Disclosure of Invention
The invention aims to provide a method and a system for determining a source signal, which solve the problem of low source signal determination accuracy caused by low efficiency of extracting a convolution blind signal in the prior art.
In order to achieve the purpose, the invention provides the following scheme:
a method of determining a source signal, comprising:
acquiring an observation signal of a convolution blind signal;
sampling observation signals of the convolution blind signals to determine an initial particle set;
according to the initial particle set, obtaining an updated particle set by adopting a pilotless Kalman particle filter algorithm;
determining a particle risk function according to the updated particle set;
determining a weight mapping function of the particle risk according to the particle risk function;
determining a particle cost function according to the updated particle set and the weight mapping function;
according to the weight mapping function and the particle cost function, fitting by adopting a least square method to obtain a predicted value of an extracted vector;
determining the determined value of the extracted vector by adopting a weighted least square method according to the predicted value of the extracted vector;
acquiring a signal mapping relation; the signal mapping relationship is a mapping relationship among the observed signal, the determined value of the extraction vector and a source signal;
and determining the source signal by utilizing the mapping relation according to the determined value of the extracted vector and the observation signal of the convolution blind signal.
Optionally, the determining a particle cost function according to the updated particle set and the weight mapping function specifically includes:
resampling the updated particle set according to the weight mapping function, and determining a cost sample estimation set of particles;
determining the particle cost function from the set of cost sample estimates for the particle.
Optionally, the obtaining the signal mapping relationship further includes:
acquiring an observation signal of a known convolution signal;
determining a convolution mixing model according to an observation signal of a known convolution signal;
converting the convolutional hybrid model to a transient hybrid model;
and determining the signal mapping relation according to the instantaneous mixing model.
Optionally, the determining the value of the extracted vector by using a weighted least square method according to the predicted value of the extracted vector specifically includes:
determining a vector variance set according to the predicted value of the extracted vector;
and obtaining the determined value of the extracted vector by adopting a weighted least square method according to the predicted value of the extracted vector and the vector variance set.
A system for determining a source signal, comprising:
the first observation signal acquisition module is used for acquiring observation signals of the convolution blind signals;
an initial particle set determining module, configured to sample an observation signal of the convolution blind signal, and determine an initial particle set;
the particle set updating module is used for obtaining an updated particle set by adopting a pilotless Kalman particle filter algorithm according to the initial particle set;
a particle risk function determining module, configured to determine a particle risk function according to the updated particle set;
the weight mapping function determining module is used for determining a weight mapping function of the particle risk according to the particle risk function;
a particle cost function determining module, configured to determine a particle cost function according to the updated particle set and the weight mapping function;
the predicted value determining module is used for obtaining a predicted value of the extracted vector by adopting least square fitting according to the weight mapping function and the particle cost function;
the determination module of the extracted vector is used for determining the determination value of the extracted vector by adopting a weighted least square method according to the predicted value of the extracted vector;
the signal mapping relation acquisition module is used for acquiring a signal mapping relation; the signal mapping relationship is a mapping relationship among the observed signal, the determined value of the extraction vector and a source signal;
and the source signal determining module is used for determining the source signal by utilizing the mapping relation according to the determined value of the extraction vector and the observation signal of the convolution blind signal.
Optionally, the particle cost function determining module specifically includes:
a cost sample estimation set determining unit of the particle, configured to resample the updated particle set according to the weight mapping function, and determine a cost sample estimation set of the particle;
a particle cost function determination unit, configured to determine the particle cost function according to the cost sample estimation set of the particle.
Optionally, the method further includes:
the second observation signal acquisition module is used for acquiring an observation signal of a known convolution signal;
the convolution mixing model determining module is used for determining a convolution mixing model according to an observation signal of a known convolution signal;
an instantaneous mixture model determination module for converting the convolutional mixture model into an instantaneous mixture model;
and the signal mapping relation determining module is used for determining the signal mapping relation according to the instantaneous mixing model.
Optionally, the determining module for determining the determined value of the extracted vector specifically includes:
a vector variance set determination unit for determining a vector variance set according to the predicted value of the extracted vector;
and the determination unit of the extraction vector is used for determining the determination value of the extraction vector by adopting a weighted least square method according to the prediction value of the extraction vector and the vector variance set.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
according to the method and the system for determining the source signal, provided by the invention, under the condition that the prior verification condition is unknown, the particle set obtained by sampling is updated by adopting a non-pilot Kalman particle filter algorithm, a particle risk function, a weight mapping function and a particle cost function are determined according to the updated particle set, the determination value of the extraction vector is further determined by adopting a least square method, and the extraction of the convolution blind signal is realized by the determination value of the extraction vector. By determining the determined value of the extracted vector, mathematical model modeling is not needed to be carried out on the introduced quantization noise, the defect that the noise signal is difficult to be modeled accurately is avoided, and the accuracy of determining the source signal is improved.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a method and a system for determining a source signal, which solve the problem of low source signal determination accuracy caused by low efficiency of extracting a convolution blind signal in the prior art.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Fig. 1 is a schematic flow chart of a method for determining a source signal according to the present invention, and as shown in fig. 1, the method for determining a source signal according to the present invention includes:
s101, acquiring an observation signal of the convolution blind signal.
And S102, sampling the observation signals of the convolution blind signals, and determining an initial particle set.
For the convolutionThe observation signal of the blind signal is sampled in a bounded region covering the initial real value of the particles, and the particles are sampled to obtain N
pAn initial particle
The cost value of each initial particle is
Initial particle cost sample set of
Wherein
Is the variance matrix of the initial particles.
And S103, obtaining an updated particle set by adopting a pilotless Kalman particle filter algorithm according to the initial particle set.
And (3) adopting a pilotless transformation to iteratively update all the particles at the time k:
wherein n is
wIs the dimension of the vector to be integrated,
is an estimated value after no pilot transformation. In order to obtain updated particles with higher accuracy, kalman particle filter updating is performed on the particles at the time k, and the updating process is as follows:
and S104, determining a particle risk function according to the updated particle set.
The process noise variance during particle update can be assumed to define a particle risk function at time k, following a gaussian distribution, as:
is the particle cost value at time k,
and the lambda epsilon (0,1) is a particle weight forgetting factor and is used for adjusting the influence of the particle cost value at the previous moment on the current particle risk value, and when the lambda is 0, no influence is shown.
And S105, determining a weight mapping function of the particle risk according to the particle risk function.
Using formulas
The particle risk value is mapped to a resampling weight for the particle, i.e. a weight mapping function that determines the particle risk. Number of particles N
pthe values of delta epsilon (0,1) and β > 1 are respectively mapping adjustment coefficients
In which is removed
The parameter delta is used for ensuring the diversity of the particles after resampling, and the parameter delta is used for ensuring that extreme conditions do not occur due to particle risk prediction values.
And S106, determining a particle cost function according to the updated particle set and the weight mapping function.
Resampling the updated particle set according to the weight mapping function, and determining a cost sample estimation set of particles;
resampling the updated particle set by using the weight mapping function of particle risk, and obtaining a cost sample estimation set of particles as
Determining the particle cost function from the set of cost sample estimates for the particle.
Particle set of k +1 time separation vectors
Is updated, wherein
Represents a mean value of
And variance of
A gaussian distribution of (a).
Using formulas
Determining the particle cost function.
Is that
An estimate of (d).
And S107, fitting by adopting a least square method according to the weight mapping function and the particle cost function to obtain a predicted value of the extracted vector.
And mapping the particle cost function into a posterior probability by using the weight mapping function.
Using formulas
And performing least square fitting on the posterior probability to obtain a predicted value of the extracted vector.
And extracting a predicted value of the vector for the k +1 moment.
And S108, determining the determined value of the extracted vector by adopting a weighted least square method according to the predicted value of the extracted vector.
And determining a vector variance set according to the predicted value of the extracted vector.
Let the extraction vector at time k be { w
1,w
2,…,w
kThe vector is extracted with the increment of k +1
Determining a set of vector variances of
And obtaining the determined value of the extracted vector by adopting a weighted least square method according to the predicted value of the extracted vector and the vector variance set.
Using formulas
Determining a determined value of the extraction vector.
S109, acquiring a signal mapping relation; the signal mapping relationship is a mapping relationship between the observed signal, the determined value of the extraction vector, and a source signal.
The signal mapping relation is
The source signal is a signal that is,
extracting the determined value of the vector, Z
k+1To observe the signal.
And S110, determining the source signal by utilizing the mapping relation according to the determined value of the extracted vector and the observation signal of the convolution blind signal.
Before the signal mapping relation is obtained, the signal mapping relation is determined by using the known convolution blind signals of the database. The determination process is as follows:
an observed signal of the known convolved signal is obtained.
A convolutional hybrid model is determined from the observed signal of the known convolutional signal.
Converting the convolutional hybrid model to a transient hybrid model.
And determining the signal mapping relation according to the instantaneous mixing model.
As shown in FIG. 2, N source signals are observed by M nodes (sensor array, signal receiving device) (M ≧ N). The observation equation (convolution mixture model) of the k-th observation node is as follows:
where K is a sequence of discrete time instants, sk=[s1,k,s2,k,...,sN,k]TFor N observed source signals, vk=[v1,k,v2,k,...,vM,k]TFor M noise vectors, xk=[x1,k,x2,k,...,xM,k]TM observation column vector signals at the k time; h (l) ═ Hij(l)]M×N,l=1,2,...,L,hij(l) Is the impulse response of L +1 order from the jth source signal to the ith observation node.
Quantization processing (uniform quantization method or weight method can be selected) is carried out on the obtained observation signal of the kth node, and quantization noise introduced in the quantization process is set as [ q ]1,k,q2,k,...,qM,k]TThe overall noise of the system can be expressed as uk=[v1,k+q1,k,v2,k+q2,k,...,vM,k+qM,k]TThe quantized observed signal is:
wherein z iskIs a quantized observed signal. Assuming that the integer L ' satisfies ML ' being greater than or equal to N (L ' + L), the source signal, the observation signal and the noise are converted into:
Si,k=[si,k,si,k-1,…,si,k-L-L'+1]T,i=1,2,…,N,Zi,k=[zi,k,zi,k-1,…,zi,k-L'+1]Ti-1, 2, …, M and Ui,k=[ui,k,ui,k-1,…,ui,k-L'+1]T,i=1,2,…,M。
Further, the converted source signal, observation signal and noise are transformed into a vector mode:
determining an ML '× N (L' + L) dimensional coefficient matrix A:
wherein A isijIs an L '× (L + L') dimensional matrix:
by the defined coefficient matrix A and matrix AijConverting the convolution mixture model to a transient mixture model:
Zk=ASk+Uk。
determining a value w of a time invariant extraction vector for determining an ML' dimension
TFurther using the formula
Determining a source signal
Obtaining a mapping equation of s for the source signalj,k=fj(sj,k-1)。
Determine an observation equation for the node as
The extraction vector or the separation matrix of the instantaneous hybrid model has the characteristics of singleness and time invariance, and the state observation equation of the nonlinear filtering algorithm is further determined as follows:
corresponding to the method for determining a source signal provided by the present invention, the present invention further provides a system for determining a source signal, as shown in fig. 3, the system for determining a source signal provided by the present invention includes: a first observed signal obtaining module 301, an initial particle set determining module 302, a particle set updating module 303, a particle risk function determining module 304, a weight mapping function determining module 305, a particle cost function determining module 306, a predicted value of extracted vector determining module 307, a determined value of extracted vector determining module 308, a signal mapping relation obtaining module 309, and a source signal determining module 310.
The first observation signal obtaining module 301 is configured to obtain an observation signal of the convolved blind signal.
The initial particle set determining module 302 is configured to sample the observation signals of the convolution blind signals to determine an initial particle set.
The particle set updating module 303 is configured to obtain an updated particle set by using a pilotless kalman particle filter algorithm according to the initial particle set.
The particle risk function determination module 304 is configured to determine a particle risk function according to the updated set of particles.
The weight mapping function determining module 305 is configured to determine a weight mapping function of the particle risk according to the particle risk function.
The particle cost function determining module 306 is configured to determine a particle cost function according to the updated set of particles and the weight mapping function.
And the predicted value determining module 307 for extracting the vector is configured to obtain the predicted value of the extracted vector by adopting least square fitting according to the weight mapping function and the particle cost function.
The determination module 308 for determining the determination value of the extracted vector is configured to determine the determination value of the extracted vector by using a weighted least square method according to the prediction value of the extracted vector.
The signal mapping relation obtaining module 309 is configured to obtain a signal mapping relation; the signal mapping relationship is a mapping relationship between the observed signal, the determined value of the extraction vector, and a source signal.
The source signal determining module 310 is configured to determine the source signal according to the determined value of the extracted vector and the observed signal of the convolutional blind signal by using the mapping relationship.
The particle cost function determining module 306 specifically includes: a cost sample estimation set determination unit and a particle cost function determination unit of the child.
And the particle cost sample estimation set determining unit is used for resampling the updated particle set according to the weight mapping function and determining a particle cost sample estimation set.
The particle cost function determination unit is used for determining the particle cost function according to the cost sample estimation set of the particles.
The invention provides a source signal determining system, which further comprises: the device comprises a second observation signal acquisition module, a convolution mixed model determination module, an instantaneous mixed model determination module and a signal mapping relation determination module.
The second observation signal acquisition module is used for acquiring an observation signal of the known convolution signal.
The convolution mixture model determining module is used for determining a convolution mixture model according to an observation signal of a known convolution signal.
The instantaneous mixture model determination module is used for converting the convolution mixture model into an instantaneous mixture model.
And the signal mapping relation determining module is used for determining the signal mapping relation according to the instantaneous mixing model.
The determining module 308 for determining the value of the extracted vector specifically includes: a vector variance set determination unit and an extracted vector determination value determination unit.
The vector variance set determining unit is used for determining a vector variance set according to the predicted value of the extracted vector.
And the determination unit of the extraction vector is used for determining the determination value of the extraction vector by adopting a weighted least square method according to the prediction value of the extraction vector and the vector variance set.
The method and the system for determining the source signal provided by the invention are based on the Kalman particle filter theory, and carry out real-time estimation on the parameters of the mixed channel through iterative update of particles so as to reduce the dependence on the statistical characteristics of the source signal and better accord with the actual application scene. And estimating the mixed channel parameters by adopting a Kalman particle filtering method.
According to the method and the system for determining the source signal, provided by the invention, the particle filtering method is improved by utilizing the particle cost and the risk function, the iterative precision of the particles is further improved by a volume point transformation method, and the real-time extraction of the convolution blind signal is finally completed. When the condition that the prior knowledge of the mixed signal is insufficient is processed, the method has a good extraction effect, namely when unknown noise is introduced into a system and the noise cannot be described mathematically accurately, the source signal can be effectively determined by the method and the system for determining the source signal, and the authenticity and the accuracy of the source signal are ensured.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.