CN115051895B - Blind estimation method and system for spread spectrum sequence combining M estimation and K-means algorithm - Google Patents
Blind estimation method and system for spread spectrum sequence combining M estimation and K-means algorithm Download PDFInfo
- Publication number
- CN115051895B CN115051895B CN202210720245.6A CN202210720245A CN115051895B CN 115051895 B CN115051895 B CN 115051895B CN 202210720245 A CN202210720245 A CN 202210720245A CN 115051895 B CN115051895 B CN 115051895B
- Authority
- CN
- China
- Prior art keywords
- samples
- sample
- signal
- estimation
- cluster
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 45
- 238000001228 spectrum Methods 0.000 title claims abstract description 30
- 238000004891 communication Methods 0.000 claims abstract description 41
- 239000013598 vector Substances 0.000 claims abstract description 39
- 238000005070 sampling Methods 0.000 claims abstract description 12
- 230000011218 segmentation Effects 0.000 claims abstract description 10
- 238000009826 distribution Methods 0.000 claims description 18
- 239000011159 matrix material Substances 0.000 claims description 18
- 230000007480 spreading Effects 0.000 claims description 12
- 230000001360 synchronised effect Effects 0.000 claims description 8
- 230000000694 effects Effects 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 230000001629 suppression Effects 0.000 claims description 5
- 230000003321 amplification Effects 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 4
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 4
- 230000005540 biological transmission Effects 0.000 claims description 3
- 238000009827 uniform distribution Methods 0.000 claims description 3
- 239000006185 dispersion Substances 0.000 claims description 2
- 229940088594 vitamin Drugs 0.000 claims 1
- 229930003231 vitamin Natural products 0.000 claims 1
- 235000013343 vitamin Nutrition 0.000 claims 1
- 239000011782 vitamin Substances 0.000 claims 1
- 150000003722 vitamin derivatives Chemical class 0.000 claims 1
- 238000004088 simulation Methods 0.000 description 7
- 238000000354 decomposition reaction Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000007476 Maximum Likelihood Methods 0.000 description 3
- 239000000654 additive Substances 0.000 description 3
- 230000000996 additive effect Effects 0.000 description 3
- 230000006872 improvement Effects 0.000 description 2
- 230000000737 periodic effect Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
- H04L25/03006—Arrangements for removing intersymbol interference
- H04L25/03178—Arrangements involving sequence estimation techniques
- H04L25/03203—Trellis search techniques
- H04L25/03216—Trellis search techniques using the M-algorithm
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B1/00—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission
- H04B1/69—Spread spectrum techniques
- H04B1/707—Spread spectrum techniques using direct sequence modulation
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B1/00—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission
- H04B1/69—Spread spectrum techniques
- H04B1/707—Spread spectrum techniques using direct sequence modulation
- H04B1/7073—Synchronisation aspects
- H04B1/7087—Carrier synchronisation aspects
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L1/00—Arrangements for detecting or preventing errors in the information received
- H04L1/0001—Systems modifying transmission characteristics according to link quality, e.g. power backoff
- H04L1/0036—Systems modifying transmission characteristics according to link quality, e.g. power backoff arrangements specific to the receiver
- H04L1/0038—Blind format detection
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
- H04L25/03006—Arrangements for removing intersymbol interference
- H04L25/03012—Arrangements for removing intersymbol interference operating in the time domain
- H04L25/03019—Arrangements for removing intersymbol interference operating in the time domain adaptive, i.e. capable of adjustment during data reception
- H04L25/03082—Theoretical aspects of adaptive time domain methods
- H04L25/03089—Theory of blind algorithms, recursive or not
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Power Engineering (AREA)
- Quality & Reliability (AREA)
- Theoretical Computer Science (AREA)
- Noise Elimination (AREA)
Abstract
A blind estimation method and system for a spread spectrum sequence combining M estimation and K-means algorithm comprises the following steps: establishing a model, wherein the model is a non-cooperative communication scene in a short code direct spread spectrum communication system; adding impulse noise into the model, and modeling the impulse noise; the method comprises the steps of obtaining received signals of a receiver and a non-cooperative node, and modeling the received signals; sampling and dividing the received signals to obtain signal vectors and signal matrixes; forming a sample set; after the sample set is obtained, the sample set is placed into a classifier to be classified, the samples are assigned according to classification results, and then the pseudo code sequence is estimated by splicing according to the sequence of the previous segmentation. The invention improves the robustness and stability of the algorithm and greatly improves the blind estimation performance of the pseudo code sequence in the impulse noise environment. The invention can obtain better performance under the conditions of the same equivalent signal to noise ratio and worse noise condition, namely more frequent pulse impact and larger impact amplitude.
Description
Technical Field
The invention belongs to the technical field of cooperative communication, and particularly relates to a method and a system for blind estimation of a spread spectrum sequence by combining M estimation and a K-means algorithm.
Background
Because of direct spread spectrum Direct sequence spread spectrum, DSSS communication has the characteristics of high spectrum utilization rate, low interception probability, strong confidentiality, flexible configuration, multipath interference resistance and the like, and is widely applied to the military and civil fields. At the same time, these advantages present a significant challenge to how non-cooperative nodes access a direct-expansion communication network. For both parties of the cooperative communication, the time, parameters, pseudo code sequences used for spreading, etc. of the communication are known, and they can easily recover the original information in the received signal. For a non-cooperative node that wants to access the communication network, the original information of the received signal cannot be recovered under the condition of unknown spreading sequences, and then the estimation of the pseudo code sequence is very necessary.
Up to now, there have been a lot of literature on pseudo code estimation of direct spread signals. The present source polynomial method utilizes the Massey algorithm to estimate the feedback coefficients of a linear feedback shift register that produces the pseudo-code sequence, and then estimates the pseudo-code sequence. However, the performance of this scheme can drop dramatically under low signal-to-noise conditions. The third-order correlation method refers to the third-order correlation theorem of the M sequence to estimate the pseudo code sequence, so that the method is only applicable to the condition that the pseudo code sequence is the M sequence. The maximum likelihood estimation method proposes a maximum likelihood function model for estimating the pseudo code sequence. The method based on maximum likelihood estimation has a great limitation because of the problem that the complexity increases sharply with the increase of the spread spectrum code length. The eigenvalue decomposition method obtains a main eigenvector and a secondary eigenvector by eigenvalue decomposition of a covariance matrix of the signal, and then estimates a pseudo code sequence by using the eigenvector containing all information of the pseudo code sequence. Common feature decomposition methods are EVD algorithm and SVD algorithm. Although the accuracy of this method is relatively high, the complexity is also very high and the applicability is not strong. According to the thought of subspace tracking, the literature utilizes subspace tracking to rapidly realize the characteristic decomposition process. The algorithm reduces the complexity of feature decomposition to some extent, but reduces the performance of the estimation. The neural network-based method proposes a constrained Hebbian rule to update the algorithm of the tap coefficients of the adaptive FIR filter, but cannot recover the signal well in a low signal-to-noise environment.
All of the above algorithms can degrade dramatically in impulse noise environments. Most algorithms are studied in an additive white gaussian noise environment because Additive White Gaussian Noise (AWGN) is easy to derive and analyze. However, in an actual communication system, there is generally a large amount of noise having a pulse characteristic in a channel, which causes a sharp decrease in performance of an estimation method suitable for a gaussian noise environment, and at the same time, the denser the noise impact, that is, the more severe the tail of the noise distribution, the more serious the performance decrease.
Disclosure of Invention
The invention aims to provide a blind estimation method and a blind estimation system for a spread spectrum sequence by combining M estimation and a K-means algorithm, which are used for solving the problem that the performance of the spread spectrum sequence of a direct spread spectrum signal in a pulse noise environment is seriously reduced.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
a blind estimation method of a spread spectrum sequence combining M estimation and K-means algorithm comprises the following steps:
establishing a model which is a non-cooperative communication scene in a short code direct spread spectrum communication system and comprises a pair of cooperative transceiver parties and a non-cooperative node;
adding impulse noise into the model, and modeling the impulse noise;
the method comprises the steps of obtaining received signals of a receiver and a non-cooperative node, and modeling the received signals;
sampling and dividing the received signals to obtain signal vectors and signal matrixes;
then dividing columns of the signal matrix to obtain signal samples, and forming a sample set;
after the sample set is obtained, the sample set is placed into a classifier to be classified, the samples are assigned according to classification results, and then the pseudo code sequence is estimated by splicing according to the sequence of the previous segmentation.
Furthermore, the transmission signals of both parties of the cooperative communication of the system model are modulated by a spread spectrum sequence, all parameters are known, and the non-cooperative nodes need to meet the requirement of accessing the communication network without affecting the communication among the cooperative communication nodes.
Further, an alpha stable distribution is selected to model impulse noise; the characteristic function of the alpha stable distribution is expressed as follows:
wherein, alpha epsilon (0, 2) is a characteristic factor, and represents the tail flick degree of a probability density function PDF, alpha is larger, the tail is narrower, the pulse amplitude is weaker, beta epsilon < -1,1 > is called a symmetrical parameter, which represents the symmetrical degree of random variable distribution, gamma >0 is called a dispersion coefficient, represents the degree of deviation of a sample value of a random variable from an average value of the random variable, alpha-infinity < mu < <infinityis called a position parameter, and exp (j mu t) represents the distance from the peak value of the PDF to a vertical axis.
Further, modeling of the signal: assuming that the received signal is strictly time synchronized and carrier synchronized, the received signal is expressed as:
r(t)=s(t)+n(t)
wherein { a }, a k = ±1, k∈z } is periodic T s Is subject to equal distribution; τ is the random time delay subject to uniform distribution; n (t) is the noise component in the received signal; { c i = ±1, i=1, 2, …, N } is a pseudo code sequence of length N, p (T) is a duration T c Has a chip pulse of T s =NT c 。
Further, the method for sampling and dividing the received signal to obtain a signal vector and a signal matrix specifically comprises the following steps:
assuming that the delay and the information code length and chip length are known, the signals are synchronized according to the delay and then the received signals are receivedSampling at equal intervals, the sampling interval being T c The method comprises the steps of carrying out a first treatment on the surface of the Then, carrying out signal segmentation according to the information code length to obtain M signal vectors; each signal vector contains complete pseudo code sequence information; the kth signal vector is expressed as:
r k =a k h+n k
wherein { a }, a k The = ±1} is an information code, h= [ c ] 1 ,c 2 ,…,c N ]Is an N-dimensional pseudo code sequence vector, N k Is impulse noise in the signal; splicing the M signal vectors into an M x N-dimensional signal matrix R, wherein n= [ N ] 1 ,n 2 ,…,n M ]Is an M x N dimensional noise matrix;
further, the method comprises the steps of,
each column of the matrix is treated as a sample to obtain a sample set containing N samples The i-th sample vector is expressed as:
x i =c i a+n i
wherein a= [ a ] 1 ,a 2 ,…,a M ]Is an information sequence; { c i = ±1} is the spreading code corresponding to the sample vector, n i An impulse noise component added to the sample vector; recording the sequence obtained by dividing the sample, and obtaining a sample x i The smaller the value of i, the earlier the order.
Further, the classifier is a joint M estimation and K-means classifier, and specifically comprises:
1) Classifier input: after obtaining a sample set, sending the samples into a classifier for clustering; grouping samples into two classes, the classifier being input as a sample set containing N samplesClassifier cluster number k=2; setting an iteration convergence threshold eta and a maximum iteration number max;
2) Initial cluster center selection: from the slaveRandomly selects 2 samples as 2 initial cluster centers o 1 =[o 11 ,o 12 ,…,o 1M ],o 2 =[o 21 ,a 22 ,…,a 2M ];
3) Noise energy σ estimation: after the classifier obtains the centers of the two class clusters, estimating noise energy; first, two kinds of cluster centers are calculated to o 1 ,o 2 An average Euclidean distance S of (2); if o 1 ,o 2 For iteration get, there areIf o 1 ,o 2 The following are obtained for initial random:
due to c i ,c j Only two values of-1 and 1 are adopted, noise is symmetrically distributed, and the information sequences are distributed uniformly, so that the method comprises the following steps:
because ofTherefore use->To estimate the noise energy.
4) Calculating the similarity between the sample and the centers of the two class clusters: in a similarity calculation part of the samples, measuring the similarity between the samples by using the distance between the samples; the farther the distance between the two is, the lower the similarity is; the closer the distance between the two is, the higher the similarity is; the distance formula between the sample vector x with the dimension of N and the center of the kth class cluster is as follows:
the Huber loss function commonly used in M estimation is introduced to suppress noise so as to counteract the amplification effect of square sum operation in a formula on pulse impact; wherein the expression of the Huber function is:
5) Clustering and labeling of samples: clustering and tag skimming are carried out according to the similarity between the sample and the centers of different clusters; when the sample and the cluster are centered o 1 When the similarity of the (C) is highest, the label is set as o 1 Otherwise set to o 2 ;
6) Updating the cluster-like center: after obtaining sample labels of all samples, the cluster centers o of two clusters are connected according to label pairs of the samples 1 ,o 2 Updating; the thought of the location parameter in M estimation is utilized, and the cluster center o is utilized according to the following formula k Updating, wherein k is 1 or 2;
i.e. by tagging all samples as o k To obtain an updated cluster-like center o by calculating a weighted average of the sample vectors of (a) k Simultaneously, the cluster center when not updated is recorded as o' k The method comprises the steps of carrying out a first treatment on the surface of the By using the method in step 4)As an adaptive weighting function, the threshold of noise suppression is the estimated noise energy sigma;
7) Iteration termination condition: when the iteration number of the classifier does not reach the maximum iteration number max set in 1) or the Euclidean distance between the center of the current class cluster and the center of the class cluster in the last iteration is larger than the convergence threshold eta set in 1), namely II o h -o′ k ‖<η, repeating steps 3) to 6) until the above iteration termination condition is satisfied.
Further, the method comprises the steps of assigning a value to the sample according to the classification result, and then splicing according to the sequence of the previous segmentation so as to estimate and obtain a pseudo code sequence, and specifically comprises the following steps:
after the sample set passes through the classifier, all samples have been labeled o 1 Or o 2 The method comprises the steps of carrying out a first treatment on the surface of the Label o 1 Assigning a sample of (1) to be-1 and a label to be o 2 The samples of the signal are assigned a value of +1, and then are spliced according to the sample sequence in the sample vectorization of the signal; obtaining an array with the length of N and the element value of +1 or-1, wherein the array is the original code or the inverse code of the pseudo code sequence; and performing inverse processing on the array according to actual conditions or keeping the array unchanged to estimate and obtain a pseudo code sequence.
Further, a blind estimation system for a spread spectrum sequence combining M estimation and K-means algorithm comprises:
the model building module is used for building a model which is a non-cooperative communication scene in the short code direct spread spectrum communication system and comprises a pair of cooperative transceiver parties and a non-cooperative node; adding impulse noise into the model, and modeling the impulse noise;
the signal acquisition module is used for acquiring received signals of the receiver and the non-cooperative node and modeling the received signals;
the signal vectorization module is used for sampling and dividing the received signals to obtain signal vectors and signal matrixes; then dividing columns of the signal matrix to obtain signal samples, and forming a sample set;
and the classification module is used for classifying the sample set in a classifier after the sample set is obtained, assigning the value to the sample according to the classification result, and then splicing according to the sequence of the previous segmentation so as to estimate and obtain the pseudo code sequence.
Compared with the prior art, the invention has the following technical effects:
the invention includes sample vectorization; classifiers (e.g., k-means); and (5) distributing and splicing sample labels based on the clustering result. The alpha stable distribution is used to model impulse noise. It is assumed that the received signals are synchronized. Firstly vectorizing and splitting a signal into samples, then clustering the samples by using a joint M estimation and K-means classifier, and finally distributing assignment and splicing sample labels according to a clustering result to obtain an estimation result of a code sequence.
For a non-cooperative communication scene in a direct spread spectrum communication network, modeling is carried out on signals received by non-cooperative nodes, and meanwhile, modeling is carried out on noise by utilizing alpha stable distribution, and most pulse impact conditions can be covered through adjustment of the distribution parameters. Aiming at the problem of pseudo code sequence estimation in the scene, the invention provides a cluster-based pseudo code sequence blind estimation framework which adapts to different classification conditions by changing the type of a classifier. Under the framework, the invention provides an algorithm combining M estimation and k-means, and noise impact is suppressed adaptively according to the noise level of a sample in the iterative process of the algorithm, so that the robustness and stability of the algorithm are improved, and the blind estimation performance of a pseudo code sequence in a pulse noise environment is greatly improved. Meanwhile, unlike other methods, the invention can obtain better performance under the conditions of the same equivalent signal-to-noise ratio but worse noise condition, namely more frequent pulse impact and larger impact amplitude. This is of great significance for its application in uncooperative communication scenarios in impulse noise environments.
Drawings
Fig. 1 is a schematic diagram of a non-cooperative communication architecture in a direct spread spectrum communication network.
Fig. 2 is a schematic diagram of a cluster-based pseudo code sequence blind estimation framework.
Fig. 3 to 7 are graphs showing simulation verification results of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings:
referring to fig. 1 to 7, a blind estimation method for a spread spectrum sequence combining M estimation and K-means algorithm includes:
1. establishing a system model: the system model is set as a non-cooperative communication scene in a short code direct spread spectrum communication system, and comprises a pair of cooperative transceiver parties and a non-cooperative node. The transmission signals of both parties of the cooperative communication are modulated by a spread spectrum sequence and all parameters are known. Non-cooperative nodes need to access the communication network without affecting the communication between the cooperative communication nodes.
2. Modeling of noise: the alpha stable distribution is chosen to model impulse noise. The alpha stable distribution belongs to generalized Gaussian distribution, is the only distribution model conforming to the generalized center limit theorem, and is a statistical model capable of describing different pulse degrees, so that pulse noise can be well described. The characteristic function of the alpha stable distribution is expressed as follows:
where α.epsilon.0, 2] is a characteristic factor representing the tail flick degree of a Probability Density Function (PDF), the larger α represents the narrower the tail, the weaker the pulse amplitude.
3. Modeling of signals: assuming that the received signal is tightly time synchronized and carrier synchronized, the received signal can be expressed as:
r(t)=s(t)+n(t)
wherein { a }, a k = ±1, k∈z } is periodic T s Is subject to equal distribution; τ is the random time delay subject to uniform distribution; n (t) is the noise component in the received signal. { c i = ±1, i=1, 2, …, N } is a pseudo code sequence of length N, p (T) is a duration T c Has a chip pulse of T s =NT c 。
4. Sample vectorization of signals: in order not to lose generality, it is assumed that the delay and information code length and chip length are known. Synchronizing the signals according to the time delay, and then sampling the received signals at equal intervals, wherein the sampling interval is T c . And then, carrying out signal segmentation by using the information code length to obtain M signal vectors. Each signal vector contains complete pseudocode sequence information. The kth signal vector may be expressed as:
r k =a k h+n k
wherein { a }, a k The = ±1} is an information code, h= [ c ] 1 ,c 2 ,…,c N ]Is an N-dimensional pseudo code sequence vector, N k Is impulse noise in the signal. Splicing the M signal vectors into an M x N-dimensional signal matrix R, wherein n= [ N ] 1 ,n 2 ,…,n M ]Is an M x N dimensional noise matrix.
Using each column of the matrix as a sample can obtain a sample set containing N samplesThe i-th sample vector can be expressed as:
x i =c i a+n i
wherein a= [ a ] 1 ,a 2 ,…,a M ]Is an information sequence. { c i = ±1} is the spreading code corresponding to the sample vector, n i Which is an additive impulse noise component on the sample vector. Here, the sequence of sample division is recorded, and sample x i The smaller the value of i, the earlier its order is considered.
5. Combining M-estimates and K-means classifiers
1) Classifier input: after obtaining the sample set, the samples need to be fed into a classifier for clustering. From the expression of the sample vector, the main difference between samples depends on the spreading code c i And the spreading codes have only two choices-1 and 1, samples need to be grouped into two classes. The classifier input is thus a sample set containing N samplesAnd classifier cluster number k=2. Since the classifier is continuously approximated to an ideal result through iteration, the iteration convergence threshold η and the maximum iteration number max also need to be set here.
2) Initial cluster center selection: from the idea of the conventional K-means algorithmRandomly selects 2 samples as 2 initial cluster centers o 1 =[o 11 ,o 12 ,…,o 1M ],o 2 =[o 21 ,a 22 ,…,a 2M ]。
3) Noise energy σ estimation: after the classifier gets the centers of the two clusters, the noise energy needs to be estimated in order to be able to suppress the impact of impulse noise dynamically according to the noise level. First, two kinds of cluster centers are calculated to o 1 ,o 2 Is a mean euclidean distance S. If o 1 ,o 2 For iteration get, there areIf o 1 ,o 2 The following are obtained for initial random:
due to c i ,c j Only two values of-1 and 1 are adopted, noise is symmetrically distributed, and the information sequences are distributed uniformly, so that the method comprises the following steps:
because ofSo can use +>To estimate the noise energy.
4) Calculating the similarity between the sample and the centers of the two class clusters: in the similarity calculation section of samples, it is necessary to represent a difference between samples of different types, where the inter-sample similarity is measured in terms of the inter-sample distance. The farther the distance between the two is, the lower the similarity is; the closer the two are, the higher the similarity. The distance formula between the sample vector x with the dimension of N and the center of the kth class cluster is as follows:
here, the Huber loss function commonly used in M estimation is introduced to suppress noise to cancel the amplification of the impulse impact by the sum-of-squares operation in the formula. Wherein the expression of the Huber function is:
5) Clustering and labeling of samples: at this step, clustering and tagging are required based on the similarity of samples to the centers of different clusters. When the sample and the cluster are centered o 1 When the similarity of the (C) is highest, the label is set as o 1 Otherwise set to o 2 。
6) Updating the cluster-like center: after obtaining the sample labels of all samples, the cluster centers o of two clusters need to be connected according to the label pairs of the samples 1 ,o 2 And updating. The thought of the location parameter in M estimation is utilized, and the cluster center o is utilized according to the following formula k Updating is carried out, and k is 1 or 2.
I.e. by tagging all samples as o k To obtain an updated cluster-like center o by calculating a weighted average of the sample vectors of (a) k Simultaneously, the cluster center when not updated is recorded as o' k . To suppress the effect of noise, the Huber function in step 4) is used as an adaptive weighting function, and the noise suppression threshold is the estimated noise energy σ.
7) Iteration termination condition: when the iteration number of the classifier does not reach the maximum iteration number max set in 1) or the Euclidean distance between the center of the current class cluster and the center of the class cluster in the last iteration is larger than the convergence threshold eta set in 1), namely ||o k -o′ k Repeating steps 3) to 6) until the above iteration termination condition is satisfied when || < η.
6. Assignment and concatenation of sample tag pairs: after the sample set passes through the classifier, all samples have been labeled o 1 Or o 2 . Label o 1 Assigning a sample of (1) to be-1 and a label to be o 2 Is assigned a value of +1 and then spliced according to the sample order in step 4 (sample vectorization of the signal).
After the operation is completed, an array with the length of N and the element value of +1 or-1 can be obtained, wherein the array is the original code or the inverse code of the pseudo code sequence. And performing inverse processing on the array according to actual conditions or keeping the array unchanged to estimate and obtain a pseudo code sequence.
Fig. 1 is a schematic diagram of a non-cooperative communication architecture in a direct spread spectrum communication network. Consider a communication network with non-cooperative nodes in which a transmitter and a receiver communicate cooperatively with each other, with only one antenna being provided at the transmitter and non-cooperative nodes. The non-cooperative nodes want to access the communication network without affecting the normal communication of the sender and the receiver. s (t) is the direct spread spectrum signal transmitted by the transmitter. The channel between the transmitter and the non-cooperating node is a impulse noise channel. The impulse noise n (t) contains many discrete, irregular pulses or short-time large amplitude noise spikes.
Fig. 2 is a schematic diagram of a cluster-based pseudo code sequence blind estimation framework. The frame is divided into three parts: signal sample vectorization; classifiers (e.g., k-means); and (5) distributing and splicing sample labels based on the clustering result. After receiving the signals, the signals are sampled and divided to obtain signal vectors and signal matrixes. Then, the columns of the signal matrix are segmented to obtain signal samples, and a sample set is formed. After the sample set is obtained, the sample set is placed into a classifier to be classified, the samples are assigned according to classification results, and then the pseudo code sequence is estimated by splicing according to the sequence of the previous segmentation.
To demonstrate the suppression effect of the present invention on impulse noise, fig. 3 shows a partial pseudo code estimation result of the present invention, and fig. 4 shows a partial pseudo code estimation result of the EVD algorithm with the highest accuracy in the conventional algorithm under the same condition. It can be seen that if the pseudo code sequence is estimated using the EVD algorithm, there will be a pulse component in the estimation result. Therefore, the values of other components are very small and are more easily influenced by noise, so that the discrimination of pseudo code sequences is not facilitated, and the overall accuracy is not high. The real sequence and the estimated sequence in fig. 3 are completely matched, so that the estimation effect is good.
In order to verify the pseudo code sequence estimation performance of the present invention, fig. 5 shows simulation results of the proposed algorithm, EVD algorithm, conventional k-means algorithm, and cosine distance-based k-means algorithm. By comparing the simulation curves of the various algorithms, the proposed algorithm is significantly superior to other algorithms. Specifically, when gsnr= -4dB, the estimation error rate of the proposed algorithm is 0.008, which is significantly better than 0.3 of EVD. The simulation is performed under the condition of alpha=1.6, namely, the noise is not frequently impacted and the impact amplitude is not high, while the performance of the algorithm is greatly improved under the condition of alpha reduction, namely, the noise is frequently impacted and the impact amplitude is higher, and the performance of other algorithms is greatly reduced. Fig. 6 and 7 verify this. Fig. 6 shows simulation results of the algorithm under different α conditions, and it can be seen that the smaller α is, the more obvious the performance improvement of the algorithm is under the condition of low signal-to-noise ratio. Fig. 7 is a simulation result of the other three algorithms under different α conditions, and the overall trend is that the performance is continuously reduced with the decrease of α.
The simulation results show that the method has obvious advantages under the impulse noise condition and has obvious improvement compared with other algorithms. Meanwhile, the noise condition is worse and the performance is better under the same signal to noise ratio.
Claims (8)
1. A blind estimation method of a spread spectrum sequence combining M estimation and K-means algorithm is characterized by comprising the following steps:
establishing a model which is a non-cooperative communication scene in a short code direct spread spectrum communication system and comprises a pair of cooperative transceiver parties and a non-cooperative node;
adding impulse noise into the model, and modeling the impulse noise;
the method comprises the steps of obtaining received signals of a receiver and a non-cooperative node, and modeling the received signals;
sampling and dividing the received signals to obtain signal vectors and signal matrixes;
then dividing columns of the signal matrix to obtain signal samples, and forming a sample set;
after the sample set is obtained, the sample set is put into a classifier to be classified, the samples are assigned according to classification results, and then the pseudo code sequence is estimated and obtained by splicing according to the sequence of the previous segmentation;
the classifier is a joint M estimation and K-means classifier, and specifically comprises:
1) Classifier input: after obtaining a sample set, sending the samples into a classifier for clustering; the samples are gathered into two classes, and the classifier is input to containSample set of individual samples->And classifier cluster number +.>The method comprises the steps of carrying out a first treatment on the surface of the Setting an iteration convergence threshold +.>And maximum number of iterations->;
2) Initial cluster center selection: from the slaveIs selected randomly->Samples are taken as->Initial cluster-like centers;
3) Noise energyEstimating: after the classifier obtains the centers of the two class clusters, estimating noise energy; first two kinds of cluster centers are calculated +.>Average Euclidean distance>The method comprises the steps of carrying out a first treatment on the surface of the If->For iteration get, there is->The method comprises the steps of carrying out a first treatment on the surface of the If->The following are obtained for initial random:
wherein:
is impulse noise in the signal;
due to,/>Only->And->Two kinds of values are symmetrically distributed, the noise is equally distributed, and the two kinds of values are:
because ofTherefore use +.>To estimate noise energy;
4) Calculating the similarity between the sample and the centers of the two class clusters: in a similarity calculation part of the samples, measuring the similarity between the samples by using the distance between the samples; the farther the distance between the two is, the lower the similarity is; the closer the distance between the two is, the higher the similarity is; the dimension isSample vector +.>And->The distance formula of the centers of the individual clusters is:
the Huber loss function commonly used in M estimation is introduced to suppress noise so as to counteract the amplification effect of square sum operation in a formula on pulse impact; wherein the expression of the Huber function is:
5) Clustering and labeling of samples: clustering and tag skimming are carried out according to the similarity between the sample and the centers of different clusters; when the sample and the cluster are centeredWhen the similarity of (C) is highest, the tag is set to +.>Otherwise set to->;
6) Updating the cluster-like center: after obtaining sample labels of all samples, the center of the class cluster connected with the two class clusters is paired according to the labels of the samplesUpdating; by using the idea of the position parameter in the M estimation, the cluster center is +.>Update->1 or 2;
i.e. by tagging all samples asTo obtain an updated cluster center +.>At the same time, the cluster center when not updated is recorded as +.>The method comprises the steps of carrying out a first treatment on the surface of the Using the Huber function in step 4) as an adaptive weighting function, the noise suppression threshold being the estimated noise energy +.>;
7) Iteration termination condition: setting the maximum iteration number in 1) when the iteration number of the classifier does not reachOr the Euclidean distance between the center of the current cluster and the center of the last iteration cluster is greater than the convergence threshold value set in 1)>When, i.eWhen the iteration termination condition is satisfied, repeating the steps 3) to 6).
2. The method for blind estimation of spreading sequences by combining M estimation and K-means algorithm according to claim 1, wherein the transmission signals of both parties of cooperative communication of the system model are modulated by the spreading sequences and the parameters are known to each other, and the non-cooperative nodes need to meet the requirement of accessing into the communication network without affecting the communication between the cooperative communication nodes.
3. A method for blind estimation of a spreading sequence combining M estimation and K-means algorithm as claimed in claim 1, wherein the selection of the method is performed byStabilizing the distribution to model impulse noise; />The characteristic function of the stable distribution is expressed as follows:
wherein,is a characteristic factor representing the tail flick degree of the probability density function PDF, +.>The larger the tail, the narrower the tail, and the weaker the pulse amplitude; />Referred to as symmetry parameters, which represent the degree of symmetry of the random variable distribution; />Referred to as a dispersion coefficient, which indicates the degree to which the sample value of the random variable deviates from its average value; />The location parameter(s) is (are),representing the distance of the peak of the PDF to the vertical axis.
4. The method for blind estimation of a spreading sequence combining M estimation and K-means algorithm according to claim 1, wherein modeling of the signal: assuming that the received signal is strictly time synchronized and carrier synchronized, the received signal is expressed as:
wherein,is of period +.>Is subject to equal distribution; />Is a random time delay subject to uniform distribution; />Is a noise component in the received signal; />Is of length +.>Pseudo code sequence of>Is of duration +.>Has ∈k (chip pulse)>。
5. The method for blind estimation of spread spectrum sequences combining M estimation and K-means algorithm according to claim 1, wherein the steps of sampling and dividing the received signal to obtain a signal vector and a signal matrix comprise:
assuming that the delay and the information code length and the chip length are known, synchronizing the signals according to the delay, and then sampling the received signals at equal intervals with the sampling interval beingThe method comprises the steps of carrying out a first treatment on the surface of the Then signal division is carried out by using the information code length to obtain +.>A signal vector; each signal vector contains complete pseudo code sequence information; first->The individual signal vectors are expressed as:
wherein,is an information code->Is->Vitamin pseudo code sequence vector,>is impulse noise in the signal; will->The individual signal vectors are spliced to one +.>Signal matrix of dimension->Wherein->Is thatA dimensional noise matrix;
。
6. a method for blind estimation of a spreading sequence combining M-estimation and K-means algorithms according to claim 5, wherein,
each column of the matrix is treated as a sample to obtain a matrix containingSample set of individual samples->First->The individual sample vectors are expressed as:
wherein the method comprises the steps ofIs an information sequence; />For the spreading code to which the sample vector corresponds, and (2)>An impulse noise component added to the sample vector; recording the sequence obtained by dividing the sample, sample +.>Is->The smaller the value of (c), the earlier the order.
7. According to claim 1The blind estimation method of the spread spectrum sequence combining the M estimation and the K-means algorithm is characterized in that the assignment is carried out on samples according to the classification resultThen splicing according to the sequence of the previous segmentation so as to estimate and obtain a pseudo code sequence, which specifically comprises the following steps:
after the sample set passes through the classifier, all samples have been labeledOr->The method comprises the steps of carrying out a first treatment on the surface of the Label as +.>Is assigned a value of-1, and the label is +.>The samples of the signal are assigned a value of +1, and then are spliced according to the sample sequence in the sample vectorization of the signal; obtaining an array with the length of N and the element value of +1 or-1, wherein the array is the original code or the inverse code of the pseudo code sequence; and performing inverse processing on the array according to actual conditions or keeping the array unchanged to estimate and obtain a pseudo code sequence.
8. A blind estimation system for a spread spectrum sequence combining M estimation and K-means algorithm, comprising:
the model building module is used for building a model which is a non-cooperative communication scene in the short code direct spread spectrum communication system and comprises a pair of cooperative transceiver parties and a non-cooperative node; adding impulse noise into the model, and modeling the impulse noise;
the signal acquisition module is used for acquiring received signals of the receiver and the non-cooperative node and modeling the received signals;
the signal vectorization module is used for sampling and dividing the received signals to obtain signal vectors and signal matrixes; then dividing columns of the signal matrix to obtain signal samples, and forming a sample set;
the classifying module is used for classifying the sample set in the classifier after the sample set is obtained, assigning the value to the sample according to the classifying result, and then splicing according to the sequence of the previous segmentation so as to estimate and obtain a pseudo code sequence;
the classifier is a joint M estimation and K-means classifier, and specifically comprises:
1) Classifier input: after obtaining a sample set, sending the samples into a classifier for clustering; the samples are gathered into two classes, and the classifier is input to containSample set of individual samples->And classifier cluster number +.>The method comprises the steps of carrying out a first treatment on the surface of the Setting an iteration convergence threshold +.>And maximum number of iterations->;
2) Initial cluster center selection: from the slaveIs selected randomly->Samples are taken as->Initial cluster-like centers;
3) Noise energyEstimating: after the classifier obtains the centers of the two class clusters, estimating noise energy; first two kinds of cluster centers are calculated +.>Average Euclidean distance>The method comprises the steps of carrying out a first treatment on the surface of the If->For iteration get, there is->The method comprises the steps of carrying out a first treatment on the surface of the If->The following are obtained for initial random:
wherein:
is impulse noise in the signal;
due to,/>Only->And->Two kinds of values are symmetrically distributed, the noise is equally distributed, and the two kinds of values are:
because ofTherefore use +.>To estimate noise energy;
4) Calculating the similarity between the sample and the centers of the two class clusters: in a similarity calculation part of the samples, measuring the similarity between the samples by using the distance between the samples; the farther the distance between the two is, the lower the similarity is; the closer the distance between the two is, the higher the similarity is; the dimension isSample vector +.>And->The distance formula of the centers of the individual clusters is:
the Huber loss function commonly used in M estimation is introduced to suppress noise so as to counteract the amplification effect of square sum operation in a formula on pulse impact; wherein the expression of the Huber function is:
5) Clustering and labeling of samples: clustering and tag skimming are carried out according to the similarity between the sample and the centers of different clusters; when the sample and the cluster are centeredWhen the similarity of (C) is highest, the tag is set to +.>Otherwise set to->;
6) Updating the cluster-like center: after obtaining sample labels of all samples, the center of the class cluster connected with the two class clusters is paired according to the labels of the samplesUpdating; by using the idea of the position parameter in the M estimation, the cluster center is +.>Update->1 or 2;
i.e. by tagging all samples asTo obtain an updated cluster center +.>At the same time, the cluster center when not updated is recorded as +.>The method comprises the steps of carrying out a first treatment on the surface of the Using the Huber function in step 4) as an adaptive weighting function, the noise suppression threshold being the estimated noise energy +.>;
7) Iteration termination condition: setting the maximum iteration number in 1) when the iteration number of the classifier does not reachOr the Euclidean distance between the center of the current cluster and the center of the last iteration cluster is greater than the convergence threshold value set in 1)>When, i.eWhen the iteration termination condition is satisfied, repeating the steps 3) to 6).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210720245.6A CN115051895B (en) | 2022-06-23 | 2022-06-23 | Blind estimation method and system for spread spectrum sequence combining M estimation and K-means algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210720245.6A CN115051895B (en) | 2022-06-23 | 2022-06-23 | Blind estimation method and system for spread spectrum sequence combining M estimation and K-means algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115051895A CN115051895A (en) | 2022-09-13 |
CN115051895B true CN115051895B (en) | 2024-03-15 |
Family
ID=83162585
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210720245.6A Active CN115051895B (en) | 2022-06-23 | 2022-06-23 | Blind estimation method and system for spread spectrum sequence combining M estimation and K-means algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115051895B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1300999A1 (en) * | 2001-10-05 | 2003-04-09 | Siemens Information and Communication Networks S.p.A. | Channel estimation |
CN109150776A (en) * | 2018-08-15 | 2019-01-04 | 桂林电子科技大学 | A kind of short code direct sequence signal pseudo-code sequence blind estimating method |
CN110460354A (en) * | 2019-08-07 | 2019-11-15 | 桂林电子科技大学 | A kind of pseudo-code sequence blind estimate system |
CN112702080A (en) * | 2020-12-16 | 2021-04-23 | 中国人民解放军国防科技大学 | Direct sequence spread spectrum signal and pseudo code estimation method based on K-means algorithm |
-
2022
- 2022-06-23 CN CN202210720245.6A patent/CN115051895B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP1300999A1 (en) * | 2001-10-05 | 2003-04-09 | Siemens Information and Communication Networks S.p.A. | Channel estimation |
CN109150776A (en) * | 2018-08-15 | 2019-01-04 | 桂林电子科技大学 | A kind of short code direct sequence signal pseudo-code sequence blind estimating method |
CN110460354A (en) * | 2019-08-07 | 2019-11-15 | 桂林电子科技大学 | A kind of pseudo-code sequence blind estimate system |
CN112702080A (en) * | 2020-12-16 | 2021-04-23 | 中国人民解放军国防科技大学 | Direct sequence spread spectrum signal and pseudo code estimation method based on K-means algorithm |
Non-Patent Citations (3)
Title |
---|
叶 铃、等.基于 k-means 算法的直扩信号信息及 伪码序列盲估计方法.《信号处理》.2021,第第37卷卷(第第8期期),正文第2-4节. * |
基于 k-means 算法的直扩信号信息及 伪码序列盲估计方法;叶 铃、等;《信号处理》;第第37卷卷(第第8期期);正文第2-4节 * |
基于聚类的软扩频信号盲解扩方法;王航;郭静波;王赞基;;电子与信息学报(02);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN115051895A (en) | 2022-09-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wang et al. | TST-MUSIC for joint DOA-delay estimation | |
CN110113288B (en) | Design and demodulation method of OFDM demodulator based on machine learning | |
CN109951214B (en) | Signal detection method suitable for large-scale MIMO system | |
US11982698B2 (en) | Joint denoising and delay estimation for the extraction of pulse-width of signals in RF interference | |
CN112910811B (en) | Blind modulation identification method and device under unknown noise level condition based on joint learning | |
CN112115821B (en) | Multi-signal intelligent modulation mode identification method based on wavelet approximate coefficient entropy | |
CN104794264B (en) | A kind of radar communication waveform design method based on sparse frequency | |
CN111901069B (en) | Multi-user detection method based on neural network and approximate message transfer algorithm | |
CN109768816B (en) | non-Gaussian noise 3D-MIMO system data detection method | |
CN114268388A (en) | Channel estimation method based on improved GAN network in large-scale MIMO | |
Kalphana et al. | Convolutional Neural Network Auto Encoder Channel Estimation Algorithm in MIMO-OFDM System. | |
CN115883297A (en) | Information source parameter estimation-based super-large-scale MIMO mixed field channel estimation method | |
CN112702080A (en) | Direct sequence spread spectrum signal and pseudo code estimation method based on K-means algorithm | |
CN109547376A (en) | A kind of Modulation Identification method of the gauss hybrid models based on modulating characteristic | |
Pankratov et al. | Linear and nonlinear Chebyshev iterative demodulation algorithms for MIMO systems with large number of antennas | |
CN109194422B (en) | SNR estimation method based on subspace | |
CN108566227B (en) | Multi-user detection method | |
CN115051895B (en) | Blind estimation method and system for spread spectrum sequence combining M estimation and K-means algorithm | |
CN110233808B (en) | FTN system signal detection method | |
Mukhopadhyay et al. | Blind system identification in noise using a dynamic-based estimator | |
CN107465472B (en) | Multipath time delay estimation method based on path synthesis | |
Majumder | One-bit spectrum sensing using Gustafson–Kessel fuzzy clustering for cognitive radio network | |
Seidel et al. | A novel approach for combining local estimates for fully decentralized feedforward massive MIMO equalization: The multistep fusion | |
Turan et al. | Reproducible evaluation of neural network based channel estimators and predictors using a generic dataset | |
Pi et al. | On blind timing acquisition and channel estimation for wideband multiuser DS-CDMA systems |
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 |