defined as:
[0091]
[0092]
[0093] Simulation test:
[0094] The basic training parameters and settings are as follows: the model is implemented based on Python 3.8.13 / Pytorch1.11.0 / cuda11.6 environment, and the training / validation and test data are generated based on Matlab R2021a on an Intel(R) Core(TM) i7-8700U CPU@3.20GHz processor. The training optimizer and learning rate are set to Adam and 0.001 respectively, the batch size is set to 10, and the number of training rounds is 100.
[0095] Simulation experiment 1: Performance test and analysis of the present invention under different environmental noise levels and interference signal strengths
[0096] In this experiment 1, the aliasing observation at the receiving node 1 in the multi-user interference channel with N=4 is considered. The aliasing observation consists of the target signal and three interference signals. The modulation style of the target signal is BPSK, and the modulation styles of the three interference signals are 8PSK, PAM8 and 16QAM respectively. The sampling duration of the aliasing observation is 10ms, and the total number of sampling points is 30270. Among them, the target signal exists in the time interval of 0-10ms, the target signal carrier frequency is 100KHz, and the interference signal exists in the time interval of 0-5ms, 4-8ms, and 6-10ms respectively. The carrier frequencies of the three interference signals are randomly selected in the three intervals of 101KHz-103KHz, 97KHz-99KHz, and 104KHz-106KHz respectively. The carrier frequencies of the target signal and the three interference signals are 100KHZ, 103KHz, 97KHz, and 106KHz respectively. The training / validation / testing of the model is based on the number of training / validation / testing samples under a single SNR or SIR, respectively, and the number of samples is 3×10 3 / 5×10 2 / 1×10 4 The target signal is completed by samples, a single target signal sample contains 320 symbols, and the symbol rate of the target signal and the interference signal is 31.25KB (Baud).
[0097] The quality of the target signal recovered under different environmental noise interference strengths SNR is shown in Table 1, and the SIR in the aliasing observation is set to -4.77dB. Table 2 shows the quality of the target signal recovered under different interference signal strengths SIR, and the SNR in the aliasing observation is set to 20dB. P With SER E Respectively represent the signal demodulation SE before and after recovery R , WS P With WS E They represent the waveform similarity before and after recovery, respectively. Here, WSP is the waveform similarity calculated directly based on the aliasing observation for the target signal, and SER P is the SER of the target signal demodulated directly based on the aliased observations.
[0098] Table 1
[0099]
[0100] Table 2
[0101]
[0102] From Table 1, we can see that the waveform of the target signal is effectively restored, and the quality of the restored waveform is significantly improved. When SNR>=5dB, the average WS (WS E) is equal to 99.483%. Compared with the WS before recovery (WS P ) is 67.517%, the improvement is 31.966%, and when SNR=0, the improvement is 42.3%. From the perspective of signal demodulation symbol error rate, it can be seen that before recovery, due to the presence of interference signals, the SER of demodulating the target signal is extremely high, and the communication quality is extremely poor. After recovery by the present invention, the signal demodulation SER (SER E ) is significantly improved. When SNR=0.5, the SER after recovery (SER E ) were reduced to 3.749×10 -3 and 1.2022×10 -3 , compared with the initial SER (SER P ) is reduced by two orders of magnitude, and when SNR>=10, SER is reduced by three orders of magnitude.
[0103] As can be seen from Table 2, under different interference signal strengths SIR, the present invention can still ensure high-quality recovery of the target signal waveform. The average waveform similarity of the target signal after recovery reaches 99.958%, which is 57.189% higher than the average waveform similarity of 42.769% before recovery. From the perspective of signal demodulation symbol error rate, it can be found that as the interference strength increases, the demodulation symbol error rate SER (SER E ) has increased, but still remains at 10 -4 SER of the order of magnitude, compared to the SER before recovery (SER P ), an average decrease of 3 orders of magnitude.
[0104] This simulation experiment 1 verifies the interference suppression capability of the proposed method under different environmental noise levels and interference signal strengths. The experimental results show that the present invention can still maintain good interference suppression performance under different SNRs and SIRs. In addition, in order to test the adaptability of the proposed method to changes in interference types, the interference signal is changed to QPSK, PAM4, and 2FSK. The experiment is similar to the above results and will not be repeated here.
[0105] Simulation experiment 2: Performance comparison between the present invention and existing interference suppression methods
[0106] In this simulation experiment 2, the performance of the proposed method is compared with that of the existing interference suppression method. The existing methods involved in the comparison include: m. a single signal recovery algorithm based on sparse coding and template matching, n. a single signal recovery algorithm based on dictionary learning and sparse coding under constraints, and the test data and parameter settings are the same as those in simulation experiment 1. Figure 7 and Figure 8Demonstrate different methods to test BER and WS under different interference signal strengths, among which SER A and WS A Representation method m performance, SER B and WS B Indicates method n performance.
[0107] from Figure 7 and Figure 8 It can be seen that the existing methods m and n are proposed under the assumption that SSR exists. Their performance deteriorates sharply when there is no target signal SSR in the aliasing observation, and as the interference signal strength increases, they gradually lose the ability to suppress interference. The main reason is that methods m and n are target signal subspace bases established based on completely clean target signal SSR. If there is no target signal SSR, the base constructed by them will not be able to characterize the target signal. In addition, when the interference signal strength is low, the performance of method n is slightly better than that of method m. This is because method n introduces the characteristics of the target signal as a constraint condition. When the interference signal strength is low, part of the target signal information can be extracted, but as the interference strength increases, the power of the target signal in the aliasing observation will gradually decrease, eventually leading to the failure of the algorithm. In contrast, the present invention can still maintain robust interference suppression performance in low SIR scenarios. At the same time, considering that the actual interference signal is mostly uncontrollable, the possibility of a completely clean SSR is low. The communication scenario addressed by the present invention is more in line with reality and has a broader application prospect.
[0108] Simulation experiment 3: Generalization ability test and analysis of the present invention
[0109] Since in practice, the parameters of the interference signal and the channel environment may be unknown and changeable, the ability of the present invention to cope with generalized test conditions is further tested here. The training data parameters of this simulation experiment 3 are the same as those of simulation experiment 1, while the test data settings are changed in two ways: first, considering the different durations of the interference signals, the durations of the three interference signals are set to 0-4.5ms, 4-7.5ms, and 5.5-10ms; second, considering the changes in the intensity of environmental noise interference, the initial SNR is set to 15dB. Except for the parameter changes, the remaining parameters are the same as those in simulation experiment 1.
[0110] Fig. 9 The test performance of the present invention under the above generalization conditions is shown, where WS 1E and SER 1E represents the test performance under the scenario of changing duration, WS 2E and SER 2E Indicates the test performance at SNR = 15dB, WS 1P SER 1P , WS2P and SER 2P Respectively represent the performance of direct calculation of the corresponding original data. By comparing with the results of the closed set test (i.e., the test in which the parameters of the test data set are consistent with the training data set), it can be seen that the present invention copes with all the above-mentioned generalization conditions better. This excellent generalization ability can be mainly attributed to the method of constructing the sparse domain of the target signal. The sparse domain constructed based on data-driven adaptive construction makes the target signal sufficiently sparse, so the present invention has a strong adaptability to changes in interference parameters. It should also be pointed out that if different SNRs are considered during training, the performance of the interference suppression method proposed in the present invention at a lower test SNR will be predictably improved. This can be achieved simply by generating training data under a wider range of SNRs.
[0111] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A data-driven SCA interference suppression method, characterized in that: The following steps are involved: Construct an asynchronous non-stationary interference channel model to obtain aliasing observations of the receiving node; The aliased observation is subjected to a sparse transformation so that the target signal has sparseness in the transform domain; wherein the calculation formula of the sparse transformation is: W i =f Encoder (X i ) in: X i =A i S i +V i In the formula, represents the aliased observation at the receiving node i; represents the aliasing matrix, where each column represents the channel coefficient between the sending node j and the receiving node i; represents the signal matrix, each row represents the signal transmitted by the sending node j received at the receiving node i; represents the AWGN at receiving node i; is the time domain aliasing mixed measurement X i The deep high-dimensional representation of K is the number of encoder filters, and L is the output length of a single filter; f Encoder (·) represents the mapping function of the encoder; The calculation formula for recovering the target signal in the sparse domain is: In the formula, is the estimation of the target signal in the sparse domain; is the time domain aliasing mixed measurement X i Deep high-dimensional representation of is the representation matrix of the target signal in the sparse domain, which is estimated by the separator; Represents a matrix dot multiplication operation; Complete the representation matrix estimation of the target signal in the transform domain to obtain the representation matrix estimation result; Based on the representation matrix estimation result, the target signal is restored in the transform domain to obtain a restoration result; The recovery result is converted to the time domain by inverse transformation to complete interference suppression; wherein the calculation formula of the inverse transformation is: is the estimate of the target signal after recovery in the time domain; f Decoder (·) represents the mapping function of the decoder; is the estimation of the target signal in the sparse domain.
2. The data-driven SCA interference suppression method according to claim 1, characterized in that: The aliasing observation is: In the formula, a ii is the channel gain between the sending node i (i = 1, 2, ..., N) and the receiving node i (i = 1, 2, ..., N); s i (nm i ) is the transmission delay m i The signal transmitted by the i-th sending node in the sampling interval; a ji is the channel gain between the sending node j (j = 1, 2, ..., N) and the receiving node i (i = 1, 2, ..., N); s j (nm j ) indicates that the transmission delay is m j The signal transmitted by the jth sending node in the sampling interval; v i (n) represents the additive white Gaussian noise at the receiving node i; T is the total number of sampling points.
3. The data-driven SCA interference suppression method according to claim 1, characterized in that: The asynchronous non-stationary interference channel model includes N main communication links and N(N-1) interference links.
4. The data-driven SCA interference suppression method according to claim 1, characterized in that: The performance of the recovery results is evaluated, including the waveform similarity of the recovered target signal and the demodulation symbol error rate of the recovered target signal.
5. The data-driven SCA interference suppression method according to claim 4, characterized in that: The calculation formula of the waveform similarity is: In the formula, ||·|| represents the 2-norm; when hour, c is a non-zero constant; when with a ii s ii When orthogonal, The closer it is to 1, the The higher the restoration quality, the smaller the error.
6. The data-driven SCA interference suppression method according to claim 4, characterized in that: The calculation formula of the demodulation symbol error rate is: In the formula, is the number of symbols in the symbol sequence obtained by demodulating the target signal before or after recovery that are inconsistent with the real symbol sequence; Q i is the total number of target signal symbol sequences.
Citation Information
Patent Citations
Inter-cell interference mitigation method using spatial covariance matrix estimation method for inter-cell interference mitigation of MIMO antenna OFDM system
US20120063529A1
Interference suppression using modal moment estimates
WO1999039445A1
Interference mitigation in heterogeneous wireless communication networks
WO2011057037A2
