TIME-FREQUENCY BLOCK-SPARSE CHANNEL ESTIMATION METHOD BASED ON COMPRESSED SENSING

Abstract

A time-frequency block-sparse channel estimation method based on compressed sensing includes the following steps. Step 1: A channel model is established. Step 2: According to the channel model obtained in Step 1, a sparse signal estimation value is solved by a compressed sensing method to further calculate an index set. Step 3: According to the index set obtained in Step 2, a channel matrix estimation value is solved. The method provides a generalized block adaptive gBAMP algorithm, which uses time-frequency joint block sparsity of a massive MIMO system to further optimize selection of an index set in an algorithm iteration process to improve stability of the algorithm. Then, without a specified threshold parameter, based on an F norm, an adaptive iteration stop condition is determined based on a residual, and the validity of the method is proved.

Claims

1. A time-frequency block-sparse channel estimation method based on compressed sensing, wherein an orthogonal frequency-division multiplexing (OFDM) system of a downlink frequency-division duplexing (FDD) massive MIMO channel model is initialized, supposing M antennas are disposed at a base station end and U single-antenna users are simultaneously served, and let there be N subcarriers in the OFDM system, wherein N.sub.P subcarriers are used to transmit pilot signals, and L is a maximum path delay, considering observing in R adjacent OFDM symbols, the method comprising: Step 1: inputting a pilot signal and a reception signal of a transmitting end, and establishing a channel model as Y=ΨH+V according to the signal, wherein Y∈.sup.N.sup.P.sup.×R is a reception signal matrix, H∈.sup.LM×R is a channel matrix, Ψ∈.sup.N.sup.P.sup.×LM is a pilot matrix, V∈.sup.N.sup.P.sup.×R is a noise matrix; Step 2: solving a sparse signal estimation value {tilde over (H)} by a compressed sensing method according to the channel model obtained in Step 1 to further calculate an index set {tilde over (Γ)}.sub.k; and Step 3: solving a channel matrix estimation value {tilde over (H)}.sub.{tilde over (Γ)}.sub.k according to the index set {tilde over (Γ)}.sub.k obtained in Step 2, i.e., {tilde over (H)}.sub.{tilde over (Γ)}.sub.k=Ψ.sub.{tilde over (Γ)}.sub.k.sup.†Y, wherein a superscript “†” represents a pseudoinverse, i.e., Ψ.sub.{tilde over (Γ)}.sub.k.sup.† represents a pseudoinverse with respect to Ψ.sub.{tilde over (Γ)}.sub.k, and after a baseband signal is demodulated, outputting data information of the transmitting end according to the obtained channel matrix estimation value {tilde over (H)}.sub.{tilde over (Γ)}.sub.k.

2. The time-frequency block-sparse channel estimation method based on compressed sensing according to claim 1, wherein Step 1 further comprises: after establishing the channel model, since N.sub.P<<LM, determining that the channel model is an underdetermined equation, and since a joint sparsity structure is present in the massive MIMO channel, determining to reconstruct a high-dimensional channel H from a low-dimensional vector Y by a channel estimation method based on compressed sensing.

3. The time-frequency block-sparse channel estimation method based on compressed sensing according to claim 1, wherein the compressed sensing method in Step 2 specifically comprises: inputting parameters as a measurement value Y, a sensing matrix W, a step size S, and a maximum path delay L; initializing a residual vector ν.sub.0=Y, reconstructing a signal estimation value H=Ø∈.sup.LM×T, an index set r=0, letting an initial iteration count k=1, and updating a step size count I=1, the method comprising: Step 201: calculating a projection coefficient of each column of the sensing matrix on the residual vector, i.e., Z=Ψ.sup.Hν.sub.k−1; Step 202: converting a matrix Z∈.sup.LM×R into a matrix {circumflex over (Z)} of L×RM by joint sparsity of the channel, and summing {circumflex over (Z)} by row to obtain {circumflex over (Z)}=Σ.sub.i.sup.RM∥{circumflex over (Z)}∥.sub.F.sup.2∈.sup.L×1; Step 203: updating the index set: Γ.sub.k.sup.L=Γ.sub.k−1.sup.L∪{arg max ({tilde over (Z)},S)}; Step 204: extending the index set Γ.sub.k.sup.L to Γ.sub.k.sup.Li=Γ.sub.k.sup.L+iL, 1≤i≤M, and merging the index sets, Γ.sub.k=Γ.sub.k.sup.L∪Γ.sub.k.sup.L2 . . . ∪Γ.sub.k.sup.LM; Step 205: solving the estimation value of the channel H by a least squares method: Ĥ.sub.Γ.sub.k.sup.k=Ψ.sub.Γ.sub.k.sup.†Y; Step 206: converting a matrix Ĥ.sub.Γ.sub.k.sup.k∈.sup.LM×R into a matrix H̆.sub.Γ.sub.k.sup.k of L×RM, and summing H̆.sub.Γ.sub.k.sup.k by row to obtain H.sub.Γ.sub.k.sup.k=Σ.sub.i.sup.RMH̆.sub.Γ.sub.k.sup.k∈.sup.L×1; Step 207: obtaining an index set: Γ.sub.k.sup.L=arg max (H.sub.Γ.sub.k.sup.k, S); Step 208: extending the index set Γ.sub.k.sup.L to Γ.sub.k.sup.Li=Γ.sub.k.sup.L+iL, 1≤i≤M, and merging the index sets, Γ.sub.k=Γ.sub.k.sup.L∪Γ.sub.k.sup.L2 . . . ∪Γ.sub.k.sup.LM; Step 209: solving the estimation value of the channel H by a least squares method: {tilde over (H)}.sub.Γ.sub.k.sup.k=Ψ.sub.Γ.sub.k.sup.†Y; Step 210: updating the residual: ν′.sub.k=Y−Ψ{tilde over (H)}.sub.Γ.sub.k.sup.k; Step 211: if ∥ν′.sub.k∥.sub.F>∥ν.sub.k−1∥.sub.F, then {tilde over (Γ)}.sub.k={circumflex over (Γ)}.sub.k and stopping operation; Step 212: if ∥ν′.sub.k∥.sub.F=∥ν.sub.k−1∥.sub.F, then I=I+1, S=S×I, {circumflex over (Γ)}.sub.k=Γ.sub.k; Step 213: if ∥ν′.sub.k∥.sub.F<∥ν.sub.k−1∥.sub.F, then ν.sub.k=ν′.sub.k, Γ.sub.k.sup.L=Γ.sub.k.sup.L; and Step 214: k=k+1, repeating Step 201 to Step 214 until the stop condition is satisfied.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0033] FIG. 1 is a diagram showing normalized mean square errors (NMSE) at different signal-to-noise ratios (SNR) of an embodiment of the disclosure and comparative embodiments.

[0034] FIG. 2 is a diagram showing normalized mean square errors at different transmitting antenna quantities of the embodiment of the disclosure and the comparative embodiments.

DESCRIPTION OF THE EMBODIMENTS

[0035] To make the objectives, technical solutions, and advantages of the disclosure more apparent, the disclosure will be described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the disclosure and are not intended to limit the disclosure.

[0036] In an embodiment of the disclosure, an FDD downlink massive MIMO system is considered, in which an antenna quantity of a base station is M=20, and U=6 single-antenna users are simultaneously served. A total number of subcarriers of OFDM symbols is N=4096, where N.sub.P=100 subcarriers are used to transmit pilot signals. Pilots are placed all in the same manner; namely, they are distributed randomly and the pilots among different antennas are orthogonal to each other. A channel length L is 160, and a TU-6 channel model is adopted, where a number of paths S=6, path delays are respectively 0.0, 0.2, 0.5, 1.6, 2.3, 5, and path gains are respectively −3, 0, −2, −6, −8, −10. Let a coherence time T of the channel be T=4 OFDM symbols. Based on a time-frequency block sparsity and a compressed sensing framework of a massive MIMO channel, the channel estimation method includes the following steps.

[0037] Step 1: A pilot signal and a reception signal of a transmitting end are inputted, and a channel model is established as Y=ΨH+V according to the signal.

[0038] where Y∈.sup.N.sup.P.sup.×R is a reception signal matrix, H∈.sup.LM×R is a channel matrix, Ψ=∈.sup.N.sup.P.sup.×LM is a pilot matrix, V∈.sup.N.sup.P.sup.×R is a noise matrix; since N.sub.P<<LM, the channel model is an underdetermined equation, but a joint sparsity structure is present in the massive MIMO channel, and a high-dimensional channel H may be reconstructed from a low-dimensional vector Y by a channel estimation method based on compressed sensing.

[0039] Step 2: A sparse signal estimation value {tilde over (H)} is solved by a compressed sensing method according to the channel model obtained in Step 1 to further calculate an index set {tilde over (Γ)}.sub.k.

[0040] The compressed sensing method in Step 2 specifically includes the following.

[0041] Parameters are inputted as a measurement value Y, a sensing matrix Ψ, a step size S, and a maximum path delay L; a residual vector ν.sub.0=Y is initialized, a signal estimation value H=Ø∈.sup.LM×T is reconstructed, an index set Γ=Ø, letting an initial iteration count k=1, and a step size count I=1 is updated; the method includes the following steps.

[0042] Step 201: A projection coefficient of each column of the sensing matrix on the residual vector is calculated, i.e., Z=Ψ.sup.Hν.sub.k−1.

[0043] Step 202: A matrix Z∈.sup.LM×R is converted into a matrix {circumflex over (Z)} of L×RM by joint sparsity of the channel, and {circumflex over (Z)} is summed by row to obtain {tilde over (Z)}=Σ.sub.i.sup.RM∥{circumflex over (Z)}∥.sub.F.sup.2∈.sup.L×1.

[0044] Step 203: The index set is updated: Γ.sub.k.sup.L=Γ.sub.k−1.sup.L∪{arg max ({tilde over (Z)}, S)}.

[0045] Step 204: The index set Γ.sub.k.sup.L is extended to Γ.sub.k.sup.Li=Γ.sub.k.sup.L+iL, 1≤i≤M, and the index sets are merged, Γ.sub.k=Γ.sub.k.sup.L∪Γ.sub.k.sup.L2 . . . ∪Γ.sub.k.sup.LM.

[0046] Step 205: The estimation value of the channel H is solved by a least squares method: Ĥ.sub.Γ.sub.k.sup.k=Ψ.sub.Γ.sub.k.sup.†Y.

[0047] Step 206: A matrix Ĥ.sub.Γ.sub.k.sup.k∈.sup.LM×R is converted in to a matrix H̆.sub.Γ.sub.k.sup.k of L×RM, and H̆.sub.Γ.sub.k.sup.k is summed by row to obtain H.sub.Γ.sub.k.sup.k=Σ.sub.i.sup.RMH̆.sub.Γ.sub.k.sup.k∈.sup.L×1.

[0048] Step 207: An index set is obtained: Γ.sub.k.sup.L=arg max (H.sub.Γ.sub.k.sup.k, S).

[0049] Step 208: The index set Γ.sub.k.sup.L is extended to Γ.sub.k.sup.Li=Γ.sub.k.sup.L+iL, 1≤i≤M, and the index sets are merged, Γ.sub.k=Γ.sub.k.sup.L∪Γ.sub.k.sup.L2 . . . ∪Γ.sub.k.sup.LM.

[0050] Step 209: The estimation value of the channel H is solved by a least squares method: {tilde over (H)}.sub.Γ.sub.k.sup.k=Ψ.sub.Γ.sub.k.sup.†Y.

[0051] Step 210: The residual is updated: ν′.sub.k=Y−Ψ{tilde over (H)}.sub.Γ.sub.k.sup.k.

[0052] Step 211: If ∥ν′.sub.k∥.sub.F>∥ν.sub.k−1∥.sub.F, then {tilde over (Γ)}.sub.k={circumflex over (Γ)}.sub.k and operation is stopped.

[0053] Step 212: If ∥ν′.sub.k∥.sub.F=∥ν.sub.k−1∥.sub.F, then I=I+1, S=S×I, {circumflex over (Γ)}.sub.k=Γ.sub.k.

[0054] Step 213: If ∥ν′.sub.k∥.sub.F<∥ν.sub.k−1∥.sub.F, then ν.sub.k=ν′.sub.k, Γ.sub.k.sup.L=Γ.sub.k.sup.L.

[0055] Step 214: k=k+1; Step 201 to Step 214 are repeated until the stop condition is satisfied.

[0056] Step 3: According to the index set {tilde over (Γ)}.sub.k obtained in Step 2, a channel matrix estimation value {tilde over (H)}.sub.{tilde over (Γ)}.sub.k may be solved, i.e., {tilde over (H)}.sub.{tilde over (Γ)}.sub.k=Ψ.sub.{tilde over (Γ)}.sub.k.sup.†Y, where a superscript “†” represents a pseudoinverse, i.e., Ψ.sub.{tilde over (Γ)}.sub.k.sup.† represents a pseudoinverse with respect to Ψ.sub.{tilde over (Γ)}.sub.k, and according to the obtained channel matrix estimation value {tilde over (H)}.sub.{tilde over (Γ)}.sub.k, after a baseband signal is demodulated, data information of the transmitting end is outputted.

[0057] To evaluate the performance of the disclosure, when the antenna quantity M is 16, the step size s is 2, and a threshold parameter μ of the algorithm reconstruction precision is all 0.001, normalized least mean square errors of channel estimation algorithms at different signal-to-noise ratios are calculated, and the result is shown in FIG. 1.

[0058] To further evaluate the performance of the disclosure, when the signal-to-noise ratio is 20 dB, the step size s is 2, the threshold parameter μ of the algorithm reconstruction precision is all 0.001, normalized least mean square errors of channel estimation algorithms at different transmitting antennas of the base station are calculated, and the result is shown in FIG. 2.

[0059] The normalized least mean square error is defined as follows:

[00001]$\mathrm{NMSE}=\frac{\underset{j=1}{\overset{N_e}{.Math.}}\sqrt{\underset{i=1}{\overset{\mathrm{LMT}}{.Math.}}{\left(\tilde{H}-H\right)}^2}}{N_e\mathrm{MT}}$

[0060] where N.sub.e represents an operation count of the algorithm at each signal-to-noise ratio, and herein, N.sub.e is 20.

[0061] According to FIG. 1, in the embodiment of the disclosure, with the generalized adaptive mechanism introduced, the proposed gBAMP algorithm can detect the minimum precision achieved by the reconstruction and automatically end the reconstruction process without the constraint of the threshold parameter μ. The experimental result shows that the algorithm achieves good performance close to that of the exact-LS algorithm and is superior to other algorithms.

[0062] According to FIG. 2, as the antenna quantity increases, the precision of channel reconstruction is affected, and the reason lies in that the increase in the antenna quantity leads to pilot insufficiency. The reconstruction by the SAMP-Block algorithm fails at the threshold parameter. However, in the embodiment of the disclosure, when the antenna quantity is 16 more, the proposed gBAMP algorithm exhibits better performance, and exhibits optimal performance in both precision and stability that are even better than those of the exact-LS algorithm.

[0063] The time-frequency block-sparse channel estimation method based on compressed sensing in the embodiment of the disclosure, i.e., a generalized block adaptive gBAMP algorithm, has good reconstruction performance and is applicable to occasions requiring pilot-assisted channel estimation of a wireless communication system.

[0064] Simulation shows that the method of the disclosure can quickly and accurately recover massive MIMO channel information of which a sparsity degree is unknown.

[0065] It will be understood that modifications and variations may be made by persons skilled in the art according to the above description, and all such modifications and variations are intended to be included within the scope of the disclosure as defined in the appended claims.