JPH09219616A - Reception signal processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a reception signal processor for an array sensor which is capable of detecting a signal in a shorter time, as compared with a conventional example. SOLUTION: This processor detects plural (d) incoming signals received by the array sensor 100 composed of plural L sensor elements 1-1 to 1-L and outputs the signals. At this time, based on the reception signals from L sensor elements 1-1 to 1-L, the number (d) of incoming signal, the incoming direction of each incoming signal and the (L×d) array response matrix corresponding to each incoming direction of d signals and each sensor element are operated so that the (d) incoming signals may be maximum. A QR decomposition into the product of (L×L) unitary matrix Q and (L×L) matrix R is performed for the array response matrix. Based on the (L×d) matrix Q1 composed of the columns from the first column to the d-th column of the unitary column Q the (d×d) triangle matrix R1 composed of the lines from the first line to the d-th line of the matrix R and plural L reception signals, plural (d) incoming signals are detected and the signals are outputted to a demodulator 6.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a received signal processing device for detecting and outputting a plurality of incoming signals received by an array sensor having a plurality of sensor elements arranged therein.

[0002]

2. Description of the Related Art In a conventional received signal processing apparatus, a signal is detected mainly by performing eigenvalue decomposition of a received data correlation matrix. For example, in the received signal processing device of the first conventional example using the MUSIC algorithm, the noise subspace is made to correspond to the set of eigenvector values, and a plurality of incoming signals are detected and output. Further, in the received signal processing apparatus of the second conventional example using the signal subspace, the signal subspace is used to perform the eigenvalue decomposition of the received data correlation matrix and create the cost function. In this case, a multidimensional search is required for signal detection and estimation.

[0003]

However, in the received signal processing devices of the above-mentioned first and second conventional examples, the calculation time for executing the eigenvalue decomposition of the data correlation matrix becomes long, and it takes time to detect the signal. There was a problem.

An object of the present invention is to provide a received signal processing device for an array sensor which can detect a signal in a shorter detection time than the conventional example.

[0005]

According to a first aspect of the present invention, there is provided a received signal processing device, wherein a plurality of L sensor elements are received by an array sensor arranged at a predetermined interval and different arrival directions are obtained. From each,
A received signal processing device for detecting and outputting a plurality of d incoming signals, the number of which is smaller than the plurality of L, and a plurality of L received by the plurality of L sensor elements.
The number d of the incoming signals so that the signal component of the plurality d of the incoming signals is maximized based on the number of received signals,
Angles of the direction of arrival of the plurality of d incoming signals to the array sensor and a plurality of angles of input signals to the plurality L of sensor elements that correspond to the respective directions of arrival of the plurality d of incoming signals. (L × d) array response matrixes for representing the output signals of (L × d) are respectively calculated (L × d).
Is an (L × L) unitary matrix Q.
If, then QR decomposition to the product of the matrix R (L × d), the matrix Q ₁ consisting of a row of the first column of the unitary matrix Q to the d-th column (L × d), the matrix R from the first row and the triangular matrix R ₁ consisting of a row of up to d-th row (d × d), based on a plurality L received signals, inverse triangular matrix R ₁ of the above (d × d) The matrix of the product is calculated by calculating the product of the matrix, the conjugate transposed matrix of the (L × d) matrix Q ₁ and the received signal matrix X composed of the plurality of L received signals. It is characterized in that it is detected and output as an incoming signal.

Further, a received signal processing device according to a second aspect is the received signal processing device according to the first aspect, wherein each of the sensor elements is an antenna element and the array sensor is an array antenna. .

[0007]

Embodiments of the present invention will be described below with reference to the drawings. <Embodiment> FIG. 1 is a block diagram of a receiver according to an embodiment of the present invention. The receiver of this embodiment includes an array antenna 100 including array antennas 1-1 to 1-L, down converters 2-1 to 2-L, and an A / A converter.
It comprises D converters 3-1 to 3-L, a memory 4, a received signal processing circuit 5, and a demodulator 6. Here, in the receiver of FIG. 1, the reception signal processing circuit 5 includes the antenna element 1-
Corresponding to each direction of arrival of a plurality of d arriving signals such that the signal component of the plurality of d arriving signals is maximized based on the plurality of L received signals respectively received by 1 to 1-L To represent a plurality of d output signals with respect to input signals to a plurality of L antenna elements.
d) The array response matrix A is calculated, and the (L × d) array response matrix A corresponding to each of the arrival directions of the plurality of d arrival signals is converted into an (L × L) unitary matrix Q.
QR decomposition is performed on the product of (L × d) and the matrix R, and the unitary matrix Q includes columns from the first column to the d-th column (L ×
d) the matrix Q ₁ , a (d × d) triangular matrix R ₁ consisting of the first row to the d-th row of the matrix R, and the plurality of L
Based on the received signals, the inverse matrix of the (d × d) triangular matrix R ₁ and the conjugate transposed matrix of the (L × d) matrix Q ₁ and the plurality of L received signals. Received signal matrix X
By calculating the product of
The number of incoming signals is detected and output to the demodulator 6.

Next, the configuration of this embodiment will be described in detail. As shown in FIG. 1, antenna element 1-i (i = 1,
2, ..., L) are connected to a down converter 2-i and an A / D converter 3-i that are connected in cascade. Here, the down converter 2-i includes a low noise amplifier (not shown), a local oscillator (not shown), a mixer (not shown), and a low pass filter (not shown). . The down converter 2-i amplifies the high frequency signal input from the antenna element 1-i by a low noise amplifier, and then mixes the amplified high frequency signal and the local oscillation signal input from the local oscillator by a mixer. , An intermediate frequency signal having a predetermined intermediate frequency (hereinafter referred to as an IF signal) is frequency-converted, and is output to the A / D converter 3-i via a low-pass filter that removes unnecessary harmonic components. .

The A / D converter 3-i A / D converts the input intermediate frequency signal into an IF digital signal at a sampling frequency four times the intermediate frequency, and outputs the IF digital signal to the memory 4. The memory 4 stores a plurality of L IF digital signals input. The reception signal processing circuit 5 includes a memory 4
The output digital signal Sh is output to the demodulator 6 by performing the processing described later in detail on the IF digital signal stored in.

Next, the reception signal processing circuit 5 will be described in detail. Here, the principle of the received signal processing in the received signal processing circuit 5 will be described first, and then the configuration and operation of the received signal processing circuit 5 will be described.

<Principle> First, the array antenna 10 described above.
In the case of receiving a plurality of d (d <L) narrow band plane waves having a center frequency ω _o of 0, the received signal x _i (t) received by the antenna element 1-i is expressed by the following mathematical expression 1. It

[0012]

## EQU1 ## x _i (t) = f (t) cos {ω ₀ (t + τ _i ) + φ
(T + τ _i )} + n _i (t) i = 1, 2, ..., L

Here, τ _i is the time required for the plane wave to propagate from the origin (here, the origin is at the antenna element 1-1, for example), which is the reference point, to the i-th antenna element 1-i. Is. Also, the envelope signal f (t) of Equation 1 is a real value and is a function of slowly changing time. _ni (t) is additional noise received by the i-th antenna element 1-i. In the present embodiment, the narrow band is a narrow band to the extent that it can be considered that the values of the envelope signal f (t) and the phase φ (t) do not change during the time when the plane wave passes through the array antenna 100. Say. Signal x _i (t) received by antenna element 1-i
Is frequency-converted by the down converter 2-i. At this time, since the multiplication by the sine wave sin (ω ₀ t) and cos (ω ₀ t) is involved, the output of the signal after the low-pass filtering is the in-phase component f (t + τ _i ) cos {ω _o τ _i + φ ( t + τ _i )} and the quadrature component f (t + τ _i ) sin {ω _o τ _i + φ (t + τ _i )}. Here, since the incoming signal has a narrow band as described above, f (t + τ _i ) = f (t) and φ (t +
τ _i ) = φ (t) holds. When there are d signals in this way, the delay time τ _i has a value corresponding to the arrival direction of each plane wave and the position of the antenna element 1-i. The delay time of each plane wave in 1-i is represented by τ _ik . Where i = 1,
2, ..., L, and k = 1, 2 ,. Then, a complex number L-dimensional column vector x (t) which is the output of the array antenna 100 including the L antenna elements 1-i.
Can be expressed by the following equation 2.

[0014]

## EQU00002 ## x (t) = A (.tau.) S (t) + n (t)

Here, s (t) is the array antenna 10
Signal arriving at 0, and s _k (t) = f _k (t) exp
It is a d-dimensional column vector having {jφ _k (t)} as a component, where s _k (t) is the k-th (k = 1, 2, ..., d) of the d incoming signals. Signal. Also, A
(Τ) is an array response matrix, and a _ik (τ) = exp
Is a (L × d) matrix having (jω ₀ τ _ik ) as an element,
a _ik (τ) is the response value at the i-th antenna element 1-i of the unit wave signal in the same arrival direction as the k-th signal. In other words, the array response matrix A (τ) corresponds to the respective arrival directions of the plurality d of plane waves arriving at the array antenna 100, and the plurality d of input signals to the plurality L of antenna elements 1-i are input. (L × for expressing the output signal
It is the matrix of d).

In the following description, for simplification of explanation, it is assumed that the L antenna elements 1-i are linear array antennas arranged in a straight line at equal intervals l, and the propagation direction of the plane wave is It is assumed that all of the L antenna elements 1-i are on the same plane including a straight line. Therefore, it can be represented by τ _ik = {(i−1) · lsin (θ _k )} / c. Here, c is the speed of light. Incident angle θ
_k is the angle between the arrival direction of the k-th plane wave and the normal line of the straight line on which the L antenna elements 1-i are arranged. Here, the incident angle vector θ = [θ ₁ θ ₂ ... θ _d ] which is a d-dimensional column vector representing all incident angles is a parameter to be detected in this embodiment, and is the array response matrix A (θ). Is the incident angle vector θ = [θ ₁ θ ₂ … θ
Emphasize direct dependence on [ _d ]. Signal s
(T) is represented as a d-dimensional vector signal, and the noise vector n (t) is represented by an L-dimensional column vector of white noise having the same power σ ² . Where the signal vector s
It is assumed that (t) and the noise vector n (t) have no correlation. Therefore, at each time t _n (n = 1, 2, ..., N), N L-dimensional column vectors x (t ₁ ), x (t ₂ ),
, X (t _N ), the received signal matrix X can be expressed by the following equation 3.

[0017]

X = [x (t ₁ ) x (t ₂ ) ... x (t _N )] = A (θ) S + N ₀

Here, the (L × N) signal matrix S = [s
(T ₁ ) s (t ₂ ) ... s (t _N )], and (L ×
N) noise matrix N ₀ = [n (t ₁ ) n (t ₂ ) ... n
(T _N )]. When N is large, the autocorrelation matrix R _x expressed by Expression 4 can be expressed by the following Expression 5 using the signal autocorrelation matrix R _s .

[0019]

(4) R _x = (1 / N) XX †

## EQU00005 ## R _x = A (θ) R _s A (θ) † + σ ² I

Here, † is a symbol indicating a conjugate transposed matrix. It is assumed that the array response matrix A (θ) is well known in the case of mathematical model or measurement, and the unitary matrix Q (d, θ) is used in place of the eigenvalue decomposition of the data correlation matrix, and is expressed by Equation 6. Perform an efficient QR decomposition performed. QR
The decomposition is described in “Iwanami Mathematics Dictionary”, 3rd edition, edited by The Mathematical Society of Japan, published by Iwanami Shoten, December 10, 1985, etc.

[0021]

## EQU6 ## A (θ) = Q (d, θ) R

Here, the array response matrix A (θ) is a (L × d) matrix as shown in Equation 7, and in order to simplify the description in the following description, (θ) is omitted and simply It is referred to as an array response matrix A. Also, the unitary matrix Q (d,
(θ) is a (L × L) matrix as shown in Equation 8, and (d, θ) is omitted in the following description and is simply referred to as a unitary matrix Q. Further, R in the equation 6 is a (L × d) matrix and is represented by the equation 9.

[0023]

(Equation 7)

(Equation 8)

[Equation 9]

Here, the unitary matrix Q is composed of the (L × d) matrix Q ₁ consisting of the first column to the d-th column of the unitary matrix Q shown in Formula 10 and the unitary matrix Q shown in Formula 11. (D + 1) th to Lth columns {L
× (L−d)} matrix Q ₂ and Q = [Q ₁ Q ₂ ]
It expresses. Further, the matrix R of the equation 9 is represented by the equation 12 (d
It is expressed as shown in Expression 13 using the triangular matrix R ₁ of × d). Here, R ₀ has all zero components {(L−
d) × d} zero matrix.

[0025]

(Equation 10)

[Equation 11]

(Equation 12)

(Equation 13)

Next, when the equation 2 is transformed by using A = QR of the equation 6, it can be expressed as the following equation 14, and the conjugate transposed matrix Q of the unitary matrix Q from the left on both sides of the equation 14 is obtained.
By multiplying by, the conversion can be performed as shown in Expression 15.

[0027]

X (t) = QR · s (t) + n (t)

[Expression 15] Q † x (t) = R · s (t) + Q † n (t)

Further, as described above, since the matrix R has the structure shown in the equation (13), the (d + 1) th to the Lth components after the L-dimensional column vector Q † x (t) of the converted received signal are obtained. Is a component that depends only on noise. Assuming that the array response matrix A is full rank for different incident angles θ ₁ , θ ₂ , ..., θ _d , the power of the channel corresponding to the last (L−d) components is minimized. Such a unitary matrix Q corresponds to the correct incident angle θ. Here, using the Frobenius norm ‖ [0I] Q † X‖ _F expressed by Equation 16, the cost function V (θ, d) to be minimized is
It is expressed by Equation 17.

[0029]

Equation 16] ‖ _{[0I] Q † X‖ F =} 1 / √ {N (L-d)} · [0I] Q † X = 1 / √ {N (L-d)} Q 2 † X

[Expression 17] V (θ, d) = {1 / (N (L−d))} ‖ [0
I] Q † X‖ _F ²

[0I] is a (Ld) xL matrix consisting of a zero matrix 0 of (Ld) xd and an identity matrix I of (Ld) x (Ld). And, from the (d + 1) th row to the Lth row (L−
d) Rows are elected. As the number of data increases, when the cost function V (θ, d) takes the global minimum value, the value approaches the noise power assumed to be equal in each channel (V _min (θ, d) ≈σ ² Can be observed. That is, if the cost function V (θ, d) is minimized, the number d of signals and the incident angle θ can be obtained.

Next, a detection algorithm for detecting the number d of incident signals and the incident angle θ so that the cost function V (θ, d) is minimized will be described. For example, when the number of detected signals da is less than the number of actually reaching signals d, the minimum value of the cost function V (θ, d) depends not only on noise but also on signals. That is, Equation 18 is established from Equation 5.
Here, in order to simplify the analysis, they have substantially the same power (R _s ≈ <s ² > I is satisfied using the average signal power value <s ² >) and {θ ₁ , θ ₂ , ..., θ _d } of da
Let θ _i for each (da <d) be an element (i = 1, 2, ...,
da) Assume uncorrelated signals with the smallest cost function value for the incident angle vector θ. Without loss of generality, an incident angle vector having da θ _i as elements can be set as θ = [θ ₁ θ ₂ ... θ _da ]
The following equation 19 holds.

[0032]

[Equation 18] Q † R _x Q ≈Q † AR _s A † Q + σ ² I

Q † R _x Q ≈ [R 'Q † a (θ _{d + 1} ) ... Q † a (θ _d )] R _s ≈ [R' Q † a (θ _{d + 1} ) ... Q † a (Θ _d )] † + σ ² I

Here, R ′ is a matrix having a triangular matrix R ′ ₁ and is represented by the equation 20. In Equation 20, R ′ ₁ is da
It is a triangular matrix of × da, and 0 is a zero matrix of (L-da) × da. At the global minimum value of the cost function V (θ, d), if a _i = a (θ _i ), then the cost function V (θ, d)
Can be approximated by the following equation 21.

[0034]

(Equation 20)

(Equation 21)

Here, Trace [a _i a _i † Q ₂ Q ₂ †] is
It represents the sum of the diagonal elements of the matrix [a _i a _i † Q ₂ Q ₂ †]. Number 2
As is clear from 1, the minimum value σ ² is achieved when the sum of the second terms on the right side of Expression 21 is zero, that is, da = d.
This holds particularly well at high signal-to-noise power ratio SNR. On the other hand, for the low signal-to-noise power ratio SNR, the contribution of the second term in the approximate expression of Eq. 21 is small, so that the difference in the cost function value with respect to the value of the incident angle vector is greatly influenced by σ ² . Since it does not exist, it is empirically known that it is better established. Instead of finding each minimum of the cost function V (θ, d) at different values of da, the cost function V (θ, θ,
Compare the slopes of d). The number of detected signals corresponds to a value for this slope such that this slope is reduced compared to the slope in the previous process. In summary, the algorithm of the method according to the invention performs a minimization on the number of signals while performing another minimization on the signal parameter (incident angle) to be detected, as described by It can be seen as an algorithm.

[0036]

(Equation 22)

Here, the function in which d or θ is described below min is a function in which d or θ is changed and the function value becomes the minimum. The first minimization is a coarse one-dimensional search for discrete values of increasing individual signal number d,
A subroutine is called for each value of, which is multidimensional over the space of signal parameters (in this case the incident angle vector θ) that can converge to some minimum cost function value. Perform a search. Here, since it is a multi-dimensional search, it is necessary to perform careful initialization. This value is compared with the previous two values to find the maximum decreasing slope for which an accurate detection value determination was made. Both minimization processes described above require initialization. The index for selecting the initial value for the first and second minimization processing will be described later.

The above detection algorithms are summarized as follows. (1) Setting da to a predetermined initial value, and initializing or calculating the minimum value of the cost function V (θ, d) shown in Expression 21 for d = da-1 and d = da-2, respectively. To do. Here, details of how to perform the initialization selection will be described later. (2) The operation of minimizing the cost function V (da, θ) is performed with respect to θ = [θ ₁ θ ₂ ... θ _da ]. This is the number 23
, Θ = [θ ₁ , θ ₂ , ..., θ _da ] that gives the minimum value of the cost function V (da, θ) is obtained. The minimum value of the cost function at this time is V _min (da, θ).

[0039]

(Equation 23)

(3) The zero hypothesis Hyp ₀ that the number of signals d is da is established. Zero hypothesis Hyp ₀ : d = da (4) V _min (d, θ)> 2 V _min (d-1, θ) -V
_{If min} (d−2, θ) is satisfied, the zero hypothesis Hyp ₀ is accepted, that is, the signal number ends as d = da, and if not satisfied, the zero hypothesis Hyp ₀ is excluded and d = d.
Set a + 1 and return to (2).

Here, in an actual application in which a signal is continuously received, the signal is repeated.
The initial value of is the minimum number of current signals predicted at any time. For example, in a mobile communication radio system, one specific frequency is assigned to each mobile in one cell so that interference occurs only from neighboring cells that use the same frequency. In a normal cell architecture, there are 6 cells that cause interference adjacent to each other, so the initial value is da = 7 (7 for one desired signal and 6 interference signals). In reality, the actual correct number will be much larger because the paths from non-adjacent cells are added. However, if there is no information on the number of signals, da = 1, da = 2, da = to start the algorithm.
A rough calculation of the initial cost function value of 3 is required.

Since the minimization process described above requires multidimensional calculation, it is preferable to set the initial value to an actual value. In the simulation in the example described later, the convergence is achieved even when the initial value is set with the incident angle θ deviating by 10 degrees from the actual value. In an application in which the receiver can almost confirm the directions of the interference signal and the desired signal, the initial value can be estimated. For example, in a mobile radio communication system, the position of the interfering cell with respect to the base station due to the geographical situation limits the interval in the arrival direction of the interference wave to some extent. If the estimation is impossible, the minimization process with different initial value parameters can be repeated several times to select the output having the lowest value of the minimum cost function. As another advantage, since the search operation is repeated for the increasing number of signals d, the output value of the previous minimization process requires an additional initial value in order to increase the number of signals d to be calculated this time. Can be used as an initial value in the processing of. This is described in Prior Art Document 1 "I. Ziskind.
and M.D. Wax, "Maximim Likelili
Hood of Multiple Sources
by Alternating Projectio
n ”, IEEE Transaction on Ac
oustic, Speech, SignalProce
ssing, Vol. ASSP-36, No. 10, p
p. 1553-1560, October 1988 ”is an application example of another projective algorithm.

Although the eigenvalue decomposition is not performed in the above-described detection method, multidimensional search processing is necessary to find the global minimum value of the cost function V (θ, d). Here, an order O proportional to the multiplication number Ld ² in a quadratic function when initialized so as to sufficiently approximate the global minimum value.
A modified variable projection algorithm that converges with the amount of calculation of (Ld ² ) was used. Here, the modified variable projection algorithm is described in Prior Art Document 2 “GW Stewa”.
rt, “An Updating Algolism”
for Subspace Tracking ”, IE
EE Transaction on Signal
Processing, Vol. 40, no. 6,19
June 1992 ”gives a detailed explanation. According to this, the following mathematical operation is required in the optimization process. First, the QR factorization of the array response matrix A can be expressed by the following equation 24.

[0044]

(Equation 24)

Next, the intermediate variable D necessary for calculating the gradient and the Hessian matrix during the optimization process is defined by the following equation 25.

[0046]

[Equation 25] D = [∂a (θ ₁ ) / ∂θ ₁ ∂a (θ ₂ ) / ∂θ ₂
…… ∂a (θ _d ) / ∂θ _d ]

Here, since the array response matrix A needs to be known in advance by measuring or performing an operation using a mathematical model, the factorization of the QR factorization can be performed in advance as with the intermediate variable D. . Next, the function Φ and
The error function Ψ that minimizes the output with the correct parameter value and the calculated output signal vector Ω are represented by the following equations 26, 27, and 28, respectively.

[0048]

Φ = Q ₂ † D

Ψ = M † Q ₂

[Equation 28] Ω = R ₁ ⁻¹ Q ₁ † M

Here, M in Equations 27 and 28 is a normalized reception signal matrix, and M = [1 / √ {N (L-
d)}] · X. As a result of the predetermined mathematical operation, the reference function V, the gradient vector V ′, the Hessian matrix H, and the error Δθ of the search angle are respectively the following Equation 29, Equation 30, Equation 31, and Equation 31,
It can be expressed by Equations 32 and 33.

[0050]

V = Trace {ΨΨ †}

V ′ = 2Re {diag (ΩΨΦ)}

[Equation 31] H = 2Re {(Φ † Φ) * (ΩΩ †) ^T }

H _ii = H _ii (1 + μ _k )

[Expression 33] Δθ = H ⁻¹ V ′

Then, the incident angle θ ^{k + 1} updated from the following equation 34 is calculated.

[0052]

[Equation 34] θ ^{k + 1} = θ ^k −Δθ

Here, the symbol * represents the Sure product, and μ
_k is a step size parameter selected by the Rebenberg / Markald's step length method. The method selects the step size so that it behaves as a steepest descent method when it is far from the minimum value and as an inverse Hessian method when it is close to the minimum value.
Here, with respect to the Rebenberg / Markald's step length method, the related art document 3 “WH Press.
et al. , "Numerical receipts
in C ", Cambridge University
y Press, Sec. Edition, 1992 "
Is described in detail. In practice, when the cost function increases, the step size parameter μ _k is multiplied by 10,
When the cost function decreases, the step size parameter μ _k is decreased to 1/10. The details are described in Chapter 15 of Related Art Document 3.

The initialization process is possible by using the alternating projection algorithm, which is an application of the relaxation optimization principle (one parameter at a time) to the optimization problem. The optimization problem is described in detail in the above-mentioned Prior Art Document 2, and the alternate projection algorithm is described in detail in the above-mentioned Prior Art Document 3. In this example, da
Since the search process is repeated by increasing the value of, the minimum value of the previous minimization process is directly applied as the initial value in this process that requires an additional initial value to increase the number of signals da. can do.

Next, the case where the position of the signal source is different and the SNR is different was examined by computer simulation. The result is that as long as the SNR is above a certain threshold and the angles of incidence of the signals are some distance from each other, satisfactory detection and estimates are obtained. If these conditions are not satisfied, and especially if the incident angles θ of the two signals are very close to each other, more accurate initialization processing is required. In the mobile communication system, since the position of the signal source is not so far away between the detection operations, better initialization can be performed by using the estimated value of the previous incident angle θ as the next initial value. When the incident angle θ is sufficiently far apart, the algorithm of the method of the present invention makes it possible to create an estimated value that is sufficiently accurate as an initial value for the next detection calculation, and has a high convergence in terms of the resolution of the incident angle θ. Sex was obtained.

Since the basic contribution of the method simplifies the computation without severely degrading its performance, a simple beamforming method can be provided. It is proposed that the relational expression of the following expression 35 is established from the expression 3 and the expression 24 using the signal matrix S, and the output signal vector Sh is represented by the approximate expression of the following expression 36. That is, the output signal vector Sh is the inverse matrix of the triangular matrix R ₁ of d × d, and the conjugate transpose matrix Q ₁ † matrix Q _1,
It is represented by the product with the received signal matrix X.

[0057]

[Formula 35] R ₁ ^-1 Q ₁ † X = S + R ₁ ^-1 Q ₁ † N

[Formula 36] Sh≈R ₁ ^-1 Q ₁ † X

Based on the principle described in detail above, in the reception signal processing device 5, the number of signals d and the incident angle vector θ are calculated so that the cost function V (θ, d) is minimized, and the signals concerned are calculated. It outputs the number d, the incident angle vector θ, and the output signal vector Sh.

Next, the received signal processing circuit 5 which executes received signal processing based on the above-mentioned principle will be described. As shown in FIG. 2, the reception signal processing circuit 5 includes a CPU 10, a ROM 11, a RAM 12, an interface 14 and a memory 13. Here, the memory 13 includes a normalized received signal matrix memory 21, a cost function value memory 22, an output signal vector memory 23, an incident angle vector memory 24, an array response matrix memory 25, a QR decomposition matrix memory 26, and a gradient vector memory 27. Hessian matrix vector memory 28
It is divided into and.

In the reception signal processing circuit 5, the ROM 11
3 stores a received signal processing program shown in FIGS. 3 and 4 and data necessary for executing the program, and the CPU 10 stores the required data in the memory 4 according to the received signal processing program stored in the ROM 11. ,
RAM 13 used as memory 13 and work area
2), execute the received signal processing described later,
The output signal vector Sh is output to the demodulator 6 via the interface 14 that executes interface processing such as signal conversion between the CPU 10 and the demodulator 6.

The received signal processing circuit 5 will be described in detail below with reference to FIGS. 2, 3 and 4. First, FIG.
In step S1 of the flowchart of FIG.
Is the signal number parameter da and the cost function values V _da-1 , V
Set _da-2 and a predetermined initial value, and set the signal number parameter d
a is stored in the RAM 12 and the cost function values V _da-1 , V _da
Store _da-2 in the cost function value memory 22. Next, in step S2, the CPU 10 calculates the normalized reception signal matrix M using the reception signal matrix X read from the memory 4 and the signal number parameter da read from the RAM 12, and calculates the calculated normalized reception signal. The matrix M is stored in the normalized reception signal matrix memory 21.

Next, in step 3, the following cost function value V _da calculation processing is executed to store the calculated cost function value V _da in the cost function value memory 22. That is, in step S3 of the subroutine, as shown in FIG. 4, in step S31, the CPU 10 sets the incident angle vector θ ^k to the initial value θ ⁰ , and then in step S32.
And the array response matrix A based on the incident angle vector θ ^k
(Θ ^k ) is calculated, and the calculated array response matrix A (θ ^k )
Are stored in the array response matrix memory 25. Further, in step S33, the matrix Q ₁ , Q ₂ , R ₁ is calculated using the conversion formula ^obtained by QR decomposition of the array response matrix A (θ ^k ) read out from the array response matrix memory 25, and the matrix Q ₁ ，
Q ₂ and R ₁ are stored in the QR decomposition matrix memory 26.

In step S 34, the CPU 10 reads the normalized received signal matrix M read from the normalized received signal matrix memory and the matrix Q read from the QR decomposition matrix memory 26.
_1, Q _2, with the R ₁ and number 24 to number 30 by calculating the cost function value V _da and gradient vector V _'da and Hessian matrix H, stores the cost function value V _da cost function value memory 22 The gradient vector V ′ _da is stored in the gradient vector memory 27, and the Hessian matrix H is stored in the Hessian matrix memory 28. In step S35, the CPU 10 calculates the gradient vector V ′ _da read from the gradient vector memory 27, the Hessian matrix H read from the Hessian matrix memory 28, and the number 32.
The error Δθ of the search angle is calculated using and, and the error Δθ is stored in the RAM 12. In step S36,
The CPU 10 compares the error Δθ read from the RAM 12 with a preset reference error Δθ _a, and if the error Δθ is smaller than the reference error Δθ _a , the process proceeds to step S4 of the main program, and otherwise. Proceeds to step S37. In step S37, the CPU 10 sets θ ^k.
Θ ^k is updated by substituting θ ^k −Δθ into, and the process proceeds to step S32 and steps S32 to S3.
Repeat 6

In step S4 of FIG. 3, the CPU 10
The cost function value V read from the cost function value memory 22
_{Based on da} , V _da-1 , V _da-2 , V _da > 2V _da-1 −V
By determining whether or not _da-2 , V _da > 2V _da-1 −V _da-2
When it is, it progresses to step S6 and V _da > 2V _da-1
If not −V _da-2 , the process proceeds to step S5. In step S5, the CPU 10 sets d in the signal number parameter da.
The signal parameter da is updated by substituting a + 1, the signal parameter da is substituted for the signal parameter da ′, the signal parameter da ′ is updated, and the output signal vector Ω is substituted for the output signal vector Ω ′. The vector Ω ′ is set, the process proceeds to step S2, and the processes of steps S2, S3, and S4 are repeatedly executed.

In step S6, the number d of incoming signals and
The angle of each of the arrival directions of the d arrival signals to the array antenna 100 is detected, and √ {N
The output signal vector Sh is calculated by multiplying (L-da ′)} and output to the demodulator 6 in step S7. As described above, the output signal vector Sh can be obtained by a relatively simple calculation of multiplying the already calculated output signal vector Ω ′ by √ {N (L-da ′)}. It is one of the advantages of the method of the embodiment used.

The reception signal processing circuit 5 having the above-described configuration, based on the plurality of L reception signals output from the array antenna 100, makes the array so that the signal components of the plurality of d arrival signals are maximized. The response matrix A is calculated, and the array response matrix A is converted into an (L × L) unitary matrix Q,
QR decomposition into a product of (L × d) with the matrix R, and the columns from the first column to the d-th column of the unitary matrix Q (L × d)
Based on the matrix Q ₁ and the (d × d) triangular matrix R _{1 including} the rows from the first row to the d-th row of the matrix R and a plurality of L received signals, the above (d × d) By calculating the product of the inverse matrix of the triangular matrix R ₁ of the above, the conjugate transpose matrix of the matrix Q ₁ and the received signal matrix X.
The h is output to the demodulator 6. Thereby, the receiver of the embodiment can detect a plurality of incoming signals and output the received data signal from the demodulator 6.

The processing algorithm executed by the received signal processing circuit 5 of the embodiment described in detail above calculates the array response matrix A for a plurality of different d incoming signals by using the QR transform to calculate d signals. It is outputting. In this signal processing method, most of the required signal processing operations are simple matrix data operations such as QR factorization, matrix product and sum, inverse matrix calculation, and Sure product. The calculation time can be shortened and the signal can be detected in a short detection time.

The reception signal processing circuit 5 of the embodiment described in detail above calculates the array response matrix A for a plurality of different arrival signals by using QR decomposition, and makes the arrival direction of the narrow-band signal beam high speed. Can be detected with. As a result, it is particularly effective in a field where a narrow-band signal beam is used to increase the communication capacity in a communication system such as mobile communication and satellite communication in which the allocated frequency is saturated rapidly. Is. It is also effective for spatial localization in cellular telephone systems that require channel switching.

<Modification> The receiver of the above embodiment is a radio receiver that receives an incoming radio signal by using the array antenna 100. However, the present invention is not limited to this, and the vibration propagating in water is used. The arrival directions of other vibration waves such as waves and seismic waves may be detected and output. That is, based on the detection signal detected by the array sensor in which a plurality of sensor elements for detecting a predetermined vibration are arranged, the signal processing algorithm executed by the reception signal processing circuit 5 described above is used to determine the arrival direction of the signal. Can be configured to output a desired signal.

[0070]

EXAMPLES The present inventors conducted various simulations in order to confirm the effects of the present invention. The results of each simulation will be described below. Here, in each simulation, the array antenna 100 is composed of eight antenna elements 1-1 to 1-8, and a linear array antenna arranged in a straight line at intervals of half a wavelength of a signal coming from each antenna element is used. It was assumed that four signals having the same power were input from different directions. Each input signal and noise vector was individually created in the computer by using a normal random number generation subroutine as initial values of different species. The vector size of the data in each simulation was set to 100 corresponding to the snapshot of the signal, and the points shown in the graph represent the signal and noise as an average of 200 measurements. In the case of coherent signals, the corresponding data vectors are assumed to be the same signal because they are initialized with the same kind of initial value in the normal random number generation subroutine.

Example 1 FIG. 5 shows an array antenna 10
The incident angles θ of the four signals input to 0 are -30
The cost function V (θ, da) for each signal parameter da in the process of detecting the number of signals d of the reception signal processing circuit 5 when fixed to ゜, 20 ゜, 30 ゜, 60 ゜
It is a graph showing the value of. Here, in this simulation, the incident angles θ of the four signals input to the array antenna 100 are −30 °, 20 °, 30 °, and 60, respectively.
It was fixed at °. In the graph of FIG. 5, the signal-to-noise power ratio SNR is changed in 2 dB steps from −10 dB to 20 dB, and is shown for each SNR value. As is clear from the graph of FIG. 5, when the signal-to-noise power ratio SNR> 8 dB, the cost function V (θ, da) shows the minimum value when the number of signals to be detected is da = 4. That is, it is shown that by detecting the minimum value of the cost function V (θ, da), it is possible to detect the correct signal number da = 4 when the signal-to-noise power ratio SNR> 8 dB.

If the above-mentioned method of utilizing the slope of the cost function V (θ, da) with respect to the continuous detection number da is used, the number of signals d = 4 when the signal-to-noise power ratio SNR> -6 dB or less. It was found that can be detected. That is, according to the received signal processing circuit 5 of the embodiment, it was confirmed that when the signal-to-noise power ratio SNR> -6 dB, the correct number of signals, 4, can be detected.

As a result of further studies, it was found that the algorithm of the method of the present invention is effective for both noncoherent signals and coherent signals. When the coherent signal was used, it was judged that the signals with incident angles θ of 20 ° and 30 ° were the same.

<Embodiment 2> In Embodiment 2, the incident angles θ of the four signals input to the array antenna 100 are −.
Under the conditions fixed at 30 °, 20 °, 30 ° and 60 °, the standard deviation of the error of the detected incident angle θ when the correct number of signals 4 was detected was evaluated. FIG.
Graph shows the result as the standard deviation of the error of the incident angle θ with respect to the signal-to-noise power ratio SNR. Here, in the graph of FIG. 6, the standard deviation of the error of the detected incident angle of the signal with respect to the incident angle θ = 30 ° is shown assuming that the four signals are incoherent. Further, the graph of FIG. 6 also shows a case where the WSF method is used for comparison. Here, the WSF method is described in Related Art 6
"M. Viberg et al.," Detecti
on and Estimation in Sens
or Arrays Using WeightedS
"subspace Fitting", IEEE Tra
nsactionon Signal Process
ing, Vol. 39, no. 11, 1991 11
It is described in detail in "Month". As is clear from the graph of FIG. 6, it can be seen that the method of the embodiment has detection accuracy equivalent to that of the WSF method in the range of the signal-to-noise power ratio SNR shown in the figure. The graph of FIG. 7 shows a case where the four input signals are coherent signals. As is clear from the graph of FIG. 7, even when the four input signals are coherent signals, the method of the embodiment has detection accuracy equivalent to that of the WSF method.

<Third Embodiment> In the third embodiment, 4 is input.
Of the two signals, the incident angle of the signal corresponding to the incident angle θ = 20 ° is changed between 20 ° and 26 ° so that the signal-to-noise power ratio SNR is between -5 dB and 10 dB. The ratio of detection error to the specific SNR value was measured. In Example 3, the incident angles θ of three signals other than the incident angle θ = 20 ° of the four signals are −30 ° and 30, respectively.
The test was carried out by fixing at 60 °. The initial value in the detection process is such that when the reception signal processing circuit 5 is actually used, the detection process of the incident angle θ is continuously performed, and the variation of the incident angle θ during each detection period does not occur. Considering the smallness, the output result of the incident angle θ in the previous detection process was used as the initial value in the detection process. As a result, it has been clarified that the detection error between the signal-to-noise power ratio SNR between −5 dB and 10 dB is 10% or less when the difference between the incident angles θ of adjacent signals is 4 ° or more.

<Embodiment 4> Next, the result of evaluating the ratio of the interference signal power and the noise power included in the output signal output from the reception signal processing circuit 5 will be described.
Here, the ratio of the output signal power to the added value of the noise power and the interference signal power, which is represented by Expression 37, {output signal power /
(Noise power + interference signal power)} OSINR and the ratio of the input signal power to the added value of the noise power and the interference signal power expressed by Equation 38 {input signal power / (noise power + interference signal power)} ISINR It evaluated using and.

[0077]

(37)

(38)

Here, E (·) is a statistical expected value, and w
Are the corresponding weighting vectors. Also, * represents a conjugate complex number. The difference between OSINR and ISINR represents the degree of improvement in signal processing using the adaptive beam forming method.

In the fourth embodiment, the signal-to-noise power ratio SNR and the incident angle θ are changed to change the respective signal-to-noise power ratio S.
A simulation is performed for each incident angle θ in NR to calculate OSINR of the output signal, and LCMV (Lin
ear Constrained Minimum V
and the output signal OSINR calculated using the reference signal method. Here, the LCMV method and the reference signal method are described in the related art document 4 “BD Va”.
n Veen et al. , "Beamformin
g: A Versatile Approach to
Spatial Filtering ”, IEEE
ASSP Mag. Pp. 4-24, 1988 4
Moon ”for more details.

Here, first, the signal-to-noise power ratio SNR is set to 0 dB, and the incident angles θ of the three signals are fixed to -30 °, 30 °, and 60 ° as in the case of the above-described third embodiment. The incident angle θ of the other one signal is set in steps of 1 ° from 20 ° to 26 °, and the OSNIR at each incident angle θ is obtained, and shown in the graph of FIG. In the graph of FIG. 8, the results calculated using the LCMV method and the reference signal method under the same conditions are shown for comparison.
Here, the ISNIR under the above-mentioned conditions was a constant value of about -6 dB. Further, both the coherent signal and the non-coherent signal were evaluated by the method of the present invention and the LCMV method, but the same result was obtained for the coherent signal and the non-coherent signal. Of course, the coherent signal cannot be detected by the reference signal method. The gain obtained with the array antenna 100 in the present Example 4 varied between 14 dB and 7 dB.

FIG. 9 shows the signal-to-noise power ratio SNR of 10d.
9 is a graph showing a result of simulation by setting B and setting other conditions as in the case of obtaining the graph of FIG. 8. In this case, ISNIR = -5 dB. Further, when the incident angle θ = 20 °, that is, when the difference between the incident angle and the signal of the adjacent incident angle θ = 30 ° is 10 °, the maximum gain is 22 dB, and the incident angle θ = 26 °.
In other words, that is, when the difference in the incident angle from the adjacent signal having the incident angle θ = 30 ° was 4 °, the minimum gain was 16 dB.

Example 5 Next, in the improvement of the received signal processing method in the embodiment, an attempt was made while paying attention to the following two constraint conditions for performing real-time processing by a digital circuit. One is the operation speed obtained by the parallel pipeline structure, and the second is the accuracy of the numerical result obtained by the algorithm which is surely brought into a good state. The degree of numerical instability when numerically solving simultaneous linear equations is determined by the condition number of the coefficient matrix. The condition number Cn (Y) of the matrix Y is represented by the ratio between the maximum eigenvalue and the minimum eigenvalue of the matrix Y, and the larger the condition number Cn (Y), the more unstable numerically. Here, the signal processing unit (SPU) required by the present algorithm includes the following first to third parts.

The first part is a signal processor which prepares the data for digital signal processing after it is received by the array antenna 100. Here, instead of calculating the autocorrelation value R _x , the data matrix Cn represented by the following equation 39
(X) was used. This is a numerical advantage.

[0084]

Cn (X) = √ {Cn (R _x )}

The second part is QR decomposition of A to obtain values of the matrices Q ₁ , Q ₂ , R ₁ and the inverse matrix R ₁ ^{-1 for} the incident angle θ different from the initial value and the number of signals da to be detected. Is a signal processing unit for obtaining. Two algorithms, the Givens formula and the Householder formula, are known to efficiently perform the conversion. These algorithms are the Givens formula and the Householder formula, which are disclosed in the prior art document 5 “NL Ownsly,“ System ”.
olic Array Adaptive Beamf
orming ”, Naval Underwater
Sys. Center, Technical Repo
rt 7981, September 1987 ", prior art document 6. The Givens rotation method operates on the left (or right) of one matrix, and zeros one of the two rows (or columns) of the column (or row) of interest, as shown in the following Equation 40. .

[0086]

(Equation 40)

Here, the rotation coefficients c and s satisfy the equations 41 to 43.

[0088]

[Expression 41] -sx _i + cy _i = 0

[Expression 42] s * s + c * c = 1

[Expression 43] c * = c

That is, in order to triangulate the (n ₁ × n ₂ ) matrix Y, the erasing method as described above can be used. If n ₁ > n ₂ then matrix Y
The columns of Y can be zeroed from top to bottom, left to right by processing the left side of the and multiplying sequentially. Applying the transformation to the QR factorization of the array response matrix A to obtain the equation (6) requires a further plane rotation of {(2L-d-1) / 2} · d. Here, in order to execute the plane rotation, an efficient algorithm that does not require a multiplier operation can be used. The algorithm is described in Related Art 7
"JM Delosme," Cordic Algo
ritms: Theory and Extensi
on ”, SPIE Vol. 1152 Adv. Alg
orithms and Architectures
for Signal Processing I
V, pp. 131-145, 1989 ". Furthermore, in order to improve the processing efficiency, the Givens rotation method algorithm can be applied to the above parallel pipeline structure.

The second Householder algorithm described above is used to perform the QR decomposition. Since the unitary symmetric matrix P is represented by the following formula 44, the conversion formula is
When u = x ± ‖x‖e ₁ , it can be expressed as in the following Expression 45.

[0091]

P = I−2 · (uu †) / (u † u)

[Expression 45] Px = ‖x‖e ₁

Therefore, until all the triangulation processing is completed, all the elements below a predetermined row are acted on and a plurality of columns are made zero at the same time. The parallel pipeline structure applied here is preferable to the Givens algorithm because the processing device can be simplified as compared with the Householder transformation.

The third part is a basic linear arithmetic processing unit necessary for the optimization processing shown in the equations 21 to 32. Here, eight matrix for multiplying but the most the number of operations required computation is proportional to Ld ² orders O (L
It was a complex number operation of d ² ). This is equivalent to QR decomposition, which requires a considerable amount of calculation. Also, the inverse matrix of the two matrices must be calculated. One is an operation to reverse matrix triangular matrix which requires the calculation of the order O (d ³⁾ which is proportional to d ^3, the other is an operation of SURE product that requires calculation of d ^2.

The above calculation was executed and the number of calculations was compared with the conventional example. Here, since the number of antenna elements L> the number of signals d, the cost value of signal processing in a complicated operation is in the order O (L ³ ) operation which is proportional to L ³ in the conventional proper decomposition, and in the method of the present invention. expressed by comparing the calculation of the order O (ld ²⁾ proportional to ld ^2.
Further, to simplify the explanation, the number of operations of the computer processing shown in FIGS. 10 and 11 is calculated by the cube of L (L ³ ) as the number of operations of eigenvalue decomposition, and the QR decomposition of the present invention is used. The number of operations is shown by being calculated by L × d ² . Here, the graph of FIG. 10 shows the number of calculations for the number of antenna elements L when d = 7 is set, and the graph of FIG. 11 is the number of antenna elements when the number of signals d = L−2 is set. The number of calculations for L is shown. As a result, FIG.
As can be seen from FIG. 0 and FIG. 11, the signal processing method of the present invention using QR decomposition can reduce the number of operations as compared with the conventional signal processing method using eigenvalue decomposition. Further, as is clear from FIG. 11, the effect is more remarkable as d is smaller than the number L of antenna elements.

In the QR decomposition, in the matrix multiplication or the inverse matrix arithmetic processing, if the parallel processing architecture such as the wavefront array is used, the above arithmetic cost can be reduced. Further, in the QR decomposition and the calculation of the inverse matrix of the triangular matrix, the systolic array can be used as another possible selection example. Regarding the parallel processing architecture, the prior art document 8 “SY Kung et a.
l. , "Wavefront Array Proce
ssor: Language, Archectur
e, and Application ", IEEE
Transaction on Computer, V
ol. C-31, November 1982 ". For systolic arrays,
This is described in detail in the above-mentioned Prior Art Document 5.

From the results described in Examples 1 to 5 above, the characteristics of the received signal processing method used in the received signal processing circuit 5 of the embodiment can be summarized as follows. (1) Compared with the reference signal method of the conventional example, the incident angle θ of the incoming signals can be detected more accurately, and even when the coherent signals are incident, the incident angles θ of the incoming signals can be calculated. Can be detected. (2) As compared with the conventional LCMV method, the number of operations for detecting the incident angle θ of the incoming signal can be reduced,
It is possible to detect the incident angle θ of a signal that arrives at high speed. That is, the received signal processing method described in the embodiments can be used in the same manner as a detection method using a computer operation such as a linear constraint minimum variance (LCMV) method or a reference signal method, and in particular, cellular mobile communication with call channel switching. It is effective for detecting the incident angle θ of the incoming signal in the system.

[0097]

According to the first aspect of the present invention, the received signal processing device according to the first aspect of the present invention includes the plurality of L received signals respectively received by the plurality L of sensor elements, based on the plurality of L received signals. In order to maximize the signal component, the number d of the incoming signals, the angles of the incoming directions of the multiple d incoming signals to the array sensor, and the incoming directions of the multiple d incoming signals are respectively set. Correspondingly, an (L × d) array response matrix for representing a plurality of d output signals with respect to input signals to the plurality of L sensor elements is calculated, and each arrival direction of the plurality of d arrival signals is calculated. The (L × d) array response matrix corresponding to each of the above is QR decomposed into the product of the (L × L) unitary matrix Q and the (L × d) matrix R, and one column of the unitary matrix Q is obtained. Consists of rows from the eye to the d-th row (L × d) Matrix Q ₁ and ₁ of the above matrix R
A triangular matrix R ₁ consisting of a row from row to d-th row (d × d), based on a plurality L received signals, and the inverse matrix of the triangular matrix R ₁ of the above (d × d) , Above (L × d)
By calculating the product of the conjugate transposed matrix of the matrix Q ₁ and the received signal matrix X composed of the plurality L of received signals, the matrix of the product is detected and output as the plurality d of incoming signals. As a result, it is possible to provide a received signal processing device for an array sensor that can detect signals in a shorter time than in the conventional example.

According to a second aspect of the present invention, there is provided the received signal processing device according to the first aspect, wherein each of the sensor elements is an antenna element and the array sensor is an array antenna. As a result, it is possible to detect and output a plurality of signals which are radio waves arriving at the array antenna.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a receiver of an embodiment according to the present invention.

FIG. 2 is a block diagram showing a configuration of a reception signal processing circuit 5 of FIG.

3 is a flowchart of a received signal processing program executed by the received signal processing circuit 5 of FIG.

4 is a flowchart of a subroutine of calculation processing of a cost function value V _da in the reception signal processing program of FIG.

5 is a graph showing the cost function value V _da with respect to the signal number parameter da in the received signal processing shown in the flowchart of FIG.

FIG. 6 is a graph showing the standard deviation of the error of the detected incident angle θ with respect to the signal-to-noise power ratio of the incident signal when a coherent signal is incident in the reception signal processing circuit 5 of FIG. 1.

7 is a graph showing the standard deviation of the error of the detected incident angle θ with respect to the signal-to-noise power ratio of the incident signal when a non-coherent signal is incident in the reception signal processing circuit 5 of FIG.

8 is a graph showing the ratio of the output signal power to the added value of the noise power and the interference signal power when the signal-to-noise power ratio SNR of the input signal is 0 dB in the reception signal processing circuit 5 of FIG.

9 is a graph showing the ratio of the output signal power to the added value of the noise power and the interference signal power when the signal-to-noise power ratio SNR of the input signal is 10 dB in the reception signal processing circuit 5 of FIG.

10 is a graph showing the number of calculations per iteration with respect to the number L of antenna elements when the number d of incident signals is set to 7 in the received signal processing circuit 5 of FIG.

11 is a graph showing the number of calculations per one iteration with respect to the number of antenna elements L when the number d of incident signals is set to L-2 in the reception signal processing circuit 5 of FIG. .

[Explanation of symbols]

 1-1 to 1-L ... Antenna element, 2-1 to 2-L ... Down converter, 3-1 to 3-L ... A / D converter, 4, 13 ... Memory, 5 ... Received signal processing circuit, 6 Demodulator, 10 ... CPU, 11 ... ROM, 12 ... RAM, 14 ... Interface, 21 ... Normalized reception signal matrix memory, 22 ... Cost function value memory, 23 ... Output signal memory, 24 ... Incident angle vector memory, 25 Array response matrix memory, 26 QR decomposition matrix memory, 27 gradient vector memory, 28 Hessian matrix memory, 100 array antenna.

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] May 13, 1996

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] 0013

[Correction method] Change

[Correction contents]

Here, τ _i is the time required for the plane wave to propagate from the origin (here, the origin is at the antenna element 1-1, for example), which is the reference point, to the i-th antenna element 1-i. Is. Also, the envelope signal f (t) of Equation 1 is a real value and is a function of slowly changing time. _ni (t) is additional noise received by the i-th antenna element 1-i. In the present embodiment, the narrow band is a narrow band to the extent that it can be considered that the values of the envelope signal f (t) and the phase φ (t) do not change during the time when the plane wave passes through the array antenna 100. Say. Signal x _i (t) received by antenna element 1-i
Is frequency-converted by the down converter 2-i. At this time, since the multiplication by the sine wave sin (ω ₀ t) and cos (ω ₀ t) is involved, the output of the signal after the low-pass filtering is the in-phase component f (t + τ _i ) cos {ω _o τ _i + φ ( t + τ _i )} and the orthogonal component f (t + τ _i ) sin {ω _o τ _i + φ (t + τ _i )}. Here, since the incoming signal has a narrow band as described above, f (t + τ _i ) = f (t) and φ (t +
τ _i ) = φ (t) holds. When there are d signals in this way, the delay time τ _i has a value corresponding to the arrival direction of each plane wave and the position of the antenna element 1-i. The delay time of each plane wave in 1-i is represented by τ _ik . Where i = 1,
2, ..., L, and k = 1, 2 ,. Then, a complex number L-dimensional column vector x (t) which is the output of the array antenna 100 including the L antenna elements 1-i.
Can be expressed by the following equation 2.

[Procedure amendment 2]

[Document name to be amended] Statement

[Name of item to be corrected] 0024

[Correction method] Change

[Correction contents]

Here, the unitary matrix Q is composed of the (L × d) matrix Q ₁ consisting of the first column to the d-th column of the unitary matrix Q shown in Formula 10 and the unitary matrix Q shown in Formula 11. (D + 1) th to Lth columns {L
× (L−d)} matrix Q ₂ and Q = [Q ₁ Q ₂ ]
It expresses. Further, the matrix R of the equation 9 is represented by the equation 12 (d
It is expressed as shown in Expression 13 using the triangular matrix R ₁ of × d). Here, 0 means that all the components are 0 {(Ld)
Xd} is a zero matrix.

Front page continued (72) Inventor Takashi Sekiguchi, Kyoto, Soraku-gun, Seika-cho, Osamu Osamu, Osamu Osamu, 5 Hiratani Sanriya, Inc. AT Optical Hikari Radio Communications Research Laboratory, Inc. (72) Inventor, Ryu Miura, Seiraku-cho, Soraku-gun, Kyoto Kenjiya, Mihiratani No.5, ATR Optical Co., Ltd.

Claims (2)

[Claims] 1. The plurality of L, which are received by an array sensor in which a plurality of L sensor elements are arranged at a predetermined interval and which respectively arrive from different arrival directions.
A received signal processing device for detecting and outputting a smaller number of a plurality of d incoming signals, the plurality of L based on the plurality of L received signals respectively received by the plurality of L sensor elements. The number d of the arriving signals, the angles of the arriving directions of the plurality of d arriving signals to the array sensor, and the plurality of d arriving signals so that the signal components of the d arriving signals are maximized. Of the array response matrix of (L × d) corresponding to the input signals to the plurality of L sensor elements and representing the plurality of d output signals corresponding to the input signals to the plurality of L sensor elements. The (L × d) array response matrix corresponding to each arrival direction of the incoming signal is QR decomposed into the product of the (L × L) unitary matrix Q and the (L × d) matrix R, and The 1st to dth columns of the unitary matrix Q (L × d) matrix Q ₁ having columns up to and a (d × d) triangular matrix R _{1 having} rows from the first row to the d-th row of the above matrix R
Based on the plurality of L received signals and (d ×
d) the inverse matrix of the triangular matrix R ₁ ; the conjugate transposed matrix of the (L × d) matrix Q ₁ ; and the received signal matrix X composed of the plurality of L received signals. A received signal processing device, characterized by detecting and outputting a product matrix as the plurality of d incoming signals. 2. The received signal processing device according to claim 1, wherein each of the sensor elements is an antenna element, and the array sensor is an array antenna.