CN104950282B - Sparse reconstruct is realized in continuous domain broadband signal super-resolution direction-finding method and device - Google Patents

Abstract

The invention relates to a broadband signal super-resolution direction finding method and device realized by sparse reconstruction in a continuous domain, and the invention belongs to the field of signal estimation and signal processing. It solves the problems of low accuracy and slow implementation speed of wideband signal super-resolution direction finding estimation realized by the existing sparse reconstruction method in the discrete domain. The sampling frequency point G of the present invention is divided into W groups, and each group contains G/W frequency points. According to the broadband signal super-resolution direction finding method realized by sparse reconstruction in the continuous domain, the estimated value of the signal arrival direction is solved. First, the G in each group The estimated values corresponding to /W frequency points are averaged as the estimated value of each group, and then the estimated values of W groups are averaged as the final estimated result. Each digital signal processor processes a set of observation data, and then averages the processed results to obtain the final result. The invention is suitable for super-resolution direction finding of broadband signals realized by sparse reconstruction in continuous domain.

Description

Method and device for wideband signal super-resolution direction finding realized by sparse reconstruction in continuous domain

technical field

The invention belongs to the field of signal estimation and signal processing.

Background technique

Sparse reconstruction-like direction finding method is a novel super-resolution direction finding method emerging in recent years. Sparse reconstruction belongs to the category of compressed sensing, and has had an important impact in many applied science fields, such as remote sensing imaging, coding, face recognition, ultra-wideband signal processing, etc. Candes proposed the Restricted Isometry Property (RIP) in 2006, which mathematically proved the conditions for the establishment of the compressed sensing method, and this criterion has also become a criterion that must be followed by the sparse reconstruction theory. Blacquiere, G et al. discussed the joint sparsity of wideband array signals, but did not give a better super-resolution direction finding method using joint sparsity. The earliest sparse reconstruction method was only used to estimate the direction of narrowband signals. For wideband signals, the most noteworthy feature of wideband array signal processing based on frequency division is the joint sparsity between different frequency bands. Hyder et al. took advantage of this property , combined with the l ₁ norm to estimate the direction of arrival of the signal, it is verified that it can estimate the broadband and narrowband signals with strong correlation and small intervals in the case of small snapshots. Zakharov proposed a solution to the basis tracking problem based on the idea of homotopy. This method obtains the optimal warping factor and reconstruction results through iteration. It is very suitable for scenarios where the amount of data is small and the error limit cannot be determined. It is connected with broadband signals. The joint sparsity of the method is extended to wideband signal direction finding at a single frequency-domain snapshot.

In 2011, Liu Zhangmeng and Zhou Yu et al. used the steering vector composed of the lower left corner elements of the covariance matrix to complete super-resolution DOA (Direction of arrival) estimation under sparse constraints, and used the sparsity of multiple subband signals to perform Reconstruction, which is a generalization of the standard sparse Bayesian learning algorithm for sparse reconstruction, can be applied to signal models with multiple redundant dictionaries.

In 2010, Eldar proposed the concept of block sparseness. On this basis, Hu Nan referred to the prolate ellipsoidal wave function model to extract the elements in the output matrix of the broadband signal array to form a vector. The representation method has problems such as high complexity and tedious calculation. At the same time, the orthogonal matching pursuit method is used to reconstruct the signal, but the calculation time of the tracking process is still long. In addition, when the sparse reconstruction method is used for super-resolution direction finding, there is often a grid mismatch phenomenon. Tan and A. Nehorai et al. and some other scholars have studied the grid mismatch problem in compressive sensing and proposed some The parameter estimation algorithm of iterative grid optimization compressive sensing estimates the original signal and the grid mismatch value at the same time; Candes studies the signal reconstruction in the continuous domain and solves the grid mismatch in the signal reconstruction process in the discrete domain. This is a further extension of the application of compressed sensing from the discrete domain to the continuous domain, but they did not give the realization of sparse reconstruction in the continuous domain in super-resolution direction finding of broadband signals method. Based on the sparse spatial sampling of the signal, He uses the broadband coefficient spectrum fitting method to expand the array degrees of freedom and is suitable for non-uniform noise backgrounds.

When estimating broadband signals, the sparse reconstruction method reduces the requirements on the signal-to-noise ratio and the number of sampling snapshots compared with the previous focusing methods, and greatly improves the estimation performance. However, it has two main shortcomings. : First, the algorithm first divides the search interval into multiple discrete angle grids. This process usually causes grid mismatch, which makes the estimation results have certain errors; second, the sparse reconstruction method is cumbersome. The calculation process leads to a long estimation time, especially when dealing with broadband signals, even exceeding the broadband focusing methods, which bring great difficulties to the application of sparse reconstruction methods in practical systems.

Contents of the invention

The invention aims to solve the problems of low accuracy and slow implementation speed of wideband signal super-resolution direction finding estimation realized by the existing sparse reconstruction method in the discrete domain. A method and device for super-resolution direction finding of broadband signals realized by sparse reconstruction in continuous domain are proposed.

The broadband signal super-resolution direction finding method realized by sparse reconstruction in the continuous domain of the present invention, the specific steps of the method are:

Step 1. Use a broadband uniform linear array composed of M broadband omnidirectional array elements to receive N far-field broadband signals s _n (t), and obtain an output signal y(t) of the antenna array; n=1,2,... , N, where M and N are both positive integers;

Step 2. Carry out weighted summation to N far-field broadband signals s _n (t), obtain the far-field broadband signal weighted sum s(t), and perform Fourier transform on the far-field broadband signal weighted sum s(t) to obtain frequency domain signal x(f _k );

Step 3, perform Fourier transform on the array output signal y(t) obtained in step 1, and obtain the frequency domain signal Y(f _k ), the _sn (t) Fourier transformed signal of N far-field broadband signals under frequency f _k S(f _k ) and the direction matrix A(f _k ) at frequency f _k ;

Step 4. Reconstruct and optimize the signal _S (f _k ) in the continuous domain, and use the convex optimization method to _solve the Lagrange multiplier _matrix V( f _k ) optimal solution V _o (f _k );

Step five, calculate the inverse Fourier transform of the vector V _o (f _k ), convert the vector V _o (f _k ) in the discrete domain into the variable s _o (t) in the continuous time domain;

Step 6. Let ψ _detect ＝{τ _detect [1],τ _detect [2],...,τ _detect [n],...,τ _detect [N]} be N signals arriving at two adjacent arrays The support set of inter-element delay, according to the direction matrix A (f _k ) construction time delay under the frequency f _k described in step 3, is the direction matrix A _detect (f _k ) of τ _detect [n];

Step seven, obtain the adjoint matrix A ^* _detect (f _k ) of A _detect (f _k ) according to A _detect (f _k ) described in step six, obtain τ _detect [1], τ _detect [2] in ψ _detect , ...,τ _detect [n],...,τ _detect [N] value, so as to obtain the value of the signal arrival direction θ ₁ , θ ₂ ,...,θ _N at the frequency point f _k ;

Step 8. Divide the G frequency points into W groups, G and W are both positive integers, and W is less than or equal to G, and use W processors to simultaneously obtain the signal of each frequency point through the methods described in steps 1 to 7 The direction of arrival of the signal of G frequency points is averaged to obtain the direction of the broadband signal.

A direction-finding device using the broadband signal super-resolution direction-finding method realized by sparse reconstruction in the continuous domain, the device includes a wide-band uniform linear array, a multi-channel wide-band digital receiver and a broadband signal super-resolution direction-finding device;

The wideband uniform linear array is used to receive far-field wideband signals, and send the received far-field wideband signals to multi-channel wideband digital receivers;

The multi-channel broadband digital receiver receives the far-field broadband signal sent by the broadband uniform linear array, and extracts the quadrature component and the in-phase component of the far-field broadband signal; and sends the extracted quadrature component and the in-phase component to the broadband signal super-resolution measurement to the device;

The broadband signal super-resolution direction finding device includes P digital signal processor chips, a crystal oscillator, a reset circuit, a static memory, a complex programmable logic device, a flash memory, and a read-only memory; the P digital signal processor chips are divided into one A master digital signal processor chip and P-1 slave digital signal processor chips;

The main digital signal processor chip and P-1 slave digital signal processor chips are cascaded connected through the serial input and output ports, and the data, address and logic control signal input and output terminals of the complex programmable logic device are simultaneously connected with the P digital signal The data, address and logic control signal input and output terminals of the processor chip are connected, the storage signal input and output terminals of the main digital signal processor chip are respectively connected to the storage signal input and output terminals of the static memory and the flash memory 5, and the clock signal output terminals of the crystal oscillator are connected to The clock signal input terminal of the main digital signal processor chip and the reset signal output terminal of the reset circuit are connected to the reset signal input terminal of the main digital signal processor chip, and the storage signal input terminal of the complex programmable logic device is connected to the storage signal output terminal of the read-only memory end.

The direction-finding method and device described in the present invention utilizes the characteristics of high estimation accuracy and strong resolution of the sparse reconstruction in the continuous domain, and combines the fast calculation speed of the digital signal processor chip (Digital Signal Processor, DSP). , parallel processing and high precision, it can realize fast and accurate estimation of the direction of arrival of broadband signals in a relatively short period of time. The present invention can be applied in the background of strong real-time performance and high precision requirements, provides good technical means for many engineering fields, can be widely used in radar, sonar, missile-borne systems and other radio detection systems, and has excellent Application prospect and value.

Description of drawings

Fig. 1 is a flowchart of the method of the present invention;

Fig. 2 is the antenna array composed of M omnidirectional array elements described in the first embodiment to receive N far-field broadband signal models;

Fig. 3 is a schematic block diagram of the direction finding device obtained by the broadband signal super-resolution direction finding method realized by the sparse reconstruction in the continuous domain described in the eighth embodiment;

Fig. 4 is a block diagram of the electrical principle of the broadband signal super-resolution direction finding device described in the eighth specific embodiment.

detailed description

Specific embodiments 1. This embodiment is described in conjunction with FIG. 1 and FIG. 2. The broadband signal super-resolution direction finding method realized by sparse reconstruction in the continuous domain described in this embodiment, the specific steps of the method are:

Step 1. Use a broadband uniform linear array composed of M broadband omnidirectional array elements to receive N far-field broadband signals s _n (t), and obtain an output signal y(t) of the antenna array; n=1,2,... , N, where M and N are both positive integers;

Step 2. Carry out weighted summation to N far-field broadband signals s _n (t), obtain the far-field broadband signal weighted sum s(t), and perform Fourier transform on the far-field broadband signal weighted sum s(t) to obtain frequency domain signal x(f _k );

Step 3, perform Fourier transform on the array output signal y(t) obtained in step 1, and obtain the frequency domain signal Y(f _k ), the _sn (t) Fourier transformed signal of N far-field broadband signals under frequency f _k S(f _k ) and the direction matrix A(f _k ) at frequency f _k ;

Step 4. Reconstruct and optimize the signal _S (f _k ) in the continuous domain, and use the convex optimization method to _solve the Lagrange multiplier _matrix V( f _k ) optimal solution V _o (f _k );

Step five, calculate the inverse Fourier transform of the vector V _o (f _k ), convert the vector V _o (f _k ) in the discrete domain into the variable s _o (t) in the continuous time domain;

Step 6. Let ψ _detect ＝{τ _detect [1],τ _detect [2],...,τ _detect [n],...,τ _detect [N]} be N signals arriving at two adjacent arrays The support set of inter-element delay, according to the direction matrix A (f _k ) construction time delay under the frequency f _k described in step 3, is the direction matrix A _detect (f _k ) of τ _detect [n];

Step seven, obtain the adjoint matrix A ^* _detect (f _k ) of A _detect (f _k ) according to A _detect (f _k ) described in step six, obtain τ _detect [1], τ _detect [2] in ψ _detect , ...,τ _detect [n],...,τ _detect [N] value, so as to obtain the value of the signal arrival direction θ ₁ , θ ₂ ,...,θ _N at the frequency point f _k ;

Step 8. Divide the G frequency points into W groups, G and W are both positive integers, and W is less than or equal to G, and use W processors to simultaneously obtain the signal of each frequency point through the methods described in steps 1 to 7 The direction of arrival of the signal of G frequency points is averaged to obtain the direction of the broadband signal.

In the actual hardware implementation, the sampling frequency G of the present invention is divided into W groups, each group contains G/W frequency points, and the estimated value of the direction of arrival of the signal is calculated according to the above method. The estimated values corresponding to the G/W frequency points in each group may be firstly averaged as the estimated value of each group, and then the estimated values of the W groups may be averaged again as the final estimated result. Here, W pieces of digital signal processors can be used to process the W sets of observation data at the same time. Each piece of DSP can process a set of observation data, and then average the processed results to obtain the final result.

Specific embodiment 2. This embodiment is a further description of the broadband signal super-resolution direction finding method realized by sparse reconstruction in the continuous domain described in specific embodiment 1. In this embodiment: step 1 obtains the output signal y( The expression of t), with one array element as the phase reference point:

Where y _m (t) is the signal vector output by the mth array element, b _m (t) is the Gaussian white noise received by the mth array element; m=1,2,...,M, the nth array element The delay between the arrival of the signal between two adjacent array elements is:

c is the electromagnetic wave propagation speed, and d is the spacing of M broadband omnidirectional array elements.

Specific embodiment three. This embodiment is a further description of the broadband signal super-resolution direction finding method implemented by sparse reconstruction in the continuous domain described in specific embodiment one. In this embodiment: the far-field broadband signal weighting described in step two and s(t) and its frequency-domain signal x(f _k ) are:

Among them, f _k is the frequency independent variable of s(t) after Fourier transform, G is the number of frequency points, k represents the k-th frequency point, and k and G are both positive integers. The far-field broadband signal weighted sum s(t) and its frequency domain signal x(f _k ) described in step 2 are:

Among them, f _k is the frequency independent variable of s(t) after Fourier transform, k represents the k-th frequency point, and k is a positive integer.

Embodiment 4. This embodiment is a further description of the wideband signal super-resolution direction finding method implemented by sparse reconstruction in the continuous domain described in Embodiment 1. In this embodiment, the frequency domain signal Y described in step 3 is obtained (f _k ), the Fourier transformed signal S(f _k ) at frequency f _k and the method of direction matrix A(f _k ) at frequency f _k are:

Y(f _k )=A(f _k )S(f _k )+B(f _k )k=1,2,...,G (5)

In the formula, B(f _k ) is the noise at frequency f _k ;

Among them, a(f _k , θ _n ) is the direction vector of the broadband array to the broadband signal with the incident direction as θ _n at the frequency point f _k .

Specific Embodiment 5. This embodiment is a further description of the wideband signal super-resolution direction finding method implemented by sparse reconstruction in the continuous domain described in the specific embodiment 1. In this embodiment: step 4 in the continuous domain. S(f _k ) reconstruction optimization, and use the convex optimization method to solve the optimal solution of the Lagrange multiplier matrix V(f _k ) that makes Y(f _k )=A(f _k )S(f _k ) The method of V _o (f _k ) is:

By formula:

Optimizing the reconstruction of the signal S(f _k ) in the continuous domain, where σ ² is the noise power on each array element;

Restrictions:

||Y(f _k )-A(f _k )S(f _k )|| ₂ ≤ε ₁

ε ₁ is the threshold value, transform formula (7) into a positive semi-definite problem, and obtain the formula:

Restrictions:

Among them, A ^* (f _k ) is the adjoint matrix of A(f _k ); Simplify formula (8):

Restrictions:

in,

H(f _k ) is an intermediate variable used to solve the optimization problem, and the position of the elements in the matrix is determined by p, q;

I(f _k ) is the identity matrix; V ^* (f _k ) is the adjoint matrix of V(f _k ), through this step, the optimal solution V _o (f _k ) of V(f _k ) is obtained.

Specific Embodiment 6. This embodiment is a further description of the wideband signal super-resolution direction finding method implemented by sparse reconstruction in the continuous domain described in Embodiment 1. In this embodiment: Step 6 is based on Step 3. Direction matrix A(f _k ) at frequency f _k The method to obtain direction matrix A _detect (f _k ) with time delay τ _detect [n] is:

Bring the support set ψ _detect into formula (6) to obtain A _detect (f _k ):

Embodiment 7. This embodiment is a further description of the wideband signal super-resolution direction finding method implemented by sparse reconstruction in the continuous domain described in Embodiment 1. In Step 7, τ _detect [1] in ψ _detect is obtained. The method of τ _detect [2],...,τ _detect [n],...,τ _detect [N] is: calculate the inverse Fourier transform of the vector V _o (f _k ), and convert the vector V _o ( f _k ) is transformed into a variable s _o (t) in the continuous time domain;

A ^* (f _k )V _o (f _k )＝sgn(s _o (t)) (11)

Among them, s _o (t) is the weighted sum of signal s _n (t) after fitting, s _o (t)≠0;

Take the absolute value of both ends of formula (11), and combine the information of all frequency points:

Put A _detect (f _k ), V _o (f _k ) and G into formula (12) to obtain τ _detect [1],τ _detect [2],...,τ _detect [n],... , the value of τ _detect [N]; then according to the relationship between τ[n] and θ _n in formula (2), the signal arrival angle θ ₁ , θ ₂ ,...,θ _N can be calculated.

Embodiment 8. This embodiment is described in conjunction with FIG. 3 and FIG. 4. This embodiment is a direction-finding device obtained by a broadband signal super-resolution direction-finding method realized by sparse reconstruction in a continuous domain. The device includes a wide-band uniform linear array 1, multiple Channel broadband digital receiver 2 and broadband signal super-resolution direction finding device 3;

The wideband uniform linear array 1 is used to receive the far-field wideband signal, and send the received far-field wideband signal to the multi-channel wideband digital receiver 2;

The multi-channel broadband digital receiver 2 receives the far-field broadband signal sent by the broadband uniform linear array 1, and extracts the quadrature component and the in-phase component of the far-field broadband signal; and sends the extracted quadrature component and the in-phase component to the broadband signal super Resolution direction finding device 3;

The broadband signal super-resolution direction finding device 3 includes P digital signal processor chips, a crystal oscillator 3-1, a reset circuit 3-2, a static memory 3-3, a complex programmable logic device 3-4, a flash memory 3-5, a Read memory 3-6; the P digital signal processor chips are divided into 1 master digital signal processor chip 3-7 and P-1 slave digital signal processor chips 3-8;

The main digital signal processor chip 3-7 and P-1 slave digital signal processor chips 3-8 are cascaded connected through the serial input and output Serial Rapid Input Output, SRIO port, and the data of the complex programmable logic device 3-4 , address and logic control signal input and output terminals are connected to the data, address and logic control signal input and output terminals of P digital signal processor chips at the same time, and the storage signal input and output terminals of the main digital signal processor chips 3-7 are respectively connected to the static memory 3-3 and the storage signal input and output terminals of the flash memory 3-5, the clock signal output terminal of the crystal oscillator 3-1 is connected to the clock signal input terminal of the main digital signal processor chip 3-7, and the reset signal output of the reset circuit 3-2 terminal is connected to the reset signal input terminal of the main digital signal processor chip 3-7, and the storage signal input terminal of the complex programmable logic device 3-4 is connected to the storage signal output terminal of the read-only memory 3-6.

The digital signal processor described in the present embodiment adopts TMS320C6678 of Texas Instruments (TI) company, adopts 6 slices of processors to process the above method in parallel, 6 slices of DSPs are connected by SRIO ports, after power-on, the read-only memory first loads the program to The complex programmable logic device, the flash memory loads the program of these 6 digital signal processor chips to the main digital signal processor chip, and the main digital signal processor chip then sequentially passes the program through the SRIO port one by one from the digital signal processing The program of the processor chip is passed to them, and then the main digital signal processor chip starts to receive the observation data of J frequency points transmitted by the multi-channel wideband digital receiver 2, and divides them into W groups, assuming J=30, W=6 , then each digital signal processor chip can process the observed data of U=30/6=5 frequency points, and the main DSP chip 3-7 passes other observed data that are responsible for processing from the digital signal processor chip to them through the SRIO port , and then each digital signal processor chip solves according to the steps of the above theoretical derivation, and then the 5 slave digital signal processor chips transmit the estimated values at 5 frequency points to the main digital signal processor chip through the SRIO port, and the main digital signal processor chip The digital signal processor chip averages the estimated values under these 30 frequency points to obtain the final result. Among them, JTAG3-11 is responsible for debugging the digital signal processor chip, the power supply is responsible for the overall power supply, the crystal oscillator is responsible for providing the clock, and the reset is responsible for providing the reset signal.

Specific Embodiment 9. This embodiment is a direction-finding device obtained by the wideband signal super-resolution direction-finding method realized by sparse reconstruction in the continuous domain described in Embodiment 8. The difference is that the main digital signal processor chip 3-7 and P-1 slave digital signal processor chips 3-8 are connected through a shared bus tight coupling instead of master digital signal processor chips 3-7 and P-1 slave digital signal processor chips 3-8. cascade connection.

The digital signal processor chip described in this embodiment adopts 6 pieces of ADSP-TS201S of Texas Instruments, which are connected through a shared bus tight coupling mode. After power-on, the read-only memory first loads the program to the complex programmable logic device to configure the DSP. Afterwards, the flash memory loads the program to the DSP chip, and the main digital signal processor chip starts to receive the observation data of G frequency points transmitted by the multi-channel broadband digital receiver, and divides them into W groups, assuming G=30, W =6, then each digital signal processor chip can process the observed data of U=30/6=5 frequency points, and the main digital signal processor chip transmits other observed data that are responsible for processing from the digital signal processor chip to the From the digital signal processor chip, each digital signal processor chip then solves according to the steps of the above theoretical derivation, and then the 5 slave digital signal processor chips transmit the estimated values at 5 frequency points to the main digital signal through the bus. Signal processor chip, the main digital signal processor chip averages the estimated values under these 30 frequency points to get the final result. Among them, the power supply is responsible for the overall power supply, the crystal oscillator is responsible for providing the clock, and the reset is responsible for providing the reset signal.

Specific Embodiment 10. This embodiment is a direction finding device obtained by the wideband signal super-resolution direction finding method realized by sparse reconstruction in the continuous domain described in Embodiment 8. The difference is that the main digital signal processor chip 3-7 and P-1 slave digital signal processor chips 3-8 are connected by link port cascading loose coupling to replace the main digital signal processor chip 3-7 and P-1 slave digital signal processor chips 3-8 The cascade connection method adopted between them.

The DSP chip described in this embodiment adopts the ADSP-TS201S of Analog Devices Instruments (ADI) company, and adopts 6 processors to process the above methods in parallel, and the DSP chip is connected in a cascaded loose coupling mode through the link port. The program is loaded to the complex programmable logic device, the flash memory loads the program to the main digital signal processor chip, and the main digital signal processor chip sequentially transmits the programs of other slave digital signal processor chips through the link port level by level. Give the slave digital signal processor chip, and then the main digital signal processor chip starts to receive the observation data of G frequency points transmitted by the multi-channel wideband digital receiver, divide them into W groups, suppose G=30, W=6, Then each digital signal processor chip can process the observation data of U=30/6=5 frequency points, and the main digital signal processor chip passes through the link port the observation data that other digital signal processor chips are responsible for processing. The first level is passed to the slave digital signal processor chip one by one, and then each digital signal processor chip is solved according to the steps of the above theoretical derivation, and then the five slave digital signal processor chips will each estimate the value under the five frequency points Through the link port, it is uploaded to the main digital signal processor chip step by step, and the main digital signal processor chip averages the estimated values under 30 frequency points to obtain the final result. The power supply is responsible for the overall power supply, the crystal oscillator is responsible for providing the clock, and the reset is responsible for providing the reset signal.

Claims (4)

1. A broadband signal super-resolution direction finding method realized by sparse reconstruction in a continuous domain, characterized in that the specific steps of the method are: Step 1. Use a broadband uniform linear array composed of M broadband omnidirectional array elements to receive N far-field broadband signals s _n (t), and obtain an output signal y(t) of the antenna array; n=1,2,... , N, where M and N are both positive integers; Step 2. Carry out weighted summation to N far-field broadband signals s _n (t), obtain the far-field broadband signal weighted sum s(t), and perform Fourier transform on the far-field broadband signal weighted sum s(t) to obtain frequency domain signal x(f _k ); Step 3, perform Fourier transform on the array output signal y(t) obtained in step 1, and obtain the frequency domain signal Y(f _k ) and N far-field broadband signals s _n (t) at the frequency point f _k after Fourier transform The signal S(f _k ) and the direction matrix A(f _k ) at the frequency point f _k ; Step 4. Reconstruct and optimize the signal _S (f _k ) in the continuous domain, and use the convex optimization method to _solve the Lagrange multiplier _matrix V(f _k ) optimal solution V _o (f _k ); Step 5. Calculate the inverse Fourier transform of the optimal solution V _o (f _k ), and convert the optimal solution V _o (f _k ) in the discrete domain into a variable s _o (t) in the continuous time domain; Step 6. Let ψ _detect ＝{τ _detect [1],τ _detect [2],...,τ _detect [n],...,τ _detect [N]} be N signals arriving at two adjacent arrays The support set of inter-element delay, according to the direction matrix A (f _k ) construction time delay under the frequency point f _k described in step 3, is the direction matrix A _detect (f _k ) of τ _detect [n]; Step seven, obtain the adjoint matrix A ^* _detect (f _k ) of A _detect (f _k ) according to A _detect (f _k ) described in step six, obtain τ _detect [1], τ _detect [2] in ψ _detect , ...,τ _detect [n],...,τ _detect [N] value, so as to obtain the value of the signal arrival direction θ ₁ , θ ₂ ,...,θ _N at the frequency point f _k ; Step 8: Divide the G frequency points into W groups, G and W are both positive integers, and W is less than or equal to G, and use W processors to simultaneously calculate the arrival direction of the signal at each frequency point through steps 1 to 7 , take the average value of the signal arrival directions of G frequency points to obtain the direction of the broadband signal. 2. the direction-finding device obtained by the broadband signal super-resolution direction-finding method that sparse reconstruction realizes in the continuous domain according to claim 1, is characterized in that, the device comprises a wide-band uniform linear array (1), a multi-channel wide-band digital receiver (2) and broadband signal super-resolution direction finding device (3); The wideband uniform linear array (1) is used to receive far-field wideband signals, and send the received far-field wideband signals to multi-channel wideband digital receivers (2); The multi-channel broadband digital receiver (2) receives the far-field broadband signal sent by the broadband uniform linear array (1), and extracts the quadrature component and the in-phase component of the far-field broadband signal; and sends the extracted quadrature component and the in-phase component To the broadband signal super-resolution direction finding device (3); The broadband signal super-resolution direction finding device (3) includes P digital signal processor chips, a crystal oscillator (3-1), a reset circuit (3-2), a static memory (3-3), a complex programmable logic device (3- 4), flash memory (3-5), read-only memory (3-6); described P digital signal processor chip is divided into 1 main digital signal processor chip (3-7) and P-1 from Digital signal processor chip (3-8); The master digital signal processor chip (3-7) and P-1 slave digital signal processor chips (3-8) are cascaded connected through serial input and output ports, and the data of the complex programmable logic device (3-4) , address and logic control signal input and output terminals are connected with the data, address and logic control signal input and output terminals of P digital signal processor chips at the same time, and the storage signal input and output terminals of the main digital signal processor chip (3-7) are respectively connected The storage signal input and output ends of the static memory (3-3) and the flash memory (3-5), the clock signal output end of the crystal oscillator (3-1) is connected to the clock signal input end of the main digital signal processor chip (3-7) 1. The reset signal output end of the reset circuit (3-2) is connected to the reset signal input end of the main digital signal processor chip (3-7), and the storage signal input end of the complex programmable logic device (3-4) is connected to the read-only memory (3-6) storage signal output. 3. the direction-finding device obtained by the broadband signal super-resolution direction-finding method that sparse reconstruction realizes in the continuous domain according to claim 2, is characterized in that, main digital signal processor chip (3-7) and P-1 from The digital signal processor chips (3-8) are connected through a shared bus tightly coupled to replace between the master digital signal processor chip (3-7) and P-1 slave digital signal processor chips (3-8) The cascade connection method used. 4. the direction-finding device obtained by the broadband signal super-resolution direction-finding method achieved by sparse reconstruction in the continuous domain according to claim 2, characterized in that, the main digital signal processor chip (3-7) and P-1 from The digital signal processor chips (3-8) are connected by link port cascading loose coupling to replace the main digital signal processor chip (3-7) and P-1 slave digital signal processor chips (3-8) The cascade connection method adopted between them.