CN114978381A - Deep learning-based compressed sensing processing method and device for broadband signals - Google Patents
Deep learning-based compressed sensing processing method and device for broadband signals Download PDFInfo
- Publication number
- CN114978381A CN114978381A CN202210450842.1A CN202210450842A CN114978381A CN 114978381 A CN114978381 A CN 114978381A CN 202210450842 A CN202210450842 A CN 202210450842A CN 114978381 A CN114978381 A CN 114978381A
- Authority
- CN
- China
- Prior art keywords
- signal
- admm
- net
- sampling
- input
- 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.)
- Granted
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/382—Monitoring; Testing of propagation channels for resource allocation, admission control or handover
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Abstract
The invention discloses a method and a device for compressed sensing processing of broadband signals based on deep learning, wherein the method comprises the following steps: performing multiple-set sampling on an input broadband signal on a time domain to obtain a discrete sampling sequence; preprocessing the obtained discrete sampling sequence and the input broadband signal and constructing a data set; designing a deep learning-based broadband signal reconstruction neural network model ADMM-net, and initializing the ADMM-net; in the model training stage, inputting signal data of a training set into the ADMM-net, and continuously minimizing a loss function through an optimization algorithm to obtain an optimal neural network parameter; and inputting the signal data of the test set into the trained ADMM-net model to obtain a reconstructed broadband signal. The ADMM-net combined with deep learning has higher signal reconstruction accuracy under the conditions of lower sampling rate and lower signal-to-noise ratio after being trained by a network model, and has certain generalization capability.
Description
Technical Field
The invention relates to the technical field of broadband signal processing, in particular to a method and a device for processing broadband signals based on deep learning by compressed sensing, an intelligent terminal and a storage medium.
Background
With the explosive growth of electromagnetic equipment and information systems, static spectrum management modes are no longer effective, the utilization rate of a large number of frequency bands is low, a few frequency bands which are not distributed and suitable for data transmission are left, and the contradiction that wireless spectrum resources are locally tensed and are wholly idle is more obvious. Meanwhile, with the rapid development of mobile services and the coming of the era of internet of things, the technical route and the key technology adopted by the future cellular network require transmission bandwidth of at least hundreds of megahertz, hundreds of transmission antennas, and ultra-densely deployed base stations and support massive users.
The Compressed Sensing (CS) theory provides a new solution for acquisition and processing of broadband signals. In the prior art: currently, the mainstream compressive Sampling framework includes Multi-set Sampling (MCS), Random Demodulation Sampling (RDS), and Modulated Wideband Converter (MWC). The multi-set sampling technology is a period non-uniform sampling under-nyquist sampling technology, and can realize compression sampling of broadband signals through a plurality of low-speed ADCs with the same sampling rate but different sampling starting moments. According to the broadband spectrum sensing scheme based on multiple-time set sampling in the prior art, under-Nyquist sampling of broadband signals is achieved through a compression sampling mode of a low-rate multi-channel architecture, and then multi-band signals are recovered through a greedy algorithm to estimate the occupied channel position, so that reconstruction of input broadband signals is achieved. In the scheme, when the sampling rate and the signal-to-noise ratio of the input broadband signal are low, the reconstruction precision of the signal is greatly improved.
Namely, compressed sensing processing of broadband signals in the prior art is insufficient in reconstruction accuracy, robustness and other aspects, and the problem of poor reconstruction performance and the like exists for broadband spectrum sensing based on under-nyquist sampling.
Thus, there is a need for improvements and enhancements in the art.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a method, an apparatus, an intelligent terminal and a storage medium for compressed sensing processing of a broadband signal based on deep learning, aiming at the above defects in the prior art, and the present invention can solve the problems of poor reconstruction performance and the like in broadband spectrum sensing based on under-nyquist sampling.
In order to solve the above technical problem, a first aspect of the present invention provides a method for compressed sensing processing of a wideband signal based on deep learning, where the method includes:
acquiring an input broadband signal, and performing multi-fold sampling on the input broadband signal on a time domain to obtain a discrete sampling sequence;
preprocessing the obtained discrete sampling sequence and an input broadband signal and constructing a data set, wherein the data set comprises: signal data of the training set and signal data of the test set;
designing a deep learning-based broadband signal reconstruction neural network model ADMM-net, and initializing the ADMM-net;
in the model training stage, inputting signal data of a training set into the ADMM-net, and continuously minimizing a loss function through an optimization algorithm to obtain an optimal neural network parameter;
and in a testing stage or an application stage, inputting the signal data of the test set into the trained ADMM-net model to obtain a reconstructed broadband signal.
The method for processing the compressed sensing of the broadband signal based on the deep learning includes the steps of obtaining an input broadband signal, and performing multiple-set sampling on the input broadband signal in a time domain to obtain a discrete sampling sequence:
obtaining input wideband signal x (t), determining sampling mode by sequence forward selection algorithm
Using multiple sets of samples att＝(mL+c i )T,i＝1,...,p,
Sampling at a moment to obtain a discrete sampling sequence x ci [n]Where T denotes the nyquist sampling time period interval of the input broadband signal, the multiple-sampling period interval is L times as large as the nyquist sampling theorem, so that the sampling frequency is reduced to 1/L of the nyquist sampling frequency.
The method for processing the compressed sensing of the broadband signal based on the deep learning includes the following steps:
for the obtained discrete sampling sequence x ci [n]Performing a discrete Fourier transform to obtain a frequency domain representation of the sample sequence
At the same time, a frequency domain representation of the resulting sample sequence is obtained
And corresponding frequency domain signal X [ k ] of input broadband signal]And forming a data set, wherein the number of the training sets and the number of the testing sets are in a set proportion.
The method for processing the compressed sensing of the broadband signal based on the deep learning comprises the following steps of designing a deep learning-based broadband signal reconstruction neural network model ADMM-net, and initializing the ADMM-net:
an ADMM algorithm is used as a signal reconstruction algorithm of compressed sensing, and an ADMM-net reconstructed broadband signal with expandable depth based on deep learning is designed;
the parameters of the ADMM-net model are initialized to the corresponding parameter values in the ADMM algorithm.
The method for processing the compressed sensing of the broadband signal based on the deep learning comprises the following steps of inputting signal data of a training set into the ADMM-net in a model training stage, and obtaining an optimal neural network parameter by continuously minimizing a loss function through an optimization algorithm:
in the model training phase, the frequency domain representation of the sampling sequence of the training set is used
Inputting the initialized ADMM-net to obtain an initially reconstructed frequency domain signal X ir [k]；
Constructing the initially reconstructed frequency domain signal X ir [k]Frequency domain signal X [ k ] with input broadband signal]A loss function E (θ) in between, where θ is the relevant neural network parameter;
and continuously minimizing the loss function E (theta) by using an optimization algorithm to obtain an optimal neural network parameter theta, so as to obtain the trained ADMM-net model.
The method for processing the compressed sensing of the broadband signal based on the deep learning includes, in a test phase or an application phase, inputting signal data of a test set into a trained ADMM-net model to obtain a reconstructed broadband signal, where the step of obtaining the reconstructed broadband signal includes:
representing the frequency domain of the sample sequence of the test set during the test phase or the application phase
Inputting the trained ADMM-net model for processing to obtain a reconstructed frequency domain signal X r [k]。
A device for compressed sensing processing of a broadband signal based on deep learning, wherein the device comprises:
the multiple-set sampling processing module is used for acquiring an input broadband signal and performing multiple-set sampling on the input broadband signal in a time domain to obtain a discrete sampling sequence;
the preprocessing module is used for preprocessing the obtained discrete sampling sequence and the input broadband signal and constructing a data set;
the method comprises the steps that a setting and initializing module is used for designing a deep learning-based broadband signal reconstruction neural network model ADMM-net and initializing the ADMM-net;
the neural network training module is used for inputting signal data of a training set into the ADMM-net in a model training stage, and continuously minimizing a loss function through an optimization algorithm to obtain an optimal neural network parameter;
and the testing and application module is used for inputting the signal data of the test set into the trained ADMM-net model in a testing stage or an application stage to obtain a reconstructed broadband signal.
An intelligent terminal comprising a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors comprises means for performing the method of the present invention.
A non-transitory computer readable storage medium having instructions stored thereon that, when executed by a processor of an electronic device, enable the electronic device to perform any of the methods described herein.
Has the advantages that: compared with the prior art, the invention provides a broadband signal compressive sensing processing method based on deep learning. Firstly, based on multiple-time set sampling, performing compression sampling on an input broadband signal on a time domain to obtain a discrete sampling sequence, then preprocessing the obtained discrete sampling sequence and the input broadband signal and constructing a data set, and meanwhile, combining deep learning to design a depth-extensible broadband signal reconstruction neural network model ADMM-net to reconstruct the input broadband signal. In the model training stage, inputting the signal data of the training set into the initialized ADMM-net to obtain an initially reconstructed frequency domain signal, then constructing a loss function between the initially reconstructed frequency domain signal and the frequency domain signal of the input broadband signal, and continuously minimizing the loss function by utilizing an optimization algorithm, thereby obtaining the optimal neural network parameters. And in a testing stage or an application stage, inputting the signal data of the test set into the trained ADMM-net model to obtain a reconstructed broadband signal. The numerical analysis of the experimental result shows that compared with the common reconstruction algorithm, the ADMM-net combined with deep learning in the invention has higher reconstruction accuracy under the conditions of lower sampling rate and lower signal-to-noise ratio after being trained by a network model, and has certain generalization capability.
Drawings
FIG. 1 is a diagram of a compressed sensing theory framework.
Fig. 2 is a flowchart of a specific implementation of a method for processing a wideband signal based on deep learning according to an embodiment of the present invention.
Fig. 3 is a diagram of a multi-fold sampling framework of a method for processing a deep learning-based wideband signal by compressed sensing according to an embodiment of the present invention.
Fig. 4 is a schematic diagram illustrating the spectrum division and transfer of an input wideband signal in a method for processing a wideband signal based on deep learning by compressed sensing according to an embodiment of the present invention.
Fig. 5 is a data flow diagram of ADMM-net of a method for compressive sensing processing of a broadband signal based on deep learning according to an embodiment of the present invention.
Fig. 6 is a diagram illustrating the reconstruction results of the SOMP, ADMM, and ADMM-net algorithms under different snr conditions, where the channel number L is 64 and the source channel k is 5 in the method for processing the compressed sensing of the deep learning-based wideband signal according to the embodiment of the present invention.
Fig. 7 is a schematic block diagram of a device for compressed sensing processing of a wideband signal based on deep learning according to an embodiment of the present invention.
Fig. 8 is a schematic block diagram of an internal structure of an intelligent terminal according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and effects of the present invention clearer and clearer, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
With the explosive growth of electromagnetic equipment and information systems, static spectrum management modes are no longer effective, the utilization rate of a large number of frequency bands is low, a few frequency bands which are not distributed and suitable for data transmission are left, and the contradiction that wireless spectrum resources are locally tensed and are wholly idle is more obvious. Meanwhile, with the rapid development of mobile services and the coming of the era of internet of things, the technical route and the key technology adopted by the future cellular network require transmission bandwidth of at least hundreds of megahertz, hundreds of transmission antennas, and ultra-densely deployed base stations and support massive users. The lowest undistorted sampling rate of a signal must be greater than or equal to twice the maximum bandwidth of the signal, according to the nyquist sampling theorem. With the increase of signal bandwidth, it becomes increasingly difficult to implement high-speed sampling of broadband signals and storage, transmission and real-time processing of large-capacity data, which brings great challenges to conventional communication systems.
The Compressed Sensing (CS) theory provides a new solution for acquisition and processing of broadband signals. Compared with a traditional Nyquist sampling system which performs high-speed sampling and then compresses data, compressed sensing directly performs compressed sampling on signals with sparsity or compressibility at a sub-Nyquist sampling rate, and then combines a reconstruction algorithm to reconstruct original signals, wherein the theoretical framework is shown in figure 1, and the original signals are subjected to sparse transformation, compressed sampling and signal reconstruction. The compressed sensing obviously reduces the requirements of signal sampling on hardware equipment such as storage, transmission, analysis and processing and the like in a communication system, thereby bringing a new opportunity for solving the challenges of high cost, low efficiency, information redundancy and the like which are difficult to break through by the traditional sampling method.
In broadband spectrum sensing based on under-nyquist sampling, selection of a compressive sampling framework and design of a sparse signal reconstruction algorithm play a main role. The compressive sampling framework determines a compressive sampling mode of the broadband signal, and is usually related to the design of a random observation matrix, so that the subsequent signal reconstruction process is influenced. The reconstruction algorithm has the largest influence on the whole broadband spectrum sensing, and determines the whole broadband spectrum sensing performance in relation to the reconstruction complexity and accuracy of the signal. Therefore, it is very important to research and design a combination of a stable compressive sampling frame with low reconstruction complexity and a reconstruction algorithm for improving the broadband spectrum sensing performance.
In the prior art: currently, the mainstream compression Sampling framework includes Multi-set Sampling (MCS), Random Demodulation Sampling (RDS), and Modulated Wideband Converter (MWC). The multi-time set sampling technology is an under-Nyquist sampling technology with non-uniform sampling period, can realize compression sampling of broadband signals through low-speed ADCs with the same sampling rate but different sampling starting moments, and compared with other time-based ADC interleaved sampling schemes, the multi-time set sampling technology only needs to select p < L channels from L channels for sampling, has low sampling matrix dimension, reduces the sampling rate of a compression sampling system, and has a simple front-end circuit, lower cost of hardware realization, but has the main difficulty of accurate design of sampling time delay. The reconstruction algorithm is essentially to reconstruct an original signal by solving an underdetermined linear equation, and the currently commonly used reconstruction algorithm is a greedy algorithm such as a Synchronous Orthogonal Matching Pursuit (SOMP); convex relaxation algorithms, such as Alternating Direction multiplier Algorithms (ADMM) and the like. The greedy algorithm has low calculation complexity, but has poor reconstruction performance when the observed value is less, and the operand gradually increases along with the increase of the observed value. While the convex relaxation algorithm can achieve approximate reconstruction of the input broadband signal with a small number of samples, but the reconstruction complexity is large. A broadband spectrum sensing scheme based on multi-fold set sampling is provided in the prior art, the under-Nyquist sampling of broadband signals is realized through a compression sampling mode of a low-rate multi-channel system structure, and then multi-band signals are recovered through a greedy algorithm to estimate the positions of occupied channels, so that the reconstruction of the input broadband signals is realized. In the scheme, when the sampling rate and the signal-to-noise ratio of the input signal are low, the reconstruction precision of the signal is greatly improved.
In order to solve the problems of the prior art, the embodiment provides a method for processing the compressed sensing of the broadband signal based on the deep learning. Firstly, based on multiple-time set sampling, performing compression sampling on an input broadband signal on a time domain to obtain a discrete sampling sequence, then preprocessing the obtained discrete sampling sequence and the input broadband signal and constructing a data set, and meanwhile, combining deep learning to design a depth-extensible broadband signal reconstruction neural network model ADMM-net to reconstruct the input broadband signal. In the model training stage, inputting the signal data of the training set into the initialized ADMM-net to obtain an initially reconstructed frequency domain signal, then constructing a loss function between the initially reconstructed frequency domain signal and the frequency domain signal of the input broadband signal, and minimizing the loss function by utilizing an optimization algorithm, thereby obtaining the optimal neural network parameters. And in a testing stage or an application stage, inputting the signal data of the test set into the trained ADMM-net model to obtain a reconstructed broadband signal.
Numerical analysis of experimental results shows that compared with a common reconstruction algorithm, the deep learning combined broadband signal reconstruction neural network model ADMM-net has higher reconstruction accuracy under the conditions of lower sampling rate and lower signal-to-noise ratio after being trained by a network model, and has certain generalization capability.
Exemplary method
The method of the embodiment can be applied to an intelligent terminal, and in specific implementation, as shown in fig. 2, the method for processing the compressed sensing of the broadband signal based on the deep learning provided by the embodiment of the present invention specifically includes the following steps:
s100, acquiring an input broadband signal, and performing multi-fold sampling on the input broadband signal in a time domain to obtain a discrete sampling sequence;
when the method is implemented specifically, an input broadband signal is acquired first, a sampling mode is determined through a sequence forward selection algorithm, and multiple-time-set sampling is performed on a time domain to obtain a discrete sampling sequence.
Step S200, preprocessing the obtained discrete sampling sequence and the input broadband signal and constructing a data set, wherein the data set comprises: signal data of the training set and signal data of the test set;
in the step, the frequency domain representation of the sampling sequence is obtained through discrete Fourier transform (DFT transform) on the obtained discrete sampling sequence;
and meanwhile, forming a data set by the frequency domain representation of the obtained sampling sequence and the corresponding frequency domain signals of the input broadband signals, wherein the number of the training sets and the number of the test sets are in a set proportion.
S300, designing a deep learning-based broadband signal reconstruction neural network model (ADMM-net), and initializing the ADMM-net;
in the step, an ADMM algorithm is used as a signal reconstruction algorithm of compressed sensing, and a deep learning-based depth-extensible broadband signal reconstruction neural network model ADMM-net reconstruction broadband signal is designed;
the parameters of the ADMM-net model are initialized to the corresponding parameter values in the ADMM algorithm.
S400, in a model training stage, inputting signal data of a training set into the ADMM-net, and constantly minimizing a loss function through an optimization algorithm to obtain an optimal neural network parameter;
in the step, in the model training phase, the frequency domain representation of the sampling sequence of the training set is input into the initialized ADMM-net to obtain an initially reconstructed frequency domain signal;
constructing a loss function between the initially reconstructed frequency domain signal and a frequency domain signal of an input wideband signal;
and (3) continuously minimizing a loss function by using a Conjugate Gradient (CG) method to obtain an optimal neural network parameter, so as to obtain the trained ADMM-net model.
And S500, inputting the signal data of the test set into the trained ADMM-net model in a test stage or an application stage to obtain a reconstructed broadband signal.
In this step, in a testing stage or an application stage, the frequency domain representation of the sampling sequence of the test set is input into the trained ADMM-net model for processing, and a reconstructed frequency domain signal is obtained.
The process of the invention is described in further detail below by way of more detailed examples:
in the embodiment of the present invention, regarding that the input wideband signal is obtained in step S100, the obtaining of the discrete sampling sequence by performing multiple-times-set sampling on the input wideband signal in the time domain is specifically implemented as follows:
obtaining input wideband signal x (t), using multiple set sampling at t ═ (mL + c) i )T,i＝1,2,…,p,
Sampling is carried out at time to obtain a discrete sampling sequence, wherein T represents the Nyquist sampling time period interval of an input broadband signal, compared with the Nyquist sampling theorem, the time interval of the multiple-time-set sampling period is L times, so the sampling frequency is reduced to 1/L of the Nyquist sampling frequency, and fig. 3 is a multiple-time-set sampling frame diagram of the embodiment of the invention. Collection
Represents p selected from {0,1,2, …, L-1}<The L different integers are referred to as the (L, p) sampling pattern in the multiple set sampling. The ith discrete sample sequence x ci [n]Can be defined as equation (1).
Wherein
i＝1,2,…,p，
The sampling delay of the ith channel is indicated,
representing the nyquist sampling frequency.
In the embodiment of the invention, a sequence forward selection algorithm is adopted to determine the sampling mode.
Further, in the embodiment of the present invention, the following is specifically implemented with respect to step S200, the preprocessing of the obtained discrete sampling sequence and the input wideband signal, and the construction of the data set:
in the present invention, in step S200, the sequence x obtained by sampling ci [n]Performing a discrete Fourier transform (DFT transform) to obtain a frequency domain representation of the ith sample sequence
Namely that
Wherein k is c 0,1,2, …, M-1, i-1, 2, …, p, M represents the number of sampling points obtained for each sampling channel.
Set at the Nyquist sampling frequency f s The input broadband signal x (T) is sampled uniformly with time interval T to obtain discrete sequence x [ n ] with total length K]. Then aligning the sequences x [ n ]]Performing discrete Fourier transform to obtain frequency domain signal X [ k ] of input broadband signal]I.e. by
Where K is 0,1,2, …, K-1, K representing the number of points in the discrete sequence x [ n ].
Frequency domain representation of the obtained sampling sequence
i-1, 2, …, p and corresponding frequency domain signal X [ k ] of the input broadband signal]Data sets were composed with the number of training sets and test sets in a set ratio (4: 1). The training set is used for training the network model, and the testing set is used for testing the reconstruction performance of the network model and evaluating the generalization capability of the network model.
Frequency domain representation of sampling sequence based on formulas (2) and (3)
Frequency domain signal X [ k ] with input broadband signal]The matrix expression of the multiple-time-set sampling input and output can be obtained as
Y＝AX(4)
Wherein Y ∈ C p×M A frequency domain representation matrix representing a sequence of samples, whose expression is:
t represents the nyquist sampling time period interval of the input broadband signal; x belongs to C L×M A frequency domain representation matrix for representing the input broadband signal, and a frequency domain signal X [ k ] of the input broadband signal]Dividing the signal into L channels by 1/LT unit, moving L-1 channels out of the first channel to left by 1/LT unit, and obtaining frequency domain signal X [ k ] of X and input broadband signal by the frequency spectrum division transfer diagram as shown in FIG. 4]A relational expression of (i), i.e
Where L denotes the number of channels of the spectrum division, M denotes the number of signal points per channel, and K-L · M denotes the number of discrete signal points of the input wideband signal based on the nyquist sampling frequency.
Further, in step S300, the present invention designs a deep learning based broadband signal reconstruction neural network model ADMM-net, and initializes the ADMM-net as follows:
specifically, the present invention reconstructs the input wideband signal by using the convex relaxation method, and the regularization unconstrained model thereof can be expressed as:
where X denotes the frequency domain representation matrix of the input wideband signal to be reconstructed, a is the measurement matrix of the multiple set samples, Y denotes the frequency domain representation matrix of the sample sequence, μ is the regularization parameter, and the auxiliary variables Z ═ Z [1], Z [2], …, Z [ p ] ].
Extended Lagrange expansion L of the above formula (8) P (X, Z, α) is:
where α represents the lagrangian multiplier for the dual variable and ρ represents the lagrangian multiplier coefficient.
By using the ADMM algorithm as a signal reconstruction algorithm of compressed sensing, the { X, Z, alpha } can be alternately optimized and solved through three subproblems separated by the above formula, and the following results are obtained:
where k denotes the kth iteration, X (k) A frequency domain representation matrix, Z, representing the kth iterative reconstruction (k) Auxiliary variables Z, M representing the kth iterative update (k) Dual variables α, ρ, η and representing the kth iterative update
And respectively representing the relevant parameters of the reconstruction signal operation, the auxiliary variable updating operation and the dual variable updating operation.
Combining deep learning and data flow graph ideas, the iterative solution of the three sub-problems of the ADMM algorithm can be designed into a data flow graph, so that the data flow graph is expanded into an alternating direction multiplier-sub network capable of setting any number of layers, the data flow graph is shown as fig. 5, and fig. 5 is a data flow graph of the embodiment of the inventionA data flow diagram of the ADMM-net model of (a). The data flow graph is composed of nodes corresponding to different operations in the ADMM algorithm, and the directed edges correspond to data flow between the operations. At the k-th stage of the graph, three operation types in the ADMM algorithm map three types of nodes, i.e., defined as signal reconstruction layers (X), respectively (k) ) Nonlinear conversion layer (Z) (k) ) And a multiplication update layer (M) (k) ). The entire data flow graph is a multiple repetition of the above stages, corresponding to successive iterations in the ADMM algorithm.
Parameters for layers in ADMM-net model
It is initialized to the corresponding parameter value in the ADMM algorithm.
In a further embodiment, in the step S400 of the present invention, in the model training phase, the signal data of the training set is input into the ADMM-net, and the loss function is continuously minimized through the optimization algorithm, so as to obtain the optimal neural network parameters specifically:
in the model training phase, the frequency domain representation of the sampling sequence of the training set is used
Inputting the initialized ADMM-net to obtain an initially reconstructed frequency domain signal X ir [k]。
Initial reconstruction frequency domain signal X for constructing training set in embodiment of the invention ir [k]And the frequency domain signal X [ k ] of the input broadband signal]The loss function of (a) is defined as:
wherein
An initial reconstruction frequency domain signal representing the ith signal data in the N signal data of the training set, (Y, theta) respectively representing a frequency domain representation matrix Y of the sampling sequence and a network parameter theta, wherein theta comprises a signal reconstruction layer (X) (k) ) A nonlinear conversion layer (Z) (k) ) And a multiplication update layer (M) (k) ) Parameter (d) of
X (i) A frequency domain signal representing the input wideband signal of the training set ith signal data.
In the embodiment of the invention, in model training, an optimization algorithm of network parameters is a Conjugate Gradient method (CG), and the Gradient descending direction of the optimization algorithm is determined by an Armijo criterion of a non-precise one-dimensional search method. And continuously minimizing the loss function E (theta) by a conjugate gradient method to obtain an optimal neural network parameter theta, thereby obtaining the trained ADMM-net model.
Further, in step S500 of the embodiment of the present invention, in the testing stage or the application stage, the signal data of the test set is input into the trained ADMM-net model, and the obtaining of the reconstructed wideband signal specifically includes:
in the embodiment of the invention, in the test stage or the application stage, the reconstruction effect of the broadband signal under different signal-to-noise ratios is tested by using the trained ADMM-net model. Frequency domain representation of a sample sequence of a test set
Inputting the trained ADMM-net model for processing to obtain a reconstructed frequency domain signal X r [k]。
In the test phase experiment, the final reconstruction error of the frequency domain signal is expressed by mean square error MSE:
wherein, X r [k]Frequency domain signal, X k, representing a reconstructed wideband signal]A frequency domain signal representing the input wideband signal.
As can be seen from the above, the method for processing compressed sensing of a broadband signal based on deep learning according to the embodiments of the present invention mainly includes five steps, where first, an input broadband signal is obtained, and multiple sets of sampling are performed on the input broadband signal in a time domain to obtain a discrete sampling sequence; secondly, preprocessing the obtained discrete sampling sequence and the input broadband signal and constructing a data set; designing a deep learning-based broadband signal reconstruction neural network model ADMM-net, and initializing the ADMM-net; fourthly, in the model training stage, inputting the signal data of a training set into the ADMM-net, and continuously minimizing a loss function through an optimization algorithm to obtain an optimal neural network parameter; and the fifth step is that in the testing stage or the application stage, the signal data of the testing set is input into the trained ADMM-net model to obtain a reconstructed broadband signal.
The implementation method of the invention comprises the following steps: the method comprises the steps of performing multiple-time-set sampling on 300 input broadband signals at two different compression sampling rates of p/L (17%) and p/L (42%), preprocessing the signals to form a data set, wherein the number of data pairs of a training set and a test set is a set proportion (4:1), the signal-to-noise ratio of the broadband signals of the training set is uniformly set in a range of [5,30] dB, and the signal-to-noise ratio of the broadband signals of the test set is set to be 5dB, 10dB, 15dB, 20dB, 25dB and 30 dB.
In the model training stage, signal data of a training set obtained by two different compression sampling rates of p/L (17%) and p/L (42%) are respectively input into an ADMM-net model, and an optimization algorithm is used for constantly minimizing a loss function to obtain an optimal neural network parameter, so that the trained ADMM-net model is obtained.
And in a test stage or an application stage, inputting the signal data of the test set into the trained ADMM-net model to obtain the reconstruction result of the broadband signal under the condition of different signal-to-noise ratios.
In the present invention, the channel L ═ 64 of the wideband signal x (t) includes 5 source channels with different bandwidths, for the signal data with two different compressed sampling rates, p/L ═ 17% and p/L ═ 42%, under different snr conditions, the signal reconstruction results of the three reconstruction algorithms SOMP, ADMM and ADMM-net are shown in fig. 6, fig. 6 is a schematic diagram of the reconstruction results of the three reconstruction algorithms SOMP, ADMM and ADMM-net in the embodiment of the present invention, where the channel number L ═ 64, the source channel k ═ 5, and under different snr conditions.
According to the results of the test set, compared with a common reconstruction algorithm, the ADMM-net combined with deep learning has higher reconstruction accuracy under the conditions of lower sampling rate and lower signal-to-noise ratio after being trained by a network model, and has certain generalization capability.
The invention also has the following advantages:
1) and combining the idea of deep learning, firstly carrying out frequency domain reconstruction on the broadband signal by using the alternative direction multiplier network capable of expanding the depth, and the result proves that the reconstruction performance of the broadband signal can be improved.
2) And training neural network parameters of the alternative direction multiplier network by constructing a loss function between a network reconstruction result and a true value and using an Armijo criterion as a conjugate gradient method of a gradient descent direction to continuously minimize the loss function, so that the reconstruction performance of the broadband signal and the generalization of a network model are improved.
3) And the alternative direction multiplier network based on multiple-time set sampling and deep learning is used as a compressed sensing frame of the broadband signal, so that the compressed sensing performance of the broadband signal is improved under the conditions of a lower sampling rate and a lower signal-to-noise ratio.
Exemplary device
As shown in fig. 7, an embodiment of the present invention provides a device for compressed sensing processing of a wideband signal based on deep learning, including:
a multiple sampling processing module 410, configured to acquire an input wideband signal, and perform multiple sampling on the input wideband signal in a time domain to obtain a discrete sampling sequence;
a preprocessing module 420, configured to preprocess the obtained discrete sampling sequence and the input wideband signal and construct a data set;
the setting and initializing module 430 is used for designing a deep learning-based broadband signal reconstruction neural network model ADMM-net and initializing the ADMM-net;
the neural network training module 440 is used for inputting the signal data of the training set into the ADMM-net in the model training stage, and continuously minimizing the loss function through an optimization algorithm to obtain the optimal neural network parameters;
a testing and applying module 450, configured to, in a testing phase or an applying phase, input signal data of the test set into the trained ADMM-net model to obtain a reconstructed wideband signal, as described above.
Based on the above embodiment, the present invention further provides an intelligent terminal, and a schematic block diagram thereof may be as shown in fig. 8. The intelligent terminal of the embodiment of the invention can be an intelligent television, and comprises a processor, a memory, a network interface and a display screen which are connected through a system bus. Wherein, the processor of the intelligent terminal is used for providing calculation and control capability. The memory of the intelligent terminal comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program to run on the non-volatile storage medium. The network interface of the intelligent terminal is used for being connected and communicated with an external terminal through a network. The computer program is executed by a processor to implement a method for compressed sensing processing of a wideband signal based on deep learning. The display screen of the intelligent terminal can be a liquid crystal display screen or an electronic ink display screen.
It will be understood by those skilled in the art that the block diagram of fig. 8 is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation to the intelligent terminal to which the solution of the present invention is applied, and a specific intelligent terminal may include more or less components than those shown in the figure, or combine some components, or have different arrangements of components.
In one embodiment, an intelligent terminal is provided that includes a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for:
acquiring an input broadband signal, and performing multi-fold sampling on the input broadband signal on a time domain to obtain a discrete sampling sequence;
preprocessing the obtained discrete sampling sequence and the input broadband signal and constructing a data set;
designing a deep learning-based broadband signal reconstruction neural network model ADMM-net, and initializing the ADMM-net;
in the model training stage, inputting signal data of a training set into the ADMM-net, and continuously minimizing a loss function through an optimization algorithm to obtain an optimal neural network parameter;
and in a testing stage or an application stage, inputting the signal data of the test set into the trained ADMM-net model to obtain a reconstructed broadband signal.
The step of acquiring an input wideband signal and performing multiple-set sampling on the input wideband signal in a time domain to obtain a discrete sampling sequence includes:
obtaining input wideband signal x (t), determining sampling mode by sequence forward selection algorithm
Sampling at a moment to obtain a discrete sampling sequence x ci [n]Where T denotes the nyquist sampling time period interval of the input broadband signal, the multiple-sampling period interval is L times as large as the nyquist sampling theorem, so that the sampling frequency is reduced to 1/L of the nyquist sampling frequency.
The steps of preprocessing the obtained discrete sampling sequence and the input broadband signal and constructing a data set comprise:
for the obtained discrete sampling sequence x ci [n]Performing a discrete Fourier transform to obtain a frequency domain representation of the sample sequence
At the same time, a frequency domain representation of the resulting sample sequence is obtained
And corresponding frequency domain signal X [ k ] of input broadband signal]And forming a data set, wherein the number of the training sets and the number of the testing sets are in a set proportion.
The method for processing the compressed sensing of the broadband signal based on the deep learning comprises the following steps of designing a deep learning-based broadband signal reconstruction neural network model ADMM-net, and initializing the ADMM-net:
an ADMM algorithm is used as a signal reconstruction algorithm of compressed sensing, and an ADMM-net reconstructed broadband signal with expandable depth based on deep learning is designed;
and initializing the parameters of the ADMM-net model into corresponding parameter values in the ADMM algorithm.
In the model training stage, inputting signal data of a training set into the ADMM-net, and continuously minimizing a loss function through an optimization algorithm to obtain an optimal neural network parameter, wherein the step of obtaining the optimal neural network parameter comprises the following steps:
in the model training phase, the frequency domain representation of the sampling sequence of the training set is used
Inputting the initialized ADMM-net to obtain an initially reconstructed frequency domain signal X ir [k]；
Constructing the initially reconstructed frequency domain signal X ir [k]Frequency domain signal X [ k ] with input broadband signal]A loss function E (θ) in between, where θ is the relevant neural network parameter;
and continuously minimizing the loss function E (theta) by using an optimization algorithm to obtain an optimal neural network parameter theta, thereby obtaining the trained ADMM-net model.
In the test phase or the application phase, inputting the signal data of the test set into the trained ADMM-net model to obtain a reconstructed broadband signal, wherein the step of obtaining the reconstructed broadband signal comprises:
representing the frequency domain of the sample sequence of the test set during the test phase or the application phase
Inputting the trained ADMM-net model for processing to obtain a reconstructed frequency domain signal X r [k]。
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above may be implemented by hardware instructions of a computer program, which may be stored in a non-volatile computer-readable storage medium, and when executed, may include the processes of the embodiments of the methods described above. Any reference to memory, storage, databases, or other media used in embodiments provided herein may include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
In summary, the present invention discloses a method, an apparatus, an intelligent terminal and a storage medium for processing a compressed sensing of a broadband signal based on deep learning, wherein the method comprises: acquiring an input broadband signal, and performing multi-fold sampling on the input broadband signal on a time domain to obtain a discrete sampling sequence; preprocessing the obtained discrete sampling sequence and the input broadband signal and constructing a data set; designing a deep learning-based broadband signal reconstruction neural network model ADMM-net, and initializing the ADMM-net; in the model training stage, inputting signal data of a training set into the ADMM-net, and continuously minimizing a loss function through an optimization algorithm to obtain an optimal neural network parameter; and in a testing stage or an application stage, inputting the signal data of the test set into the trained ADMM-net model to obtain a reconstructed broadband signal. The numerical analysis of the experimental result shows that compared with the common reconstruction algorithm, the ADMM-net combined with deep learning in the invention has higher reconstruction accuracy under the conditions of lower sampling rate and lower signal-to-noise ratio after being trained by a network model, and has certain generalization capability.
The invention also has the following advantages:
1) and combining the idea of deep learning, firstly carrying out frequency domain reconstruction on the broadband signal by using the depth-extensible alternative direction multiplier network, and the result proves that the reconstruction performance of the broadband signal can be improved.
2) And training neural network parameters of the alternative direction multiplier network by constructing a loss function between a network reconstruction result and a true value and using an Armijo criterion as a conjugate gradient method of a gradient descent direction to continuously minimize the loss function, so that the reconstruction performance of the broadband signal and the generalization of a network model are improved.
3) And the alternative direction multiplier network based on multiple-time set sampling and deep learning is used as a compressed sensing frame of the broadband signal, so that the compressed sensing performance of the broadband signal is improved under the conditions of a lower sampling rate and a lower signal-to-noise ratio.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (9)
1. A method for processing compressed sensing of a broadband signal based on deep learning is characterized in that the method comprises the following steps:
acquiring an input broadband signal, and performing multi-fold sampling on the input broadband signal on a time domain to obtain a discrete sampling sequence;
preprocessing the obtained discrete sampling sequence and an input broadband signal and constructing a data set, wherein the data set comprises: signal data of the training set and signal data of the test set;
designing a deep learning-based broadband signal reconstruction neural network model ADMM-net, and initializing the ADMM-net;
in the model training stage, inputting signal data of a training set into the ADMM-net, and continuously minimizing a loss function through an optimization algorithm to obtain an optimal neural network parameter;
and in a testing stage or an application stage, inputting the signal data of the test set into the trained ADMM-net model to obtain a reconstructed broadband signal.
2. The method according to claim 1, wherein the step of obtaining the input wideband signal and performing multiple-set sampling on the input wideband signal in the time domain to obtain the discrete sampling sequence further comprises:
obtaining input wideband signal x (t), determining sampling mode by sequence forward selection algorithm
Using multiple sets of samples at
Sampling at a moment to obtain a discrete sampling sequence x ci [n]Where T denotes the nyquist sampling time period interval of the input broadband signal, the multiple-sampling period interval is L times as large as the nyquist sampling theorem, so that the sampling frequency is reduced to 1/L of the nyquist sampling frequency.
3. The method according to claim 1, wherein the step of preprocessing the obtained discrete sampling sequence and the input wideband signal and constructing the data set comprises:
for the obtained discrete sampling sequence x ci [n]Performing a discrete Fourier transform to obtain a frequency domain representation X of the sample sequence ci [k c ]；
At the same time, a frequency domain representation of the resulting sample sequence is obtained
And corresponding frequency domain signal X [ k ] of input broadband signal]And forming a data set, wherein the number of the training sets and the number of the testing sets are in a set proportion.
4. The method for processing compressed sensing of a deep learning based broadband signal according to claim 1, wherein the step of designing a deep learning based signal reconstruction neural network model ADMM-net and initializing the ADMM-net comprises:
designing a depth-learning-based depth-extensible signal reconstruction neural network model ADMM-net by using an ADMM algorithm as a signal reconstruction algorithm of compressed sensing;
the parameters of the ADMM-net model are initialized to the corresponding parameter values in the ADMM algorithm.
5. The method for processing compressed sensing of broadband signals based on deep learning of claim 1, wherein the step of inputting signal data of a training set into the ADMM-net in a model training phase, and continuously minimizing a loss function through an optimization algorithm to obtain optimal neural network parameters comprises:
in the model training phase, the frequency domain representation of the sampling sequence of the training set is used
Inputting the initialized ADMM-net to obtain an initially reconstructed frequency domain signal X ir [k]；
Constructing the initially reconstructed frequency domain signal X ir [k]Frequency domain signal X [ k ] of input broadband signal]A loss function E (θ) in between, where θ is the relevant neural network parameter;
and continuously minimizing the loss function E (theta) by using an optimization algorithm to obtain an optimal neural network parameter theta, so as to obtain the trained ADMM-net model.
6. The method according to claim 1, wherein the step of inputting the signal data of the test set into the trained ADMM-net model during the test phase or the application phase to obtain the reconstructed wideband signal comprises:
representing the frequency domain of the sample sequence of the test set during the test phase or the application phase
Inputting the trained ADMM-net model for processing to obtain a reconstructed frequency domain signal X r [k]。
7. An apparatus for compressed sensing processing of a wideband signal based on deep learning, the apparatus comprising:
the multiple set sampling processing module is used for acquiring an input broadband signal and performing multiple set sampling on the input broadband signal in a time domain to obtain a discrete sampling sequence;
the preprocessing module is used for preprocessing the obtained discrete sampling sequence and the input broadband signal and constructing a data set, wherein the data set comprises: signal data of the training set and signal data of the test set;
the device comprises a setting and initializing module, a processing module and a processing module, wherein the setting and initializing module is used for designing a signal reconstruction neural network model ADMM-net based on deep learning and initializing the ADMM-net;
the neural network training module is used for inputting signal data of a training set into the ADMM-net in a model training stage, and continuously minimizing a loss function through an optimization algorithm to obtain an optimal neural network parameter;
and the testing and application module is used for inputting the signal data of the test set into the trained ADMM-net model in a testing stage or an application stage to obtain a reconstructed broadband signal.
8. An intelligent terminal comprising a memory, and one or more programs, wherein the one or more programs are stored in the memory, and wherein the one or more programs being configured to be executed by the one or more processors comprises instructions for performing the method of any of claims 1-6.
9. A non-transitory computer-readable storage medium, wherein instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the method of any of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210450842.1A CN114978381B (en) | 2022-04-27 | 2022-04-27 | Compressed sensing processing method and device for broadband signal based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210450842.1A CN114978381B (en) | 2022-04-27 | 2022-04-27 | Compressed sensing processing method and device for broadband signal based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114978381A true CN114978381A (en) | 2022-08-30 |
| CN114978381B CN114978381B (en) | 2023-06-23 |
Family
ID=82978551
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210450842.1A Active CN114978381B (en) | 2022-04-27 | 2022-04-27 | Compressed sensing processing method and device for broadband signal based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114978381B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150287223A1 (en) * | 2014-04-04 | 2015-10-08 | The Board Of Trustees Of The University Of Illinois | Highly accelerated imaging and image reconstruction using adaptive sparsifying transforms |
| US20200074244A1 (en) * | 2018-08-29 | 2020-03-05 | Université De Genève | Signal sampling with joint training of learnable priors for sampling operator and decoder |
| CN111224671A (en) * | 2020-01-15 | 2020-06-02 | 高跃 | Signal processing apparatus |
| CN111680666A (en) * | 2020-06-30 | 2020-09-18 | 西安电子科技大学 | Under-sampling frequency hopping communication signal deep learning recovery method |
| CN112615801A (en) * | 2020-12-16 | 2021-04-06 | 西安交通大学 | Channel estimation method, medium, and apparatus based on compressed sensing and deep learning |
-
2022
- 2022-04-27 CN CN202210450842.1A patent/CN114978381B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150287223A1 (en) * | 2014-04-04 | 2015-10-08 | The Board Of Trustees Of The University Of Illinois | Highly accelerated imaging and image reconstruction using adaptive sparsifying transforms |
| US20200074244A1 (en) * | 2018-08-29 | 2020-03-05 | Université De Genève | Signal sampling with joint training of learnable priors for sampling operator and decoder |
| CN111224671A (en) * | 2020-01-15 | 2020-06-02 | 高跃 | Signal processing apparatus |
| CN111680666A (en) * | 2020-06-30 | 2020-09-18 | 西安电子科技大学 | Under-sampling frequency hopping communication signal deep learning recovery method |
| CN112615801A (en) * | 2020-12-16 | 2021-04-06 | 西安交通大学 | Channel estimation method, medium, and apparatus based on compressed sensing and deep learning |
Non-Patent Citations (4)
| Title |
|---|
| 苏汉宁: "《基于深度学习的LFM信号参数估计算法研究》", 《硕士学位论文》 * |
| 苏汉宁: "《基于深度学习的LFM信号参数估计算法研究》", 《硕士学位论文》, 1 December 2020 (2020-12-01), pages 44 - 66 * |
| 董宁斐: "《稀疏宽带信号压缩采样与重构技术研究》", 《博士学位论文》 * |
| 董宁斐: "《稀疏宽带信号压缩采样与重构技术研究》", 《博士学位论文》, 15 July 2019 (2019-07-15), pages 60 - 78 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114978381B (en) | 2023-06-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111680666B (en) | Under-sampling frequency hopping communication signal deep learning recovery method | |
| CN107192878A (en) | A kind of trend of harmonic detection method of power and device based on compressed sensing | |
| CN112967357B (en) | Spectrum map construction method based on convolutional neural network | |
| CN105610525B (en) | A kind of broadband compression frequency spectrum sensing method based on sub-band match selection | |
| Zhao et al. | Adaptive gradient-based block compressive sensing with sparsity for noisy images | |
| CN111030954A (en) | Multichannel sampling broadband power amplifier predistortion method based on compressed sensing | |
| He et al. | Maximum correntropy adaptation approach for robust compressive sensing reconstruction | |
| CN112688746A (en) | Spectrum prediction method based on space-time data | |
| CN114978381A (en) | Deep learning-based compressed sensing processing method and device for broadband signals | |
| CN113222812A (en) | Image reconstruction method based on information flow reinforced deep expansion network | |
| CN110954860A (en) | DOA and polarization parameter estimation method | |
| Wang et al. | Parallel non-Cartesian spatial-temporal dictionary learning neural networks (stDLNN) for accelerating 4D-MRI | |
| CN111478706B (en) | Compressed sensing-oriented sparse multi-band signal blind reconstruction method | |
| Wang et al. | Neural network based point spread function deconvolution for astronomical applications | |
| Srinivas et al. | Performance comparison of measurement matrices in compressive sensing | |
| CN113095215B (en) | Solar radio filtering method and system based on improved LSTM network | |
| Zhao et al. | Sparse multiband signal spectrum sensing with asynchronous coprime sampling | |
| Liu et al. | SRN-SZ: Deep Leaning-Based Scientific Error-bounded Lossy Compression with Super-resolution Neural Networks | |
| CN114244458A (en) | Total-blind spectrum sensing method of sub-Nyquist sampling front end | |
| Lyons et al. | A deep compound Gaussian regularized unfolded imaging network | |
| Onose et al. | A randomised primal-dual algorithm for distributed radio-interferometric imaging | |
| Rateb et al. | A generic top-level mechanism for accelerating signal recovery in compressed sensing | |
| CN108092668A (en) | A kind of linear FM signal restructing algorithm based on minimum sparse ruler sampling | |
| Wang et al. | WDU-Net: Wavelet-Guided Deep Unfolding Network for Image Compressed Sensing Reconstruction | |
| CN117195650B (en) | FDTD calculation method and system based on high-order matrix index perfect matching layer |
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 |