CN104022978A - Half-blindness channel estimating method and system - Google Patents

Abstract

The invention provides a half-blindness channel estimating method and system. The method comprises the steps that a pilot frequency sequence is extracted from a receiving signal, and initial estimation is carried out on a channel according to the extracted pilot frequency sequence and a known pilot frequency sequence in the receiving signal to obtain an initial channel estimation value; the initial channel estimation value is adjusted according to RBF neural network to obtain a needed channel estimation result. The method and system overcome the nonlinearity distortion caused when an existing half-blindness channel estimating method is applied to a nonlinearity system.

Description

A kind of half-blind channel estimating method and system

Technical field

The present invention relates to channel estimating field, relate in particular to a kind of half-blind channel estimating method and system.

Background technology

Multi-input multi-output-orthogonal frequency division multiplexing (MIMO-OFDM) can adapt to the demand of the broadband multimedia application such as high channel capacity, higher bit information rate, is considered to one of most potential technology in rear 3G technology.Channel estimating is the necessary condition that is concerned with in MIMO-OFDM system and decodes while detecting with sky, and the importance of channel estimating is self-evident as can be seen here.

In MIMO-OFDM system, receiving terminal carries out can carrying out channel estimating after Time and Frequency Synchronization to the signal receiving, and estimates the frequency response on each subcarrier between every pair of sending and receiving antenna.The position of channel estimating in MIMO-OFDM system as shown in Figure 1.

Existing channel estimating can be divided into the semi-blind channel estimation that blind Channel Estimation, non-blind Channel Estimation and the advantage in conjunction with blind Channel Estimation and non-blind Channel Estimation produce.The feature of blind Channel Estimation is not need pilot tone or reference signal, only utilize modulation signal itself intrinsic, estimate channel characteristics parameter with the irrelevant feature of user profile bit, the method has improved the availability of frequency spectrum, receiver and the requirement of transmitter concertedness are reduced, there is very large prospect improving on the reliability of communication system and capacity, its shortcoming is that operation time is long, and convergence rate is slow, has hindered its application in real system.The feature of non-blind Channel Estimation is in transmitting terminal, to insert pilot tone or the training sequence that receiving terminal is known, not transmitting subscriber information of described pilot tone or training sequence, know whole channel response at receiving terminal by inference by known frequency pilot sign or training sequence, better performances, realize simple, but owing to needing circulation transmission training sequence or pilot frequency sequence to upgrade channel estimation value in time varying channel, so busy channel capacity and effective bandwidth, reduce the efficiency of transmission of channel utilization and system, caused the waste of radio frequency resources.Semi-blind channel estimation is based on blind Channel Estimation, part pilot frequency sequence or training sequence are added at receiving terminal simultaneously, compare the complexity that blind estimation has reduced channel estimating, compare non-blind estimation and reduced the occupied bandwidth of pilot frequency sequence or training sequence, improved the availability of frequency spectrum.But existing semi-blind channel estimation adopts linear computational methods conventionally, while being applied to this non linear system of MIMO-OFDM, tend to occur nonlinear distortion, affect the accuracy of channel estimating.

Summary of the invention

The invention provides a kind of half-blind channel estimating method and system, be applied to the technical problem of the nonlinear distortion that non linear system brings to overcome existing half-blind channel estimating method.

For solving the problems of the technologies described above, the invention provides a kind of half-blind channel estimating method, described method comprises:

Extract pilot frequency sequence from receiving signal, according to this extraction pilot frequency sequence and known to receiving the pilot frequency sequence inserting in signal and carry out the initial estimation of channel, obtain channel initial estimate;

Utilize radial basis (RBF) neural net to adjust channel initial estimate, obtain required channel estimation results.

Further, described radial basis (RBF) neural net of utilizing is adjusted channel initial estimate, obtains required channel estimation results, comprising:

According to the pilot frequency sequence extracting, RBF neural net is arranged, comprise: the hidden layer Centroid that the pilot frequency sequence of extraction is arranged on to RBF neural net, the radial basis width that the pilot frequency sequence of extraction is set to the each node of hidden layer, in the pilot frequency sequence of described extraction, the interval of each frequency pilot sign is identical;

The RBF neural net channel that utilization sets is adjusted channel initial estimate;

Receive the channel estimation results y adjusting through RBF neural net _m:

Y _m=w ₁h ₁+ w ₂h ₂+ ... + w _mh _m, wherein, m is the node number that neural net hidden layer comprises;

x is the initial estimate of channel, x=[x ₁, x ₂..., x _n] ^t, n is subcarrier number, b _jfor the radial basis width of node j in neural net hidden layer; c _jfor the center vector of node j in neural net hidden layer, C _j=[c _j1, c _j2c _jic _jn] ^t, i=1,2 ..., n; W=[w ₁, w ₂..., w _m] be the weight vector of RBF neural net, w _jfor the weight of node j, j=1 ..., m.

Further, described method comprises:

Repeatedly extract pilot frequency sequence from receiving signal, according to the pilot frequency sequence of each extraction and known to receiving the pilot frequency sequence inserting in signal and carry out the initial estimation of channel, obtain k secondary channel initial estimate, K >=2;

According to the pilot frequency sequence of the k time extraction, RBF neural net is arranged;

The RBF neural net that utilization sets is adjusted k secondary channel initial estimate;

Receive the k secondary channel estimated result y adjusting through RBF neural net _m(k)=w ₁h ₁+ w ₂h ₂+ ... + w _mh _m.

Further, described method also comprises, according to k secondary channel estimated result y _m(k) the weight w to RBF neural net node j _jadjust the w after adjustment _j(k) be:

w _j(k)＝w _j(k-1)+Δw _j(k)+α×(w _j(k-1)-w _j(k-2))，

Δw _j(k)＝η×(y(k)-y _m(k))×h _j

Wherein, k>=3, w _j(1)=0, w _j(2) be default random number, y (k) for the weight as RBF neural net node j be w _j(k) in situation, the k secondary channel estimated result of adjusting through RBF neural net; η is learning rate, and α is factor of momentum, and η and α are the intrinsic parameter of RBF neural net.

For solving the problems of the technologies described above, the present invention also provides a kind of semi-blind channel estimation system, and described system comprises: channel initial estimate computing module and channel initial estimate adjusting module, wherein,

Described channel initial estimate computing module, for extracting pilot frequency sequence from receiving signal, according to this extraction pilot frequency sequence and known to receiving the pilot frequency sequence inserting in signal and carry out the initial estimation of channel, obtain channel initial estimate, and this initial estimate is sent to channel initial estimate adjusting module;

Described channel initial estimate adjusting module, for utilizing radial basis (RBF) neural net to adjust the initial estimate of channel, obtains required channel estimation results.

Further, described channel initial estimate adjusting module, for utilizing radial basis (RBF) neural net to adjust the initial estimate of channel, obtains required channel estimation results, comprising:

According to the pilot frequency sequence extracting, RBF neural net is arranged, comprise: the hidden layer Centroid that the pilot frequency sequence of extraction is arranged on to RBF neural net, the radial basis width that the pilot frequency sequence of extraction is set to the each node of hidden layer, in the pilot frequency sequence of described extraction, the interval of each frequency pilot sign is identical;

The RBF neural net channel that utilization sets is adjusted channel initial estimate;

Receive the channel estimation results y adjusting through RBF neural net _m:

Y _m=w ₁h ₁+ w ₂h ₂+ ... + w _mh _m, wherein, m is the node number that neural net hidden layer comprises;

x is the initial estimate of channel, x=[x ₁, x ₂..., x _n] ^t, n is subcarrier number, b _jfor the radial basis width of node j in neural net hidden layer; c _jfor the center vector of node j in neural net hidden layer, C _j=[c _j1, c _j2c _jic _jn] ^t, i=1,2 ..., n; W=[w ₁, w ₂..., w _m] be the weight vector of RBF neural net, w _jfor the weight of node j, j=1 ..., m.

Further,

Described channel initial estimate computing module, for repeatedly extracting pilot frequency sequence from receiving signal, according to the pilot frequency sequence of each extraction and known to receiving the pilot frequency sequence inserting in signal and carry out the initial estimation of channel, obtain k secondary channel initial estimate, k >=2;

Described channel initial estimate adjusting module, arranges RBF neural net for the pilot frequency sequence extracting according to the k time; The RBF neural net that utilization sets is adjusted k secondary channel initial estimate; Receive the k secondary channel estimated result Y adjusting through RBF neural net _m(k)=w ₁h ₁+ w ₂h ₂+ ... + w _mh _m.

Further, described system also comprises neural net weight adjusting module,

Described neural net weight adjusting module, for according to k secondary channel estimated result y _m(k) the weight w to RBF neural net node j _jadjust the w after adjustment _j(k) be:

w _j(k)＝w _j(k-1)+Δw _j(k)+α×(w _j(k-1)-w _j(k-2))，

Δw _j(k)＝η×(y(k)-y _m(k))×h _j

Wherein, k>=3, w _j(1)=0, w _j(2) be default random number, y (k) for the weight as RBF neural net node j be w _j(k) in situation, the k secondary channel estimated result of adjusting through RBF neural net; η is learning rate, and α is factor of momentum, and η and α are the intrinsic parameter of RBF neural net.

Technique scheme transfers to RBF neural net to adjust the channel initial estimate that utilizes semi-blind channel estimation to obtain, utilize RBF network to there is the characteristic of the Nonlinear Mapping of approaching, make half-blindness estimated result there is nonlinear characteristic, can be applied to preferably non linear system.

Brief description of the drawings

Fig. 1 is the position view of channel estimating in MIMO-OFDM system;

Fig. 2 is the half-blind channel estimating method flow chart of the present embodiment;

Fig. 3 is the schematic network structure of RBF neural net;

Fig. 4 is the semi-blind channel estimation system composition diagram of the present embodiment.

Embodiment

For making the object, technical solutions and advantages of the present invention clearer, hereinafter in connection with accompanying drawing, embodiments of the invention are elaborated.It should be noted that, in the situation that not conflicting, the combination in any mutually of the feature in embodiment and embodiment in the application.

Fig. 2 is the half-blind channel estimating method flow chart of the present embodiment.

S201 extracts pilot frequency sequence from receive signal;

S202 carries out the initial estimation of channel according to the pilot frequency sequence extracting and the known pilot frequency sequence inserting in reception signal, obtain channel initial estimate;

The same prior art of this step, can adopt least-squares estimation algorithm to realize;

Channel initial estimate is sent to radial basis (RBF) neural net by S203;

RBF network is a kind of three layers of (input layer, hidden layer and output layer) feedforward network, its hidden layer is linear to the mapping of output layer, but its input layer is nonlinear to the mapping of output layer, therefore RBF network has the characteristic of the Nonlinear Mapping of approaching, and the structural representation of RBF neural net as shown in Figure 3;

S204 utilizes RBF neural net to adjust the initial estimate of channel, obtains required channel estimation results;

According to the pilot frequency sequence extracting, RBF neural net is arranged, comprise: the hidden layer Centroid that the pilot frequency sequence of extraction is arranged on to RBF neural net, the radial basis width that the pilot frequency sequence of extraction is set to the each node of hidden layer, in the pilot frequency sequence of described extraction, the interval of each frequency pilot sign is identical;

The RBF neural net channel that utilization sets is adjusted channel initial estimate;

Receive the channel estimation results y adjusting through RBF neural net _m:

Y _m=w ₁h ₁+ w ₂h ₂+ ... + w _mh _m, wherein, m is the node number that neural net hidden layer comprises, and is also the frequency pilot sign number comprising in pilot frequency sequence;

$h_j=\exp (-\frac{{\left|\right|x-C_j\left|\right|}^2}{2b_j^2}),j=\mathrm{1,2},...,m$

X is the initial estimate of channel, x=[x ₁, x ₂..., x _n] T, n is subcarrier number, b _jfor the radial basis width of node j in neural net hidden layer; c _jfor the center vector of node j in neural net hidden layer, C _j=[c _j1, c _j2..., c _ji] ^t, i=1,2 ..., n, is utilizing RBF neural net to carry out initial treatment stage, C to channel initial estimate _jfor the random one group of parameter arranging, thereby approach gradually and obtain C by learning training in to the processing procedure of channel initial estimate in RBF neural net _j; W=[w ₁, w ₂..., w _m] be the weight vector of RBF neural net, w _jfor the weight of node j, j=1 ..., m.

In above-described embodiment, for the actual change of the channel response reaction channel that ensures to estimate, can from receive signal, repeatedly extract pilot frequency sequence, from receive signal, extract after pilot frequency sequence at every turn, all carry out the initial estimation of channel according to this extraction pilot frequency sequence and known pilot sequence, obtain k secondary channel initial estimate, k>=2; According to the interval of frequency pilot sign in the pilot frequency sequence of the k time extraction, the hidden layer radial basis width of RBF neural net is arranged, and utilize the RBF neural net setting to adjust k secondary channel initial estimate, receive the k secondary channel estimated result y adjusting through RBF neural net _m(k)=w ₁h ₁+ w ₂h ₂+ ... + w _mh _m.

Above-described embodiment in semi-blind channel estimation, makes half-blindness estimated result have nonlinear characteristic RBF Application of Neural Network, can be applied to preferably non linear system.

In the weight vector of RBF neural net, the selection of parameter is extremely important, the weight that neural net hidden layer node is selected is if improper, can cause dispersing of the RBF neural net approximation accuracy even whole RBF neural net of decline, for ensureing the accuracy of weight vector of neural net, the present embodiment also provides the method that neural net weight vector is adjusted:

The first, adjusts neural net weight vector according to gradient descent method, comprising:

According to the above-mentioned k secondary channel estimated result of mentioning, to the weight w of RBF neural net node j _jadjust, k>=3, wherein, w _j(1)=0, w _j(2) be random number, can arrange; y _m(k) the k secondary channel estimated result for adjusting through RBF neural net; Y (k) for the weight as the node j of RBF neural net be w _j(k) in situation, the k secondary channel estimated result of adjusting through RBF neural net; η is learning rate, and α is factor of momentum, and η and α are the intrinsic parameter of RBF neural net;

w _j(k)＝w _j(k-1)+Δw _j(k)+α×(w _j(k-1)-w _j(k-2))

Δw _j(k)＝η×(y(k)-y _m(k))×h _j

The second, adjusts the weight vector of RBF neural net according to genetic algorithm, comprising:

1) initialization population P, comprises crossover scale, crossover probability Pc, mutation probability Pm;

2) be RBF neural net each weight selection individuality the evaluation function that calculates each individuality, described individuality is the possible adaptation value of the each weight of RBF neural net;

Can select individuality, p according to following new probability formula _slarger, more forward in individuality queuing troop to be selected, selected chance is larger;

$p_s=f_i/\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{f}_i$

Wherein, N is chromosomal number in genetic algorithm, and N can arrange; F (i) is i individuality to be selected, $f\left(i\right)=1/E\left(i\right),E\left(i\right)=\underset{k=1}{\overset{N}{\mathrm{\Σ}}}{(y_i\left(k\right)-y_{\mathrm{mi}}\left(k\right))}^2;$

3) in the individuality of choosing, with probability P c to individual G _iand G _i+1interlace operation produces new individual G _i' and G _i+1', the individuality that does not carry out interlace operation directly copies, i ∈ N; In the individuality after intersecting and copying, utilize probability P m sudden change to produce individual G again _jnew individual G _j', j ∈ N, is inserted into new individuality in population P; Calculate new individual evaluation function;

4) whether what judgement was new individual is less than preset value, if be less than defaultly, illustrates that this new individuality is satisfied individuality, execution step 5), otherwise, execution step 3);

5) flow process finishes.

Fig. 4 is the semi-blind channel estimation system composition diagram of the present embodiment.

This system comprises channel initial estimate computing module and channel initial estimate adjusting module, wherein,

Described channel initial estimate computing module, for extracting pilot frequency sequence from receiving signal, according to this extraction pilot frequency sequence and known to receiving the pilot frequency sequence inserting in signal and carry out the initial estimation of channel, obtain channel initial estimate, and this initial estimate is sent to channel initial estimate adjusting module;

Described channel initial estimate adjusting module, be used for utilizing radial basis (RBF) neural net to adjust the initial estimate of channel, obtain required channel estimation results, comprise: according to the pilot frequency sequence extracting, RBF neural net is arranged, comprise: the hidden layer Centroid that the pilot frequency sequence of extraction is arranged on to RBF neural net, the radial basis width that the pilot frequency sequence of extraction is set to the each node of hidden layer, in the pilot frequency sequence of described extraction, the interval of each frequency pilot sign is identical; The RBF neural net channel that utilization sets is adjusted channel initial estimate; Receive the channel estimation results y adjusting through RBF neural net _m: y _m=w ₁h ₁+ w ₂h ₂+ ... + w _mh _m, wherein, m is the node number that neural net hidden layer comprises, and is also the frequency pilot sign number comprising in pilot frequency sequence;

x is the initial estimate of channel, x=[x ₁, x ₂..., x _n] ^t, n is subcarrier number, b _jfor the radial basis width of node j in neural net hidden layer; c _jfor the center vector of node j in neural net hidden layer, C _j=[c _j1, c _j2c _jic _jn] ^t, i=1,2 ..., n, is utilizing RBF neural net to carry out initial treatment stage, C to channel initial estimate _jfor the random one group of parameter arranging, thereby approach gradually and obtain C by learning training in to the processing procedure of channel initial estimate in RBF neural net _j; W=[w ₁, w ₂..., w _m] be the weight vector of RBF neural net, w _jfor the weight of node j, j=1 ..., m.

RBF network is a kind of three layers of (input layer, hidden layer and output layer) feedforward network, its hidden layer is linear to the mapping of output layer, but its input layer is nonlinear to the mapping of output layer, therefore RBF network has the characteristic of the Nonlinear Mapping of approaching.

In above-described embodiment, for the actual change of the channel response reaction channel that ensures to estimate, described channel initial estimate computing module, can be used for repeatedly extracting pilot frequency sequence from receive signal, according to the pilot frequency sequence of each extraction and known to receiving the pilot frequency sequence inserting in signal and carry out the initial estimation of channel, obtain k secondary channel initial estimate, k >=2;

Described channel initial estimate adjusting module, arranges RBF neural net for the pilot frequency sequence extracting according to the k time; The RBF neural net that utilization sets is adjusted k secondary channel initial estimate; Receive the k secondary channel estimated result y adjusting through RBF neural net _m(k)=w ₁h ₁+ w ₂h ₂+ ... + w _mh _m.

Above-described embodiment in semi-blind channel estimation, makes half-blindness estimated result have nonlinear characteristic RBF Application of Neural Network, can be applied to preferably non linear system.

In the weight vector of RBF neural net, the selection of parameter is extremely important, the weight that neural net hidden layer node is selected is if improper, can cause dispersing of the RBF neural net approximation accuracy even whole RBF neural net of decline, for ensureing the accuracy of weight vector of neural net, above-mentioned semi-blind channel estimation system also can comprise neural net weight adjusting module, for neural net weight vector being adjusted according to gradient descent method, specifically comprise: according to k secondary channel estimated result y _m(k) the weight w to RBF neural net node j _jadjust the w after adjustment _j(k) be:

w _j(k)＝w _j(k-1)+Δw _j(k)+α×(w _j(k-1)-w _j(k-2))，

Δw _j(k)＝η×(y(k)-y _m(k))×h _j

Wherein, k>=3, w _j(1)=0, w _j(2) be default random number, y (k) for the weight as RBF neural net node j be w _j(K) in situation, the k secondary channel estimated result of adjusting through RBF neural net; η is learning rate, and α is factor of momentum, and η and α are the intrinsic parameter of RBF neural net.

Above-mentioned neural net weight adjusting module, except according to gradient descent method, neural net weight vector being adjusted, also can adjust the weight vector of RBF neural net according to genetic algorithm, comprise: initialization population P, comprises crossover scale, crossover probability Pc, mutation probability Pm; Select individual and calculate the evaluation function of each individuality for the each weight of RBF neural net, described individuality is the possible adaptation value of the each weight of RBF neural net; In the individuality of choosing, with probability P c to individual G _iand G _i+1interlace operation produces new individual G _i' and G _i+1', the individuality that does not carry out interlace operation directly copies, i ∈ N; In the individuality after intersecting and copying, utilize probability P m sudden change to produce individual G again _jnew individual G _j', j ∈ N, is inserted into new individuality in population P; Calculate new individual evaluation function; Whether what judgement was new individual is less than preset value, if be less than defaultly, illustrates that this new individuality is satisfied individuality; Otherwise, continue to obtain new individuality in the individuality of choosing, and calculate new individual evaluation function.

One of ordinary skill in the art will appreciate that all or part of step in said method can carry out instruction related hardware by program and complete, described program can be stored in computer-readable recording medium, as read-only memory, disk or CD etc.Alternatively, all or part of step of above-described embodiment also can realize with one or more integrated circuits, and correspondingly, the each module/unit in above-described embodiment can adopt the form of hardware to realize, and also can adopt the form of software function module to realize.The present invention is not restricted to the combination of the hardware and software of any particular form.

It should be noted that; the present invention also can have other various embodiments; in the situation that not deviating from spirit of the present invention and essence thereof; those of ordinary skill in the art can make according to the present invention various corresponding changes and distortion, but these corresponding changes and distortion all should belong to the protection range of the appended claim of the present invention.

Claims (8)

1. a half-blind channel estimating method, is characterized in that, described method comprises:

Extract pilot frequency sequence from receiving signal, according to this extraction pilot frequency sequence and known to receiving the pilot frequency sequence inserting in signal and carry out the initial estimation of channel, obtain channel initial estimate;

Utilize radial basis (RBF) neural net to adjust channel initial estimate, obtain required channel estimation results.

2. the method for claim 1, is characterized in that, described radial basis (RBF) neural net of utilizing is adjusted channel initial estimate, obtains required channel estimation results, comprising:

According to the pilot frequency sequence extracting, RBF neural net is arranged, comprise: the hidden layer Centroid that the pilot frequency sequence of extraction is arranged on to RBF neural net, the radial basis width that the pilot frequency sequence of extraction is set to the each node of hidden layer, in the pilot frequency sequence of described extraction, the interval of each frequency pilot sign is identical;

The RBF neural net channel that utilization sets is adjusted channel initial estimate;

Receive the channel estimation results y adjusting through RBF neural net _m:

Y _m=w ₁h ₁+ w ₂h ₂+ ... + w _mh _m, wherein, m is the node number that neural net hidden layer comprises;

x is the initial estimate of channel, x=[x ₁, x ₂..., x _n] ^t, n is subcarrier number, b _jfor the radial basis width of node j in neural net hidden layer; c _jfor the center vector of node j in neural net hidden layer, C _j=[c _j1, c _j2c _jic _jn] ^t, i=1,2 ..., n; W=[w ₁, w ₂..., w _m] be the weight vector of RBF neural net, w _jfor the weight of node j, j=1 ..., m.

3. method as claimed in claim 2, is characterized in that, described method comprises:

Repeatedly extract pilot frequency sequence from receiving signal, according to the pilot frequency sequence of each extraction and known to receiving the pilot frequency sequence inserting in signal and carry out the initial estimation of channel, obtain k secondary channel initial estimate, K >=2;

According to the pilot frequency sequence of the k time extraction, RBF neural net is arranged;

The RBF neural net that utilization sets is adjusted k secondary channel initial estimate;

Receive the k secondary channel estimated result y adjusting through RBF neural net _m(k)=w ₁h ₁+ w ₂h ₂+ ... + w _mh _m.

4. method as claimed in claim 3, is characterized in that, described method also comprises, according to k secondary channel estimated result y _m(k) the weight w to RBF neural net node j _jadjust the w after adjustment _j(k) be:

w _j(k)＝w _j(k-1)+Δw _j(k)+α×(w _j(k-1)-w _j(k-2))，

Δw _j(k)＝η×(y(k)-y _m(k))×h _j

Wherein, k>=3, w _j(1)=0, w _j(2) be default random number, y (k) for the weight as RBF neural net node j be w _j(k) in situation, the k secondary channel estimated result of adjusting through RBF neural net; η is learning rate, and α is factor of momentum, and η and α are the intrinsic parameter of RBF neural net.

5. a semi-blind channel estimation system, is characterized in that, described system comprises channel initial estimate computing module and channel initial estimate adjusting module, wherein,

Described channel initial estimate computing module, for extracting pilot frequency sequence from receiving signal, according to this extraction pilot frequency sequence and known to receiving the pilot frequency sequence inserting in signal and carry out the initial estimation of channel, obtain channel initial estimate, and this initial estimate is sent to channel initial estimate adjusting module;

Described channel initial estimate adjusting module, for utilizing radial basis (RBF) neural net to adjust the initial estimate of channel, obtains required channel estimation results.

6. system as claimed in claim 5, is characterized in that, described channel initial estimate adjusting module, for utilizing radial basis (RBF) neural net to adjust the initial estimate of channel, obtains required channel estimation results, comprising:

According to the pilot frequency sequence extracting, RBF neural net is arranged, comprise: the hidden layer Centroid that the pilot frequency sequence of extraction is arranged on to RBF neural net, the radial basis width that the pilot frequency sequence of extraction is set to the each node of hidden layer, in the pilot frequency sequence of described extraction, the interval of each frequency pilot sign is identical;

The RBF neural net channel that utilization sets is adjusted channel initial estimate;

Receive the channel estimation results y adjusting through RBF neural net _m:

Y _m=w ₁h ₁+ w ₂h ₂+ ... + w _mh _m, wherein, m is the node number that neural net hidden layer comprises;

x is the initial estimate of channel, x=[x ₁, x ₂..., x _n] ^t, n is subcarrier number, b _jfor the radial basis width of node j in neural net hidden layer; c _jfor the center vector of node j in neural net hidden layer, Cj=[c _j1, c _j2c _jic _jn] ^t, i=1,2 ..., n; W=[w ₁, w ₂..., w _m] be the weight vector of RBF neural net, w _jfor the weight of node j, j=1 ..., m.

7. system as claimed in claim 6, is characterized in that,

Described channel initial estimate computing module, for repeatedly extracting pilot frequency sequence from receiving signal, according to the pilot frequency sequence of each extraction and known to receiving the pilot frequency sequence inserting in signal and carry out the initial estimation of channel, obtain k secondary channel initial estimate, k >=2;

Described channel initial estimate adjusting module, arranges RBF neural net for the pilot frequency sequence extracting according to the k time; The RBF neural net that utilization sets is adjusted k secondary channel initial estimate; Receive the k secondary channel estimated result y adjusting through RBF neural net _m(k)=w ₁h ₁+ w ₂h ₂+ ... + w _mh _m.

8. system as claimed in claim 7, is characterized in that, described system also comprises neural net weight adjusting module,

Described neural net weight adjusting module, for according to k secondary channel estimated result y _m(k) the weight w to RBF neural net node j _jadjust the w after adjustment _j(k) be:

w _j(k)＝w _j(k-1)+Δw _j(k)+α×(w _j(k-1)-w _j(k-2))，

Δw _j(k)＝η×(y(k)-y _m(k))×h _j

Wherein, k>=3, w _j(1)=0, w _j(2) be default random number, y (k) for the weight as RBF neural net node j be w _j(k) in situation, the k secondary channel estimated result of adjusting through RBF neural net; η is learning rate, and α is factor of momentum, and η and α are the intrinsic parameter of RBF neural net.