Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a MIMO underwater acoustic channel estimation method based on an improved adaptive hybrid optimization smoothing L0 norm, which effectively improves the noise robustness and the channel estimation performance.
The invention adopts the following technical scheme for solving the technical problems:
the invention provides a MIMO underwater acoustic channel estimation method based on an improved adaptive hybrid optimization smoothing L0 norm, which comprises the following steps:
step 1, establishing a network with NtA transmitting terminal and NrMIMO system of each receiving end, forming an Nt×NrEach subchannel corresponds to an underwater acoustic channel model, and the mth transmission time t of the subchannel is obtainedChannel frequency domain response D of the terminal and the nth receiving terminal in time delay of taumn(t, τ) to obtain N for the entire MIMO systemt×NrA channel frequency domain response matrix D (t, tau) of the underwater acoustic channel under the time delay of t time tau;
step 2, vectorizing the obtained channel frequency domain response matrix D (t, tau) into a vector h of Nx 1
h=vec(D(t,τ))
Wherein vec (.) represents a vectorization function, and N is the number of elements of the vector h;
step 3, establishing an underwater acoustic channel compressed sensing equation y ═ Ax + r, wherein x represents an underwater acoustic channel vector to be estimated, y represents a one-dimensional observation vector, the dimension of the one-dimensional observation vector is Mx 1, M represents the number of elements of the observation vector y, A is a known random Gaussian matrix measurement matrix, the dimension of the A is Mx N, r represents independent and identically distributed Gaussian white noise, the mean value of the R is 0, and the variance is w2;
Step 4, performing hybrid optimization by adopting an iterative algorithm combining a Nesterov gradient acceleration method and a Newton method to obtain a final iterated underwater acoustic channel vector estimation value x; the method comprises the following specific steps:
step 4a, initialization:
step A, solving an initial value x of an underwater sound channel by using a least square method(0)Presetting the first cycle number as L and the second cycle number as K;
step B, selecting a sequence [ sigma ]1,σ2,σ3,....,σJ.]And σ1>σ2>σ3>,...,>σJWherein σ isjJ represents the value of the jth control parameter σ, J represents the sequence number of the control parameter σ, and J is 1;
step (1) of setting the control parameter σ of the objective function in the SL0 algorithm to σjCalculating L times by using Nesterov gradient acceleration method, and solving fσ(x) Minimum value x of (1)L (j)Wherein x isL (j)Representing the minimum value after L cycles in the j iteration;
selection of an objective function f in the SL0 algorithmσ(x) As a function of the norm of approximate L0SL0 algorithm, fσ(x) Is expressed as
Wherein e represents a natural base number, and sigma represents a control parameter of the objective function;
step (2), initializing a Newton method: x ═ xL (j);
Computing K times by using Newton method, and solving fσ(x) Minimum value x of (1)K (j)Wherein x isK (j)Representing the minimum value after K cycles in the j iteration;
calculating a regularization factor lambda and projecting it onto the feasible set:
x(j)=xK (j)-AH(AAH+λ-1Im)-1(AxK (j)-y)
wherein, ()HDenotes a conjugate transpose, ImRepresenting an identity matrix of order M x M, Q being a constant, x(j)Is the optimal value of the underwater acoustic channel after the jth iteration;
and (3) outputting x ═ x(j);
Step 4b, when sigmaj>σJWhen j is equal to j +1, σj=βσj-1Turning to step (1), β is the attenuation factor, 0<β<1。
As a further optimization scheme of the MIMO underwater acoustic channel estimation method based on the improved self-adaptive hybrid optimization smoothing L0 norm, sigma1=1。
As a further optimization scheme of the MIMO underwater acoustic channel estimation method based on the improved adaptive hybrid optimization smoothing L0 norm, in step 1, the channel frequency domain response D of a subchannel is obtainedmn(t, τ) is
Wherein p is the number of paths of each sub-channel, m, n respectively represent the mth transmitting end and the nth receiving end, Dmn(t, tau) represents the frequency domain response of the mth transmitting end and the nth receiving end in the time of tau time delay at the time of t, taui,mnRepresenting the delay, h, of the ith path between the mth transmitting end and the nth receiving endi,mn(t) represents the impulse response of the ith path between the mth transmitting end and the nth receiving end at the time t, delta (t-tau)i,mn) Is expressed at taui,mnA unit impulse function of the time delay;
n of the whole MIMO system
t×N
rThe channel frequency domain response matrix D (t, tau) of the underwater acoustic channel at the time t time tau is expressed as
As a further optimization scheme of the MIMO underwater acoustic channel estimation method based on the improved adaptive hybrid optimization smoothing L0 norm, K is 3, and L is 2.
As a further optimization scheme of the MIMO underwater acoustic channel estimation method based on the improved adaptive hybrid optimization smoothing L0 norm, β is 0.5.
Compared with the prior art, the invention adopting the technical scheme has the following technical effects:
(1) on the basis of a regularization SL0(ReSL0) algorithm, a target function with a better L0 norm is approximated, and a reliable regularization factor is generated in a self-adaptive mode to balance errors based on the sparsity and residual error of a solution in the target function in an iteration process; in order to ensure that iteration accurately converges to an optimal point, the initial value after inner loop iteration of a Nesterov gradient acceleration method (NAG) is adopted as the initial value of a Newton method to carry out mixed optimization channel estimation, and the noise robustness and the channel estimation performance are effectively improved;
(2) in the actual underwater acoustic channel, because the sparsity characteristic of the channel cannot be exactly known, the method can better estimate the channel information without acquiring the prior information of the sparsity of the channel on the basis of the compressive sensing theory; the MAReSL0 algorithm performs mixed optimization of the Neisseliverv gradient acceleration method and the Newton method based on a new objective function, adaptively generates a proper regularization factor for estimation of a sparse underwater acoustic channel, and compares estimation performance of the MAReSL0 algorithm with that of SL0, improved SL0, ReSL0, improved ReSL0 and AReSL0 algorithms; the comparison result shows that the method is obviously superior to other 5 algorithms in MSE and SER performance indexes, the channel estimation performance is better, and the robustness to noise is better.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and specific embodiments.
Fig. 6 is a flow chart of MIMO underwater acoustic channel estimation based on an improved adaptive hybrid optimization smoothing L0 norm. The method specifically comprises the following steps:
1): as shown in FIG. 1, establish a channel with NtMIMO system of transmitting end and Nr receiving ends, forming one NtA sub-channel model of x Nr, each sub-channel corresponding to an underwater acoustic channel model, and obtaining the channel frequency domain response D of the sub-channelmn(t, τ) is
Wherein p is the number of paths of each sub-channel, m, n respectively represent the mth transmitting end and the nth receiving end, Dmn(t, tau) represents the frequency domain response of the mth transmitting end and the nth receiving end in the time of tau time delay at the time of t, taui,mnRepresenting the delay, h, of the ith path between the mth transmitting end and the nth receiving endi,mn(t) represents the impulse response of the ith path of the mth transmitting end and the nth receiving end at the time t, Dmn(t, τ) channel frequency domain responses of the mth transmitting end and the nth receiving terminal channels;
the channel frequency domain response matrix D (t, tau) of the Nt multiplied by Nr underwater acoustic channel of the whole MIMO system under the time delay of t time tau is expressed as
2) Vectorizing the obtained channel frequency domain response matrix D (t, tau) into an Nx 1 vector h
h=vec(D(t,τ))
Wherein vec (.) represents a vectorization function, and N is the number of elements of the vector h;
3) establishing an underwater acoustic channel compressed sensing equation y as Ax + r, wherein x represents an underwater acoustic channel vector to be estimated, y represents a one-dimensional observation vector with dimension of M × 1, M represents the number of elements of the observation vector y, A is a known random Gaussian matrix measurement matrix with dimension of M × N, r represents independent and uniformly distributed white Gaussian noise with mean value of 0 and variance of w2;
4) Performing hybrid optimization by adopting an iterative algorithm combining a Nesterov gradient acceleration method and a Newton method to obtain a final iterated underwater acoustic channel vector estimation value x; the method comprises the following specific steps:
4a, initialization:
A. method for solving initial value x of underwater acoustic channel by least square method(0)Presetting the first cycle number as L and the second cycle number as K;
B. selecting a sequence [ sigma ]1,σ2,σ3,....,σJ.]And σ1>σ2>σ3>,...,>σJWherein j is 1, σjRepresents the value of the jth control parameter sigma, and J represents the sequence number of the control parameter sigma;
(1) let the control parameter sigma of the objective function in the SL0 algorithm be sigmajCalculating L times by using Nesterov gradient acceleration method, and solving fσ(x) Minimum value x of (1)L (j)Wherein x isL (j)Representing the minimum value after L cycles in the j iteration;
target function selection in the SL0 algorithm approximates the better performing target function as shown in FIG. 2, target function selection f in the SL0 algorithmσ(x) SL0 algorithm was performed as a function of an approximate L0 norm, fσ(x) Is expressed as
Wherein e represents a natural base number, and sigma represents a control parameter of the objective function;
(2) initializing Newton method: x ═ xL (j);
Computing K times by using Newton method, and solving fσ(x) Minimum value x of (1)K (j)Wherein x isK (j)Representing the minimum value after K cycles in the j iteration;
calculating a regularization factor lambda and projecting it onto the feasible set:
x(j)=xK (j)-AH(AAH+λ-1Im)-1(AxK (j)-y)
wherein, ()HDenotes a conjugate transpose, ImRepresenting an identity matrix of order M x M, Q being a constant, x(j)Is the optimal value of the underwater acoustic channel after the jth iteration;
(3) x is output(j);
4b, when σj>σJWhen j is equal to j +1, σj=βσj-1Turning to (1), β is the attenuation factor, 0<β<1。。
Simulation experiment
1 underwater acoustic channel sparsity simulation
In order to verify the performance of the MAReSL0 algorithm, a bellhop underwater acoustic channel simulation model is adopted to model a sparse underwater acoustic channel, and N is sett2 emitting sources and Nr 2 receiving ends, the depth is 10m and 20m respectively, the distance is 1000m, the water depth is 150m, the uniform sound velocity is 1500m/s, and 4 channels are formed from the emitting sources to the receiving ends.
Fig. 3 shows the simulated channel normalized impulse response absolute values of 4 channels respectively. In FIG. 3, (a), (b), (c), (d) are channels 1, 2, 3, 4, respectively. As can be seen from fig. 3, the underwater acoustic channel has a significant sparsity. The number of the paths of the 4 channels in the simulation is 44, 43, 44 and 43 respectively, namely the sparsity of each channel.
2MAReSL0 algorithm performance simulation
In the simulation experiment, simulation parameters include: the length N of the vectorized channel h is 800, and M is 64; the signal to noise ratio varies from 0dB to 20dB against a background of gaussian white noise in the experiment. The simulation was run 100 times and averaged for comparison. The marcsl 0 algorithm of the present invention has two iterations K-3, L-2, and β -0.5, all of which are empirical values.
The performance indexes of the algorithm adopted by the invention are channel estimation Mean Square Error (MSE) and error rate (SER), which are respectively expressed as
In terms of performance, the smaller the values of MSE and SER, the better the algorithm estimates performance.
The MAReSL0 algorithm carries out channel estimation by using the sparse characteristic of a channel, and is compared with SL0[3], MSL0, ReSL0[6], MReSL0 and AReSL0[7] algorithms in simulation experiments, wherein the MSL0 and MReSL0 algorithms are obtained by replacing original objective functions in SL0 and ReSL0 algorithms with the objective functions. In the simulation experiment, different random gaussian matrices are taken to operate for 100 times and then an average value is taken to calculate, and the mean square error and the error rate curves of the channels of the 6 estimation algorithms are respectively shown in fig. 4 and fig. 5.
As can be seen from fig. 4 and 5, the new objective function of the present invention is adopted, so the channel estimation performance of the improved SL0 and the improved rel 0 algorithms is slightly better than that of the SL0 and the rel 0 algorithms, respectively.
As can be seen from fig. 4 and 5, in a noise environment, both MSE and SER performance of the SL0 algorithm are inferior to those of other algorithms, and the noise immunity performance of the SL0 algorithm is the worst; the ReSL0 algorithm introduces a regularization factor that estimates better performance than the SL0, modified SL0 algorithm. In a noise environment, because the regularization parameter of the ReSL0 algorithm is fixed in the iteration process, the sparse solution of the channel does not change proportionally with the residual error, and the estimation performance of the channel is greatly reduced; the AReSL0 algorithm can adaptively generate regularization factors in iterations that estimate performance superior to the SL0, the modified SL0, the ReSL0, and the modified ReSL0 algorithms.
As can be seen from fig. 4 and 5, the improved adaptive regularization SL0 (marisl 0) algorithm of the present invention adopts an adaptive regularization iterative algorithm that is a mixture of a nertiov gradient acceleration method and a newton method, which can not only obtain an optimal value accurately, but also generate a regularization error factor adaptively, maintain channel sparsity and balance of residual errors in an iterative process, and have better estimation performance, and noise robustness is better than that of SL0, improved SL0, renl 0, improved renl 0, and AReSL0 algorithms.
The above description is only for the specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.